Datasets:

mlcore

/

arxiv-classifier

like 1

1501.03486

2

1501

A spin-wave logic gate based on a width-modulated dynamic magnonic crystal

An electric current controlled spin-wave logic gate based on a width-modulated dynamic magnonic crystal is realized. The device utilizes a spin-wave waveguide fabricated from a single-crystal Yttrium Iron Garnet film and two conducting wires attached to the film surface. Application of electric currents to the wires provides a means for dynamic control of the effective geometry of the waveguide and results in a suppression of the magnonic band gap. The performance of the magnonic crystal as an AND logic gate is demonstrated.

cond-mat.mes-hall

cond-mat

A spin-wave logic gate based on a width-modulated dynamic magnonic crystal A.A. Nikitin1,2,3*, A.B. Ustinov1,3, A.A. Semenov1, A.V. Chumak2, A.A. Serga2, V.I. Vasyuchka2, E. Lähderanta4, B.A. Kalinikos1, and B. Hillebrands2 1) Department of Physical Electronics and Technology, St. Petersburg Electrotechnical University, St. Petersburg, 197376 Russia 2) Fachbereich Physik and Landesforschungszentrum OPTIMAS, Technische Universität Kaiserslautern, Kaiserslautern, 67663 Germany 3) Department of Mathematics and Physics, Lappeenranta University of Technology, Lappeenranta, 53850 Finland An electric current controlled spin-wave logic gate based on a width-modulated dynamic magnonic crystal is realized. The device utilizes a spin-wave waveguide fabricated from a single-crystal Yttrium Iron Garnet film and two conducting wires attached to the film surface. Application of electric currents to the wires provides a means for dynamic control of the effective geometry of the waveguide and results in a suppression of the magnonic band gap. The performance of the magnonic crystal as an AND logic gate is demonstrated. * [email protected] In recent years artificially patterned magnetic media, magnonic crystals (MCs), have commanded increased interest (see reviews [1-4] and literature therein). One of the important features of MCs is the presence of band gaps in the spin-wave spectrum, i.e. frequency bands in which the propagation of spin waves is forbidden [5-7]. The dispersion management introduced by an artificial periodicity of the magnetic film allows for the observation of a variety of different linear and nonlinear spin-wave phenomena. Among the formation of band gaps [8-11] and the generation of gap solitons [12, 13]. Magnonic crystals are also promising for practical applications, particularly for phase shifters [14] and generators [15], and they have also been suggested for use as temperature or magnetic field sensors [16, 17]. them are A very fascinating type of MCs is the dynamic magnonic crystal (DMC) [18-20]. This is a wave guiding structure with rapidly switchable periodic properties. Up to now, it has been realized in the form of a YIG film with a meander type conductor placed on its surface. An electric current in the conductor produces a periodic spatial modulation of the bias magnetic field [18]. Due to this property, the dynamic for unique signal processing magnonic crystal allows functions, such as all-linear frequency conversion [19], and signal storage [20]. reversal, time logic functionality can also be realized via control of the spin- wave amplitude in the interferometer arms [23, 26]. The logic gate presented here does not utilize any interferometer circuitry; instead, it consists of one WMDMC, which controls the spin-wave amplitude. This function is realized through control of the sensitivity of the propagating spin wave to the edge modulation of the magnetic-film waveguide structure. A schematic view of the WMDMC is shown in Fig. 1. The device was fabricated using an epitaxial Yttrium Iron Garnet (YIG) film. For the experimental investigation, the YIG-film waveguide was cut from a high-quality single-crystal YIG film of 8.5 µm thickness grown on 500 µm thick Gadolinium Gallium Garnet (GGG) substrate by liquid-phase epitaxy. The film has pinned surface spins that lead to standing spin-wave resonances and appearance of additional dips in the amplitude- frequency characteristic [27]. Periodic sinusoidal width modulation of the YIG-film waveguide was made by chemical etching of the film. The spatial modulation period Tm and the peak-to-peak amplitude 2Am (see Fig. 1) were 400 m and 200 m, respectively (see Fig. 1). The length of the modulated In this Letter, we report on the realization of a width- modulated dynamic magnonic crystal (WMDMC). In the designed structure, in contrast to the aforementioned DMC, electric currents control the effective geometry of the magnetic film waveguide for spin waves propagating therein. Moreover, we demonstrate the application of the WMDMC as a spin- wave logic gate. It should be noted that so far spin-wave logic gates have only been realized using a spin-wave interferometer geometry [21-26]. For example, elements providing controllable phase shifts for spin waves propagating in the arms of a Mach- Zehnder interferometer were included in each arm. The logic operations were based on changing the phase of the spin-wave by π with an external current signal coded as a logical ‘1’. The Fig. 1. Sketch of the width-modulated dynamic magnonic crystal. The ports A and B represent the logical inputs. The currents I1 and I2 supplied to the ports correspond to a logical ‘1’ if their values equal IMBG. Zero currents correspond to a logical ‘0’. The output spin-wave signal represents the logical output.  (a) (b) (c) (d) 1570 Oe Hext 1600 Oe Fig. 2. Distributions of the bias magnetic field (top) and transmitted spin-wave power versus frequency for the width-modulated dynamic magnonic crystal (bottom) for the cases when the electric current is not applied to the wires (a); the current IMBG is applied to only one of the wires (b,c); and the current IMBG is applied to both wires (d). The value of the current IMBG = 1500 mA. the conditions for excitation of area was 4 mm, while the width of the unmodulated section of the YIG-film waveguide was 1.5 mm. A spatially uniform bias magnetic field was applied across the YIG waveguide in order to provide surface magnetostatic spin waves [28]. These waves were exited and detected by the microwave strip-line antennas placed at equal distances from both ends of the modulated YIG-film area 10 mm away from each other. Two gold wires of 50 m in diameter were placed on the YIG-film surface along the modulated edges, as shown in Fig. 1. These wires were used for the dynamic control of the magnonic crystal band gap by supplying the electric currents. Let us consider qualitatively the formation of a band gap in the width-modulated magnonic crystal. Generally, a wave vector of traveling spin waves exited by the input antenna in the YIG film has two components, which are in-plane. The first component is longitudinal along the direction from the input to the output antenna while the second component is transverse. As it is physically clear, the value of the transverse component is dependent on the width of the waveguide. This effect leads to the periodic change in the wave guiding properties of the width-modulated YIG film and provides the magnonic band gap (MBG) in the amplitude vs. frequency characteristic of the investigated MC. The number of MBGs is defined by the type of the width modulation. For example, in the case of rectangular modulation there are several MBGs. As soon as the width modulation is sinusoidal, as is the case in the investigated structure (Fig. 1), there is only one MBG in the power vs. frequency characteristic (see Fig. 2(a)) [29]. the borders of The physical mechanism underlying the control of the magnonic crystal band gap by the electric currents applied to the wires can be understood as follows. Positive currents I1 and I2 applied to both wires create an additional negative Oersted field with respect to the applied static bias field. Therefore, the spatial distribution of the resulting magnetic field has two minima at the width modulation. The distributions of the bias magnetic field are shown in Fig. 2 as color maps. Provided the depth of the minima is sufficiently large, they screen the YIG-film modulated edges due to a decrease in the magnetic field inside the YIG waveguide. In other words, the existence of these minima cause the width modulation of the waveguide to become “invisible” to spin waves. This effect leads to the disappearance of the band gap at f0 = 6.6 GHz, as is shown in Fig. 2(d). We will now present the dynamic properties of the device structure realized upon application of the controlling dc electric currents. The experiments were carried out in the following manner. A microwave signal applied to the input antenna excited a surface magnetostatic spin wave. After passing through the magnonic crystal the spin wave was detected by the output antenna (Fig. 1). The power of the received spin waves was measured as a function of their carrier frequency. A typical dependence measured for H = 1600 Oe with no current in the edge wires is shown in Fig. 2(a). As can be seen in the graph, the center frequency of the band gap was f0 = 6.6 GHz. The dependence of the power of the output signal vs. frequency obtained under two equal controlling currents IMBG of 1500 mA is shown in Fig. 2(d). Note that the data clearly demonstrate (in comparison to Fig. 2(a)) a suppression of the band gap at 6.6 GHz. If the dc current was applied to one of the wires then only one of the modulated edges was deactivated. In this case, the input spin wave was still scattered on its path through the waveguide. In this situation, the band gap appears as shown in Figs 2(b, c). The difference between Fig. 2(b) and Fig. 2(c) could be explained by a possible asymmetry in the distribution of the spin-wave amplitude along the cross-section of the YIG-film waveguide. In this case, there is a difference in the sensitivity of the propagating wave to the boundary conditions on the different waveguide edges. Dependences of the spin-wave transmission as a function of the currents I1 and I2 measured at the frequency f0 = 6.6 GHz is shown in Fig. 3. As is visible, an increase in both currents leads to a substantial increase in the spin-wave transmission due to a reduction in the influence of the modulated edges. At the same time, there is a relatively weak influence with only one dc current applied. 6.66.76.80.1110Magnonic crystal band gap I1=I2= 0 mA Power (a. u.)Frequency (GHz) f0=6.6 GHz6.66.76.80.1110 Magnonic crystal band gap I1= IMBG I2= 0 mAFrequency (GHz)Power (a. u.) 6.66.76.80.1110 Magnonic crystal band gap I2= IMBG I1= 0 mAFrequency (GHz) Power (a. u.)6.66.76.80.1110 I1=I2=IMBG Power (a. u.)Frequency (GHz) Fig. 3 Dependence of the spin wave transmission on the value of the dc current at frequency f0=6.6 GHz. Squares: both currents I1 and I2 are changed; triangles: the current I1 is changed and I2=0; circles: I1=0 and I2 is changed. A possible application of the WMDMC for realization of an AND logic gate will now be explored. The operating principle of the AND logic gate is based on controlling the bias magnetic field distribution along the sinusoidal borders of the YIG-film waveguide through a change in the electric currents I1 and I2. The logic ‘0’ is represented by I = 0 mA and a logic ‘1’ – by I = IMBG = 1500 mA, which is enough to "switch off" the width modulation on one side of the WMDMC. The microwave pulses at the input antenna represent clock pulses. The microwave signal at the output antenna serves as the logic signal output. The operational frequency corresponds to the central frequency of the band gap, for example f0 = 6.6 GHz. Therefore, the presence of the band gap (i.e., low power level at the output) represents a logic ‘0’ and the absence of the band gap (i.e., high power at the output, Pout = P1) represents a logic ‘1’. traces of signal demonstrating the performance of the logic gate are shown in Fig. 4. It is observed that the logic ‘0’ appears at the output port in three situations: for two logic ‘0s’ applied to the input ports A and B (Fig. 4(a)) and for the combinations of the logic ‘0’ and the logic ‘1’ applied to the inputs (Fig. 4(b, c)). In the case where a logic ‘1’ is applied simultaneously to both inputs, the signal at the output port corresponds to a logic ‘1’. Typical oscilloscope the output In conclusion, width-modulated magnonic crystals may find a variety of applications. For example, they can be employed for the development of dynamic magnonic crystals and spin-wave logic gates, as demonstrated in this paper. Another example could be microwave notch filters with a stop- band frequency corresponding to the magnonic band gap. Advantages of the dynamic magnonic crystal presented above in comparison to the ones previously developed [18-20] are the potentially for miniaturization. Indeed, the current control is provided here by short conductors, which have an inductance much less than the meander type wire reported in [18]. The use of nanostructured width-modulated waveguides made of Permalloy films, as in Ref. [9], will allow for a significant reduction of the size of the dynamic magnonic crystal and the logic circuit presented here. fast performance and the possibility Fig. 4 Waveforms of the output signal of the AND logic gate. Microwave signal frequency f0=6.6 GHz. Currents I1 and I2 serve as logic inputs; logic ‘0’ corresponds to zero current, logic ‘1’- to IMBG=1500 mA. High level of the microwave output signal corresponds to a logic ‘1’, and low level to a logic ‘0’. The work was supported in part by the Russian Science Foundation (Grant 14-12-01296), the Ministry of Education and Science of Russian Federation, the Academy of Finland, EU-FET grant the Deutsche Forschungsgemeinschaft. InSpin 612759, and by 1. A. A. Serga, A. V. Chumak, and B. Hillebrands, J. Phys. D: Appl. Phys. 43, 264002 (2010). 2. V. V. Kruglyak, S. O. Demokritov, and D. Grundler, J. Phys. D: Appl. Phys. 43, 264001 (2010). 3. B. Lenk, H. Ulrichs, F. Garbs, and M. Münzenberg, Phys. Rep. 507, 107 (2011). 4. M. Krawczyk and D. Grundler, J. Phys.: Condens. Matter 26, 123202 (2014). 5. J. W. Kos, M. Krawczyk, Yu. S. Dadoenkova, N. N. Dadoenkova, and I. L. Lyubchanskii, J. Appl. Phys. 115, 174311 (2014). 6. H. T. Nguyen and M. G. Cottam, J. Appl. Phys. 111, 07D122 (2012). 7. M. Mruczkiewicz, M. Krawczyk, V. K. Sakharov, Yu. V. Khivintsev, Yu. A. Filimonov, and S. A. Nikitov, J. Appl. Phys. 113, 093908 (2013). 8. K.-S. Lee, D.-S. Han, and S.-K. Kim, Phys. Rev. Lett. 102, 127202 (2009). 9. A. V. Chumak et al., Appl. Phys. Lett. 95, 262508 (2009). 10. M. Arikan, Y. Au, G. Vasile, S. Ingvarsson, and V. V. Kruglyak, J. Phys. D: Appl. Phys. 46, 135003 (2013). 11. A. B. Ustinov and B. A. Kalinikos, Tech. Phys. Lett. 40, 568 (2014). 12. A. B. Ustinov, B. A. Kalinikos, V. E. Demidov, and S. O. Demokritov, Phys. Rev. B 81, 180406(R) (2010). 13. S. V. Grishin, Y. P. Sharaevskii, S. A. Nikitov, E. N. Beginin, and S. E. Sheshukova, IEEE Trans. Mag. 47, 3716 (2011). 14. Y. Zhu, K. H. Chi, and C. S. Tsai, Appl. Phys. Lett. 105, 022411 (2014). 0300600900120015000.000.250.500.751.00 I1=I2= I I1= I, I2= 0 I1= 0, I2= INormalized transmission (a. u.)Current I (mA)15. S. V. Grishin, E. N. Beginin, M. A. Morozova, Yu. P. Sharaevskii, and S. A. Nikitov, J. Appl. Phys. 115, 053908 (2014). 16. M. Inoue, A. Baryshev, H. Takagi, P. B. Lim, K. Hatafuku, J. Noda, and K. Togo, Appl. Phys. Lett. 98, 132511 (2011). 17. R. G. Kryshtal and A. V. Medved, Appl. Phys. Lett. 100, 192410 (2012). 18. A. V. Chumak, T. Neumann, A. A. Serga, B. Hillebrands, and M. P. Kostylev J. Phys. D: Appl. Phys. 42, 205005 (2009). 19. A. V. Chumak, V. S. Tiberkevich, A. D. Karenowska, A. A. Serga, J. F. Gregg, A. N. Slavin, B. Hillebrands, Nat. Commun. 1, 141 (2010). 20. A. D. Karenowska, J. F. Gregg, V. S. Tiberkevich, A. N. Slavin, A. V. Chumak, A. A. Serga, and B. Hillebrands, Phys. Rev. Lett. 108, 015505 (2012). 21. A. B. Ustinov and B. A. Kalinikos, Tech. Phys. Lett. 27, 403 (2001). 22. M. P. Kostylev, A. A. Serga, T. Schneider, B. Leven, and B. Hillebrands, Appl. Phys. Lett. 87, 153501 (2005). 23. T. Schneider, A. A. Serga, B. Leven, B. Hillebrands, R. L. Stamps, and M. P. Kostylev, Appl. Phys. Lett. 92, 022505 (2008). 24. A. Khitun, M. Bao, and K. L. Wang, J. Phys. D: Appl. Phys. 43, 264005 (2010). 25. Y. Au, M. Dvornik, O. Dmytriiev, and V. V. Kruglyak, Appl. Phys. Lett. 100, 172408 (2012). 26. A. V. Chumak, A. A. Serga, and B. Hillebrands, Nat. Commun. 5, 4700 (2014). 27. B. A. Kalinikos, A. N. Slavin, J. Phys. C: Solid State Phys. 19, 7013 (1986). 28. D. D. Stancil and A. Prabhakar, Spin Waves: Theory and Applications, Springer, 2009. 29. F. Ciubotaru, A. V. Chumak, N. Yu. Grigoryeva, A. A. Serga, and B. Hillebrands, J. Phys. D: Appl. Phys. 45, 255002 (2012). 

0908.4485

2

0908

Discontinuous Euler instability in nanoelectromechanical systems

We investigate nanoelectromechanical systems near mechanical instabilities. We show that quite generally, the interaction between the electronic and the vibronic degrees of freedom can be accounted for essentially exactly when the instability is continuous. We apply our general framework to the Euler buckling instability and find that the interaction between electronic and vibronic degrees of freedom qualitatively affects the mechanical instability, turning it into a discontinuous one in close analogy with tricritical points in the Landau theory of phase transitions.

cond-mat.mes-hall

cond-mat

Discontinuous Euler instability in nanoelectromechanical systems Guillaume Weick,1, 2 Fabio Pistolesi,3, 4 Eros Mariani,1, 5 and Felix von Oppen1 1Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, D-14195 Berlin, Germany 2IPCMS (UMR 7504), CNRS and Universit´e de Strasbourg, F-67034 Strasbourg, France 3CPMOH (UMR 5798), CNRS and Universit´e de Bordeaux I, F-33405 Talence, France 4LPMMC (UMR 5493), CNRS and Universit´e Joseph Fourier, F-38042 Grenoble, France 5School of Physics, University of Exeter, Stocker Road, Exeter, EX4 4QL, UK (Dated: November 1, 2018) We investigate nanoelectromechanical systems near mechanical instabilities. We show that quite generally, the interaction between the electronic and the vibronic degrees of freedom can be ac- counted for essentially exactly when the instability is continuous. We apply our general framework to the Euler buckling instability and find that the interaction between electronic and vibronic de- grees of freedom qualitatively affects the mechanical instability, turning it into a discontinuous one in close analogy with tricritical points in the Landau theory of phase transitions. PACS numbers: 73.63.-b, 85.85.+j, 63.22.Gh I. INTRODUTION The buckling of an elastic rod by a longitudinal com- pression force F applied to its two ends constitutes the paradigm of a mechanical instability, called buckling instability.1 It was first studied by Euler in 1744 while in- vestigating the maximal load that a column can sustain.2 As long as F stays below a critical force Fc, the rod re- mains straight, while for F > Fc it buckles, as sketched in Fig. 1a-b. The transition between the two states is con- tinuous and the frequency of the fundamental bending mode vanishes at the instability. There has been much recent interest in exploring buck- ling instabilities in nanomechanical systems. In the quest to understand the remarkable mechanical prop- erties of nanotubes,3 -- 5 there have been observations of compressive buckling instabilities in this system.6 The Euler buckling instability has been observed in SiO2 nanobeams and shown to obey continuum elasticity theory.7 There are also close relations with the recently observed wrinkling8 and possibly with the rippling9 of suspended graphene samples. Theoretical works have studied the quantum properties of nanobeams near the Euler instability,10 -- 13 proposing this system to explore zero-point fluctuations of a mechanical mode11 or to serve as a mechanical qubit.13 In this work, we study the interaction of current flow with the vibrational motion near such continuous me- chanical instabilities which constitutes a fundamental issue of nanoelectromechanics.14 Remarkably, we find that under quite general conditions, this problem ad- mits an essentially exact solution due to the continu- ity of the instability and the consequent vanishing of the vibronic frequency at the transition ("critical slow- ing down"). In fact, the vanishing of the frequency implies that the mechanical motion becomes slow com- pared to the electronic dynamics and an appropriate non- equilibrium Born-Oppenheimer (NEBO) approximation becomes asymptotically exact near the transition. Here, FIG. 1: (color online) Sketch of a nanobeam (a) in the flat state and (b) the buckled state with two equivalent metastable positions of the rod (solid and dashed lines). An equivalent circuit of the embedded SET is shown in (c). we illustrate our general framework by applying it to the nanoelectromechanics of the Euler instability. We find that the interplay of electronic transport and the mechanical instability causes significant qualitative changes both in the nature of the buckling and in the transport properties. In leading order, the NEBO ap- proximation yields a current-induced conservative force acting on the vibronic mode. At this order, our principal conclusion is that the coupling to the electronic dynamics can change the nature of the buckling instability from a continuous to a discontinuous transition which is closely analogous to tricritical behavior in the Landau theory of phase transitions. Including in addition the fluctua- tions of the current-induced force as well as the corre- sponding dissipation leads to Langevin dynamics of the vibrational mode which becomes important in the vicin- ity of the discontinuous transition. Employing the same NEBO limit to deduce the electronic current, we find that the buckling instability induces a current blockade over a wide range of parameters. This is a manifestation of the Franck-Condon blockade15 -- 17 whenever the buckling instability remains continuous but is caused by a novel tricritical blockade when the instability is discontinuous. The emergence of a current blockade in the buckled state suggests that our setup could, in principle, serve as a mechanically-controlled switching device. II. MODEL Close to the Euler instability, the frequency of the fun- damental bending mode of the beam approaches zero, while all higher modes have a finite frequency.1 This al- lows us to retain only the fundamental mode of amplitude X (see Fig. 1a-b) and following previous studies,10 -- 12 we reduce the vibrational Hamiltonian to18 Hvib = P 2 2m + mω2 2 X 2 + α 4 X 4, (1) which is closely analogous to the Landau theory of contin- uous phase transitions. In Eq. (1), P is the momentum conjugate to X and m denotes an effective mass. The mode frequency ω2 ∼ 1 − F/Fc changes sign when F reaches a critical force Fc. Global stability then requires a quartic term with α > 0. Thus, for F < Fc (ω2 > 0), X = 0 is the only stable minimum and the beam remains straight. For F > Fc (ω2 < 0), the beam buckles into one of the two minima at ±X+ = ±(cid:112)−mω2/α. The vibronic mode of the nanobeam interacts with an embedded metallic single-electron transistor (SET), con- sisting of a small metallic island coupled to source, drain, and gate electrode (Fig. 1c). We assume that the SET operates in the Coulomb blockade regime19 and that bias and gate voltage are tuned to the vicinity of the conduct- ing region between SET states with, say, zero and one excess electron. The electronic degrees of freedom cou- ple to the vibronic motion through the occupation n of excess electrons on the metallic island. Specifically, we assume that the electron-vibron coupling does not break the underlying parity symmetry of the vibronic dynam- ics under X → −X. This follows naturally when the coupling emerges from the electron-phonon coupling in- trinsic to the nanobeam20 and implies that the coupling depends only on even powers of the vibronic mode coor- dinate X. The dominant coupling is quadratic in X, with a coupling constant g > 0.20 When there is a signif- icant contribution to the electron-vibron coupling origi- nating from the electrostatic dot-gate interaction, we en- vision a symmetric gate setup consistent with Eq. (2). In the presence of the vibronic dynamics X(t), the elec- tronic occupation n(X, t) of the island is described by the Boltzmann-Langevin equation21 = {n, Hvib} + Γ+(1 − n) − Γ−n + δJ+ − δJ−. (3) dn dt This equation assumes that the bias is large compared to temperature so that tunneling is effectively unidi- rectional and the relevant tunneling rates Γ± for tun- neling onto (+) and off (−) the island are given by Γ± = R−1(V /2 ± ¯Vg)Θ(V /2 ± ¯Vg). Here, R denotes the tunneling resistances (R (cid:29) h/e2) between island and leads, V is the bias voltage, Θ(x) denotes the Heaviside 2 step function, and we set  = e = 1. Since both gate voltage (via the capacitances C and Cg in Fig. 1c) and vi- bronic deformations couple to the excess charge n on the island, the effective gate voltage ¯Vg = Vg − gX 2/2 com- bines the gate potential Vg (measured from the degen- eracy point between the states with zero and one excess electron) and the vibron-induced shift of the electronic energy described by Hc. The stochastic Poisson nature of electronic tunneling is accounted for by including the Langevin sources δJ± with correlators (cid:104)δJ+(t)δJ+(t(cid:48))(cid:105) = Γ+(1 − n)δ(t − t(cid:48)) and (cid:104)δJ−(t)δJ−(t(cid:48))(cid:105) = Γ−nδ(t − t(cid:48)). The vibronic dynamics enters Eq. (3) through the Pois- son bracket {n, Hvib}. III. STABILITY ANALYSIS We are now in a position to investigate the influence of the electronic dynamics on the vibronic motion. Near the instability, the vibrational dynamics becomes slow compared to the electronic tunneling dynamics. As has recently been shown,22,23 the effect of the current on the vibrational motion can then be described within a NEBO approximation in which the vibrational motion is subject to a current-induced force −gXn(X, t) originat- ing in the electron-vibron interaction (2). This current- induced force involves both a time-averaged and conser- vative force as well as fluctuating and frictional forces, resulting in Langevin dynamics of the vibronic degree of freedom. In lowest order, the Langevin dynamics only involves the conservative force which emerges from the average occupation n0(X) in the absence of fluctuations (δJ± (cid:39) 0) and vibronic dynamics ({n, Hvib} (cid:39) 0). In this limit, Eq. (3) reduces to the usual rate equation of a metallic SET so that19 1, 1 2 0, vg(x) > v/2, + vg(x) v , −v/2 (cid:54) vg(x) (cid:54) v/2, vg(x) < −v/2 (4) with v > 0 and vg(x) = vg − x2/2. Here and below, we employ dimensionless variables by introducing char- acteristic scales E0 = g2/α of energy, l0 = (cid:112)g/α of length, and ω0 = (cid:112)g/m of frequency (or time t) from a comparison of the quartic vibron potential in Hvib and the electron-vibron coupling Hc. Specifically, we intro- duce the reduced variables x = X/l0, p = P/mω0l0, τ = ω0t, v = V /E0, vg = Vg/E0, and r = Rω0/E0. In terms of these variables, we can also write Hvib + Hc = E0[p2/2 + (−
+ n)x2/2 + x4/4] in terms of a reduced compressional force
= −mω2/g. The current-induced force −xn0(x) has dramatic ef- fects on the Euler instability, as follows from a stability analysis of the vibrational motion. The (meta)stable po- sitions of the nanobeam are obtained by setting the effec- tive force feff (x) =
x − x3 − xn0(x) to zero. Our results Hc = g 2 X 2 n, (2) n0(x) =  3 FIG. 2: (color online) (Meta)stable (solid blue lines) and unstable (dashed red lines) positions of the nanobeam vs. scaled force
for (a) vg < v/2, v < 1/2, (b) vg < v/2, v > 1/2, (c) vg > v/2, v < 1/2 (for
+ > 1; a similar plot holds for
+ < 1), (d) vg > v/2, v > 1/2, as indicated in the vg -- v plane in (e). The dotted blue line is the result without electron-vibron coupling. + =
− 1, and x2− = (
−
)/(1 − 1/2v). Grey indicates conducting regions. Notation:
± = 2vg ± v,
= 1/2 + vg/v, x2 + =
, x2 are summarized in the stability diagrams in Fig. 2. The most striking results of this analysis are: (i) The current flow renormalizes the critical force required for buckling towards larger values. (ii) At low biases, the buckled state can appear via a discontinuous transition. These results can be understood most directly in terms of the potential veff (x) associated with feff (x). Focusing on the current-carrying region (shown in grey in Fig. 2 and delineated by max{0,
−} < x2 <
+ with
± = 2vg ± v), we find veff (x) = 1 2 −
+ v + 2vg 2v x2 + 1 4 1 − 1 2v x4. (5) (cid:18) (cid:19) (cid:18) (cid:19) The quadratic term shows that the current indeed stabi- lizes the unbuckled state, renormalizing the critical force to
= 1/2 + vg/v when
− < 0 <
+ (Fig. 2a-b). Re- markably, however, the current-induced contribution to the quartic term is negative at small x2 and thus desta- bilizes the unbuckled state. According to Eq. (5), the quartic term in the current-induced potential becomes in- creasingly significant as the bias voltage v decreases and we find that the overall prefactor of the quartic term be- comes negative when v < 1/2.24 It is important to note that this does not imply a globally unstable potential since the current-induced force contributes only for small x2. A sign reversal of the quartic term is also a famil- iar occurrence in the Landau theory of tricritical points which connect between second- and first-order transition lines.25 In close analogy, the sign reversal of the quartic term in the effective potential (5) signals a discontinuous Euler instability which reverts to a continuous transition at biases v > 1/2 where the prefactor of the quartic term remains positive. Specifically, when v > 1/2 (Fig. 2b,d), the current- induced potential renormalizes the parameters of the vi- bronic Hamiltonian but leaves the quartic term posi- tive. This modifies how the position of the minimum depends on the applied force in the conducting region max{0,
−} < x2 <
+, but the Euler instability remains continuous. When v < 1/2, the equilibrium position at finite x becomes unstable within the entire current- carrying region. This leads to a discontinuous Euler tran- sition when
− < 0 <
+ (Fig. 2a) and to multistability in the region
− < x2 <
+ when
− > 0 (Fig. 2c).26 At the level of the stability analysis, we can also ob- tain the current I by evaluating the rate-equation result19 RI(x)/V = 1/4 − [vg(x)/v]2 at the position of the most stable minimum. Corresponding results in the vg -- v plane are shown in Fig. 3a-f for various values of the applied force
. By comparison with the Coulomb di- amond in the absence of the electron-vibron coupling (dotted lines in Fig. 3), we see that the Euler instability leads to a current blockade over a significant parameter range. For v > 1/2, the blockade is a manifestation of the Franck-Condon blockade,15,17 caused by the induced linear electron-vibron coupling when expanding Eq. (2) about the buckled state. In contrast, for v < 1/2, the current blockade is a direct consequence of the discontinuous Euler instabil- FIG. 3: (color online) Conductance G = RI/V in the vg -- v plane for applied force (a)
(cid:54) 0, (b)
= 0.25, (c,g)
= 0.5, (d)
= 0.75, (e)
= 1, (f,h)
= 1.25, within (a-f) stability analysis and (g-h) full Langevin dynamics (r = γe = T = 0.01). Color scale: G = 0 → 1/4 from dark blue to white. Dotted lines delineate the Coulomb diamond for g = 0. 4 ity. We have seen above that in this regime, the buckled state becomes unstable throughout the entire current- carrying region. As a result, the current-induced force will always drive the system out of the current-carrying region, explaining the current blockade. An intriguing feature of this novel tricritical current blockade is the curved boundary of the apparent Coulomb-blockade dia- mond (Fig. 3), a behavior which is actually observed in nanoelectromechanical systems. Planck equation (6). Numerical results for the scaled lin- ear conductance G = RI/V are shown in Fig. 3g-h, using the same parameters as in Fig. 3c,f. We observe that the fluctuations reduce the size of the blockaded region and blur the edges of the conducting regions as the system can explore more conducting states in phase space. Nev- ertheless, the conclusions of the stability analysis clearly remain valid qualitatively. IV. LANGEVIN DYNAMICS V. CONCLUSION To investigate the robustness of the stability analysis against fluctuations, we turn to the complete vibronic Langevin dynamics x + γ(x) x = feff(x) + ξ(τ ). The fluctuating force ξ(τ ) is generated by fluctuations of the electronic occupation and the frictional force −γ(x) x by the delayed response of the electrons to the vibronic dynamics. To compute γ(x) and ξ(τ ), we solve Eq. (3) including the vibronic dynamics and the Langevin sources. Writing separate equations for average and fluctuations of the occupation by setting n = ¯n + δn, we see that the leading correction to ¯n arises from the Poisson bracket, yielding ¯n = n0 − X∂X n0. At the same time, the fluctuations δn obey the correla- tor (cid:104)δn(t)δn(t(cid:48))(cid:105) = Insert- ing these results into the expression for the current- induced force −gXn and employing reduced units, we find (cid:104)ξ(τ )ξ(τ(cid:48))(cid:105) = D(x)δ(τ−τ(cid:48)) with diffusion and damp- ing coefficients D(x) = 2rx2n0(1 − n0)/v and γ(x) = −rx∂xn0/v, respectively. Finally, we can pass from the Langevin to the equivalent Fokker-Planck equation22,23 for the probability P(x, p, τ ) that the nanobeam is at po- sition x and momentum p at time τ , Γ++Γ− n0(1 − n0)δ(t − t(cid:48)). 2 1 Γ++Γ− We have presented a general approach to the interplay between continuous mechanical instabilities and current flow in nanoelectromechanical systems, and have applied our general framework to the Euler buckling instabil- ity. The current flow modifies the nature of the buck- ling instability from a continuous to a tricritical transi- tion. Likewise, the instability induces a novel tricritical current blockade at low bias. Our nonequilibrium Born- Oppenheimer approach generalizes not only to other con- tinuous mechanical instabilities, but also to other systems such as semiconductor quantum dots or single-molecule junctions with a discrete electronic spectrum, to other types of electron-vibron coupling,28 and to further trans- port characteristics (e.g., current noise). Our proposed setup can be realized experimentally by clamping, e.g., a suspended carbon nanotube and apply- ing a force to atomic precision either using a break junc- tion or an atomic force microscope. Indeed, several recent experiments show that the electron-vibron coupling is surprisingly strong in suspended carbon nanotube quan- tum dots.4,5,16 ∂τP = −p∂xP − feff∂pP + γ∂p(pP) + pP. ∂2 (6) Acknowledgments D 2 The current I =(cid:82) dxdpPst(x, p)I(x) is now obtained Note that the diffusion and damping coefficients are non- vanishing only in the conducting region.27 from the stationary solution ∂τPst = 0 of the Fokker- We acknowledge financial support through Sfb 658 of the DFG (GW, EM, FvO) and ANR contract JCJC- 036 NEMESIS (FP). FvO enjoyed the hospitality of the KITP (NSF PHY05-51164). 1 L. D. Landau and E. M. Lifshitz, Theory of Elasticity (Pergamon Press, Oxford, 1970). 2 L. Euler, in Leonhard Euler's Elastic Curves, translated and annotated by W. A. Oldfather, C. A. Ellis, and D. M. Brown, reprinted from ISIS, No. 58 XX(1), 1744 (Saint Catherine Press, Bruges). 3 P. Poncharal et al., Science 283, 1513 (1999). 4 G. A. Steele et al., Science 325, 1103 (2009). 5 B. Lassagne et al., Science 325, 1107 (2009). 6 M. R. Falvo et al., Nature 389, 582 (1997). 7 S. M. Carr and M. N. Wybourne, Appl. Phys. Lett. 82, 709 (2003). 9 J. C. Meyer et al., Nature 446, 60 (2007). 10 S. M. Carr, W. E. Lawrence, and M. N. Wybourne, Phys. Rev. B 64, 220101(R) (2001). 11 P. Werner and W. Zwerger, Europhys. Lett. 65, 158 (2004). 12 V. Peano and M. Thorwart, New J. Phys. 8, 21 (2006). 13 S. Savel'ev, X. Hu, and F. Nori, New J. Phys. 8, 105 (2006). 14 H. G. Craighead, Science 290, 1532 (2000); M. L. Roukes, Phys. World 14, 25 (2001). 15 J. Koch and F. von Oppen, Phys. Rev. Lett. 94, 206804 (2005). 16 R. Leturcq et al., Nature Phys. 5, 327 (2009). 17 F. Pistolesi and S. Labarthe, Phys. Rev. B 76, 165317 8 W. Bao et al., Nature Nanotech. 4, 562 (2009). (2007). 18 The rotation of the plane of the buckled nanobeam is as- sumed massive due to clamped boundary conditions. 19 See, e.g., Chap. 3 in T. Dittrich et al., Quantum Transport and Dissipation (Wiley-VCH, Weinheim, 1998). 20 E. Mariani and F. von Oppen, Phys. Rev. B 80, 155411 (2009). 21 See Ya. M. Blanter and M. Buttiker, Phys. Rep. 336, 1 (2000) for a review of the Boltzmann-Langevin method. 22 Ya. M. Blanter, O. Usmani, and Yu. V. Nazarov, Phys. Rev. Lett. 93, 136802 (2004); ibid. 94, 049904(E) (2005). 23 D. Mozyrsky, M. B. Hastings, and I. Martin, Phys. Rev. B 73, 035104 (2006). 24 We note that the singularity at small v is cut off for bias voltages of the order of temperature or level broadening. 5 25 See, e.g., P. M. Chaikin and T. C. Lubensky, Principles of Condensed Matter Physics (Cambridge University Press, Cambridge, 1995). 26 In a quite different context, a discontinuous Euler insta- bility has also been predicted in: S. Savel'ev and F. Nori, Phys. Rev. B 70, 214415 (2004). 27 In some cases, a stable numerical solution of Eq. (6) re- quires a small extrinsic damping γe and temperature T . 28 E.g., a small symmetry-breaking coupling linear in X leads to a tricritical point in an external field, a purely linear coupling to a 2nd order transition in a field, G. Weick et al., unpublished. 

1007.3144

1

1007

Polarization dependence of coherent phonon generation and detection in highly-aligned single-walled carbon nanotubes

We have investigated the polarization dependence of the generation and detection of radial breathing mode (RBM) coherent phonons (CP) in highly-aligned single-walled carbon nanotubes. Using polarization-dependent pump-probe differential-transmission spectroscopy, we measured RBM CPs as a function of angle for two different geometries. In Type I geometry, the pump and probe polarizations were fixed, and the sample orientation was rotated, whereas, in Type II geometry, the probe polarization and sample orientation were fixed, and the pump polarization was rotated. In both geometries, we observed a very nearly complete quenching of the RBM CPs when the pump polarization was perpendicular to the nanotubes. For both Type I and II geometries, we have developed a microscopic theoretical model to simulate CP generation and detection as a function of polarization angle and found that the CP signal decreases as the angle goes from 0 degrees (parallel to the tube) to 90 degrees (perpendicular to the tube). We compare theory with experiment in detail for RBM CPs created by pumping at the E44 optical transition in an ensemble of single-walled carbon nanotubes with a diameter distribution centered around 3 nm, taking into account realistic band structure and imperfect nanotube alignment in the sample.

cond-mat.mes-hall

cond-mat

Polarization dependence of coherent phonon generation and detection in highly-aligned single-walled carbon nanotubes L. G. Booshehri,1, 2 C. L. Pint,2, 3, 4 G. D. Sanders,5 L. Ren,1, 2 C. Sun,1, 2 E. H. H´aroz,1, 2 J.-H. Kim,6 K.-J. Yee,6 Y.-S. Lim,7 R. H. Hauge,2, 4 C. J. Stanton,5 and J. Kono1, 2, 3, ∗ 1Department of Electrical and Computer Engineering, Rice University, Houston, Texas 77005, USA 2The Richard E. Smalley Institute for Nanoscale Science and Technology, Rice University, Houston, Texas 77005 3Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA 4Department of Chemistry, Rice University, Houston, Texas 77005, USA 5Department of Physics, University of Florida, Box 118440, Gainesville, Florida 32611-8440 6Department of Physics, Chungnam National University, Daejeon, 305-764, Republic of Korea 7Department of Applied Physics, Konkuk University, Chungju, Chungbuk, 380-701, Republic of Korea (Dated: January 5, 2018) We have investigated the polarization dependence of the generation and detection of radial breath- ing mode (RBM) coherent phonons (CP) in highly-aligned single-walled carbon nanotubes. Using polarization-dependent pump-probe differential-transmission spectroscopy, we measured RBM CPs as a function of angle for two different geometries. In Type I geometry, the pump and probe po- larizations were fixed, and the sample orientation was rotated, whereas, in Type II geometry, the probe polarization and sample orientation were fixed, and the pump polarization was rotated. In both geometries, we observed a very nearly complete quenching of the RBM CPs when the pump polarization was perpendicular to the nanotubes. For both Type I and II geometries, we have de- veloped a microscopic theoretical model to simulate CP generation and detection as a function of polarization angle and found that the CP signal decreases as the angle goes from 0◦ (parallel to the tube) to 90◦ (perpendicular to the tube). We compare theory with experiment in detail for RBM CPs created by pumping at the E44 optical transition in an ensemble of single-walled carbon nanotubes with a diameter distribution centered around 3 nm, taking into account realistic band structure and imperfect nanotube alignment in the sample. PACS numbers: 78.67.Ch, 63.22.+m, 73.22.-f, 78.67.-n I. INTRODUCTION The one-dimensionality of single-walled carbon nan- otubes (SWNTs) is attractive from both fundamental and applied points of view, where the 1D confinement of electrons and phonons results in unique anisotropic electric, magnetic, mechanical, and optical properties.1,2 Individualized SWNTs, both single-tube and in ensem- ble samples, have shown anisotropy with polarized Ra- man scattering and optical absorption measurements where maximum signals result when the nanotube axis is aligned parallel to the polarization of incident light.3–8 Additionally, due to strong anisotropic magnetic sus- ceptibilities, both semiconducting and metallic SWNTs align well within an external magnetic field, and with the added properties of the Aharonov-Bohm effect, the elec- tronic band structure of SWNTs respond anisotropically with the strength of a tube-threading magnetic flux.9–16 Such optical and magnetic anisotropy is also expected with bulk samples, but detailed measurements showing extreme anisotropy have been lacking.17 Here we investigate anisotropic optical and vibrational properties of a bulk film of highly-aligned SWNTs us- ing polarization-dependent coherent phonon (CP) spec- troscopy. CP spectroscopy is an ultrafast pump-probe technique that complements CW Raman spectroscopy, and although both techniques provide information about electron-phonon coupling, CP spectroscopy avoids the common disadvantages of Raman spectroscopy that in- clude detection of Rayleigh scattering and photolumi- nescence and the broadening and blending of peak features.18 Such advantages are useful when investigat- ing SWNTs, where a large majority of samples include ensembles dispersed in various environments and their optical properties are obscured by the collection of vary- ing species of nanotubes. Recent CP studies on SWNTs have produced direct observation of CP oscillations of both the radial breath- ing mode (RBM) and G-band phonons, in addition to their phase information and dephasing times.18–24 Fur- thermore, when pulse-shaping techniques are combined with CP spectroscopy, predesigned trains of femtosec- ond pulses selectively excite RBM CPs of a specific chirality, avoiding inhomogeneous broadening from the ensemble.23 However, it is important to note that pre- vious CP studies investigated randomly-aligned SWNT samples, and as the quasi-1D nature of SWNTs leads to optical anisotropy that is dominant in the polarization dependence of PL, absorption, and Raman scattering, CP studies on aligned SWNTs are necessary. Below, we present results of a detailed experimental and theoretical study of CPs in highly-aligned SWNT thin films and found a very strong polarization anisotropy of the RBM as a function of angle. In particular, we ob- served a very nearly complete quenching of the RBM when the optical polarization is perpendicular to the e c n a b r o s b A e c n a b r o s b A 3.0 2.5 2.0 1.5 1.0 0.5 1.0 0.8 0.6 0.4 0.2 0.0 (a) A// A⊥ 1.6 2.4 2.0 Energy (eV) 2.8 3.2 (b) 1 2 A// A⊥ 4 3 Energy (meV) 5 6 7 2 ! Sapphire Type I V V (a) SWNTs H pump probe (b) SWNTs H ! FIG. 1: (color online) Absorption spectra of aligned single- walled carbon nanotubes in the (a) near-infrared to visible and (b) far-infrared (or terahertz) ranges for parallel (Ak) and perpendicular (A⊥) polarization. tubes. We have developed a theoretical model to un- derstand this extreme anisotropy, including band struc- ture, interband optical transition elements, ultrafast car- rier and phonon dynamics, and imperfect nanotube align- ment in the sample. Fitting our results to theory, we also calculated the nematic order parameter, S,15–17 i.e., the degree of alignment of SWNTs in the sample. II. SAMPLES AND EXPERIMENTAL METHODS We investigated CPs through degenerate pump-probe spectroscopy measurements on highly-aligned SWNT thin films transferred onto sapphire substrates. Pat- terned, vertically aligned SWNT arrays grown by chemi- cal vapor deposition were subsequently etched with H2O vapor to allow transffer to sapphire via a dry contact transfer printing technique, which forms horizontally- aligned SWNT thin films.25,26 The resulting film pro- duces aligned nanotubes of the same length, with a di- ameter distribution centered around 3 nm.26 Figure 1(a) shows polarization-dependent absorption spectra of a typical aligned sample in the visible to near-infrared range, while Fig. 1(b) shows absorption spectra in the far-infrared or terahertz (THz) range. The polarization Sapphire pump probe Type II FIG. 2: (color online) Scanning electron microscopy image of aligned single-walled carbon nanotubes, and the two exper- imental configurations employed in the current pump-probe spectroscopy work. (a) Type I: pump and probe polariza- tions are fixed and sample orientation is rotated. (b) Type II: probe and sample orientations are fixed and pump polar- ization is rotated. anisotropy is obvious in both (a) and (b), but extreme in the THz regime (b), as there is virtually zero ab- sorption when the sample is perpendicular to the THz polarization.17 Therefore, with such an aligned sample, our polarization measurements can be extended to in- clude the sample as an additional rotation parameter, compared to our earlier study of CP polarization depen- dence in randomly-aligned SWNTs.21 Using a mode-locked Ti:Sapphire laser with ∼80 fs pulse width that is shorter than the period of excited CP oscillations and ∼40 mW average pump power, the laser was tuned to central wavelength of 850 nm (1.46 eV) to predominantly excite the E44 interband transitions of the SWNTs in the sample. A shaker-delay system and balance detector was employed for fast-scan, real-time observation of CPs. Extraction of CP oscillations from raw pump-probe time-domain signal were performed as previously described.18 As depicted in Fig. 2, two types of polarization measurements were investigated. Type I configuration maintained the same polarization for the 60 50 40 30 20 10 6 − 0 1 × / T T ∆ 0 0.5 (a) Type I q = 0° (b) Type II f = 0° 15° 30° 45° 60° 90° 15° 30° 45° 60° 90° ) s t i . n u b r a ( y t i s n e t n I 1.5 1.0 Time (ps) 2.0 0.5 1.0 1.5 Time (ps) 2.0 3 Type II λ = 850 nm f = 0° 30° 45° 60° 90° FIG. 3: (color online) Experimental differential transmission data in the time domain showing coherent phonon oscillations of the radial breathing mode for different polarization angles in (a) Type I and (b) Type II configurations (see Fig. 1). The traces are vertically offset for clarity. pump and probe, while the alignment axis of the sample was rotated. Type II configuration maintained the same orientation for the probe and sample, while the pump polarization was rotated. For Type II measurements, a half-wave plate provided 90-degree rotation of the pump. All measurements were performed at room temperature. III. EXPERIMENTAL RESULTS Results of our polarization-dependent ultrafast pump- probe differential transmission measurements on the aligned SWNT film for both Type I and Type II ori- entations are shown in Fig. 3. Here, it is seen that the typical differential transmission amplitude of the RBM CP oscillations is on the order of 10−6, with a CP de- cay time of ∼1.5 ps. This short decay time, compared to our earlier studies on individually-suspended SWNTs in solution,18,21,23 is expected with our sample of bun- dled SWNTs. As the polarization angle is rotated, we see a strong polarization anisotropy of the RBM CPs as a function of angle, where the strongest oscillations are observed at 0◦ while the signal is very nearly completely quenched at 90◦ for both Type I and Type II geometries. Figure 4 shows CP spectra for different polarization angles for Type II configuration. To produce these spec- tra in the frequency domain, we calculated the Fourier transform (FT) of the time-domain CP oscillations. The frequencies of the CP oscillations have a wide range, from 50 to 200 cm−1, which is consistent with the large di- ameter distribution of nanotubes within the sample.26 The polarization anisotropy is clearly observed in the CP 0 100 200 Frequency (cm-1) 300 FIG. 4: (color online) Coherent phonon spectra for different polarization angles in Type II configuration obtained through Fourier transform of the time-domain data in Fig. 3(b). The traces are vertically offset for clarity. ) s t i n u . b r a ( y t i s n e n t I t d e a r g e t n I 1.0 0.8 0.6 0.4 0.2 0.0 0 (a) Type I (b) Type II 30 60 q (°) 90 0 30 60 f (°) 90 FIG. 5: (color online) Spectrally integrated coherent phonon intensity as a function of angle, measured in (a) Type I and (b) Type II configurations. spectra. Figures 5(a) and 5(b) plot the integrated CP in- tensity of the FT for all excited nanotubes as a function of θ [in Fig. 5(a)] and φ [in Fig. 5(b)], where θ (φ) is the angle of rotation for the Type I (Type II) configuration. IV. THEORY We have developed a microscopic theory for the gener- ation of coherent phonons in single-walled carbon nan- otubes and their detection in coherent phonon spec- troscopy experiments. Our microscopic theory is de- scribed in detail in Ref. 22, so we only summarize the main points here and indicate how our earlier work has been extended to include polarization-dependent coher- ent phonon spectroscopy. We treat the π and π∗ electronic states in (n,m) car- bon nanotubes using a third-nearest-neighbor extended tight binding (ETB) formalism developed by Porezag et al.27 for carbon compounds. Using a density functional based parametrization, Porezag et al. derived analytic expressions for the Hamiltonian and overlap matrix ele- ments that depend only on the C-C bond lengths. Using the ETB formalism, we obtain tight-binding wave func- tions and electronic energy levels Esµ(k) where s = c, v labels the conduction and valence bands, µ labels the cut- ting lines, and k is the one-dimensional nanotube Bril- louin zone.1 By exploiting the screw symmetry of the nanotube, we can block-diagonalize the electronic Hamil- tonian and overlap matrices into 2×2 blocks, one for each cutting line, as described in Appendix A of Ref. 22. We treat nanotube lattice dynamics using a seven- parameter valence-force-field model described in Ap- pendix B of Ref. 22. Exploiting the nanotube screw sym- metry, the dynamical matrix can be block-diagonalized into 6 × 6 blocks, one for each cutting line, which can be solved for the phonon displacement vectors and dis- persion relations ω2 βν(q). In the the dispersion relations, β = 1,··· , 6 labels the phonon modes, ν is an angu- lar momentum quantum number, and q is the phonon wave vector in the one-dimensional nanotube Brillouin zone. Following the work of Jiang et al.28 and Lobo and Martins,29 we include four types of force field potentials, i.e., the bond stretching, in-plane bond bending, out-of- plane bond bending, and bond twisting potentials. Our force-field potential energies are invariant under rigid ro- tations and translations (force constant sum rule), and, as a result, our phonon model correctly predicts the dis- persion relation for the long-wavelength flexure modes.22 The electron-phonon interaction is treated in a second- quantized formalism where the electron-phonon inter- action matrix elements are evaluated in Appendix C of Ref. 22. In calculating electron-phonon matrix ele- ments, we use 2pz graphene atomic wave functions and screened atomic potentials obtained from an ab initio calculation.28 We obtain equations of motion for coherent phonon amplitudes using the Heisenberg equations as described by Kuznetsov and Stanton.30 We assume that the opti- cal pulse and photoexcited carrier distributions are dis- tributed uniformly over the nanotube. In this case, only the ν = q = 0 phonon modes are excited. To excite RBM coherent phonons, the laser pulse must be short in comparison with the RBM phonon oscillation period. 4 For coherent RBM phonons, the coherent phonon ampli- tude is proportional to the tube diameter D(t). The tube diameter, being proportional to the coherent phonon am- plitude, satisfies a driven oscillator equation ∂2D(t) ∂t2 + ω2D(t) = S(t) (1) where ω is the angular frequency of the ν = q = 0 RBM phonon mode. The driving function S(t) for RBM co- herent phonons is given by S(t) ∝ −Xsµk Msµ(k)(cid:2)fsµ(k, t) − f 0 sµ(k)(cid:3) (2) where Msµ(k) is the driving function kernal for RBM co- herent phonons defined in Eq. (12) of Ref. 22 and fsµ(k, t) and f 0 sµ(k) are the time-dependent and initial equilibrium carrier distribution functions, respectively. From Eq. (2) we see that the driving function depends on the photoexcited carrier distribution functions. We treat photoexcitation of carriers in a Boltzmann equa- tion formalism and obtain the photogeneration rate us- ing Fermi's golden rule. In Ref. 22 we only considered linearly-polarized laser pulses with the electric polariza- tion vector parallel to the nanotube axis. For parallel po- larization, optical transitions only occur between states with the same cutting line index µ, and the resulting photogeneration rate is given by Eq. (13) in Ref. 22. In the present work, we consider linearly-polarized pump and probe beams in which the electric polariza- tion vector makes a finite angle with the nanotube axis. If ǫ is the unit electric polarization vector for the pump, then the photogeneration rate is now given by ∂fsµ(k) ∂t = (cid:12)(cid:12)(cid:12)(cid:12)gen 8π2e2 u(t)  n2 g (ω)2 (cid:18) 2 ×(cid:16)fs′µ′ (k, t) − fsµ(k, t)(cid:17) δ(cid:16)∆Eµµ′ m0(cid:19)Xs′µ′ ǫ · ~P µµ′ ss′ (k) − ω(cid:17) ss′ (k)2 (3) where ∆Eµµ′ ss′ (k) = Esµ(k) − Es′µ′ (k) are the k- dependent transition energies, ω is the pump photon energy, u(t) is the time-dependent energy density of the pump pulse, e is the electronic charge, m0 is the free electron mass, and ng is the index of refraction in the surrounding medium. To account for spectral broaden- ing of the laser pulses, the delta function in Eq. (3) was replaced with a Lorentzian lineshape31 δ(∆E − ω) → Γp/(2π) (∆E − ω)2 + (Γp/2)2 (4) In Eq. (3), ~P µµ′ ss′ (k) are the dipole-allowed optical ma- trix elements between the initial and final states where µ and µ′ are initial and final state cutting lines. For polar- ization parallel to the tube axis (taken to be z), µ′ = µ and z · ~P µµ ss′ (k) with the right hand side being defined in Eq. (14) of Ref. 22. For linear polarization ss′ (k) = P µ parallel to the x axis, µ′ = µ ± 1 and in the notation of Ref. 22, we have ss′ x · ~P µ,µ±1 ×XrJ (k) =  √2m0 1 2 Xr′ C ∗ r′ (s′, µ ± 1, k) Cr(s, µ, k) eiφJ(k,µ) (cid:0)Mx(r′, rJ) ± iMy(r′, rJ)(cid:1) Similarly, for linear polarization along the y axis, we have (5) (a) E44 ) 1 - m c 4 0 1 ( (38,0) E77 ) V e ( y g r e n E 1 0 -1 y · ~P µ,µ±1 ss′ (k) = ∓ i(cid:0)x · ~P µ,µ±1 ss′ (k)(cid:1) (6) -0.4 -0.2 0.0 0.2 0.4 8 6 4 2 0 5 (b) 00 150 300 450 600 900 E44 2 1 0 3 Photon Energy (eV) E44 (d) RBM 10.91 meV Type II (38,0) 0 30 60 90 Angle (deg.) In Eqs. (5) and (6), the atomic dipole matrix element vectors (which can be evaluated analytically) are given by M(r′, rJ) = Z dr ϕ∗ r′ 0(r − Rr′ 0) ∇ ϕrJ(r − RrJ) (7) where the 2pz orbitals ϕrJ are defined in Eq. (C3) of Ref. 22. In our experiments, a probe pulse is used to mea- sure the time-varying absorption coefficient. In our the- ory, the time-varying optical properties measured by the probe pulse are obtained from the imaginary part of the dielectric function. For a linearly polarized probe pulse, we have ) s t i n u . b r a ( r e w o P k ( /T) E44 E77 (c) ) s t i n u . b r a ( r e w o P d e a r g e n t t I 00 150 300 450 600 900 2 1 Pump Energy (eV) 3 ε2(ω, t) = 8π2e2 At(ω)2 (cid:18) 2 m0(cid:19) Xss′µµ′Z dk π ǫ · ~P µµ′ ss′ (k)2 ×(cid:16)fsµ(k, t) − fs′µ′ (k, t)(cid:17) δ(cid:16)∆Eµµ′ ss′ (k) − ω(cid:17) (8) where At = π(dt/2)2 is the cross-sectional area of the tube and dt is the equilibrium nanotube diameter. The dirac delta function is replaced by a broadened Lorentzian of the form shown in Eq. (4). The photon energy of the probe pulse photons is ω, and ǫ is the probe unit polarization vector. In our experiments, the photon energy ω is the same for both pump and probe, while the polarization vector ǫ may be different for pump and probe. To obtain the computed coherent phonon spectrum, we take the power spectrum of the computed differential transmission signal after background subtrac- tion using the Lomb periodogram algorithm described in Ref. 32. V. COMPARISON OF THEORY AND EXPERIMENT To compare experiment with theory, we performed simulations of polarization-dependent CP spectroscopy on a (38,0) zigzag nanotube. This is a mod-2 semicon- ducting tube with a diameter of 3.01 nm. Our sample contains an ensemble of nanotubes with diameters cen- tered around 3 nm,26 and we expect that (38,0) tubes FIG. 6: (color online) (a) Theoretical bandstructure for (38,0) nanotubes showing the strong E44 and E77 transitions for a polarization vector parallel to the tube. The wavevector k is expressed in units of π/T where T = 4.31 A is the length of the translational unit cell. (b) Computed absorption spectra for polarization angles varying from 0◦ (parallel to tube) to 90◦ (perpendicular to tube). (c) Coherent phonon spectra as a function of pump energy and pump polarization angle φ for RBM coherent phonons (ω = 10.91 meV). Coherent phonon spectra are Fourier transforms of computed time dependent differential transmission for Type II pump-probe experiments. (d) Integrated power (black dots) obtained by taking the area under the computed E44 peaks in lower left panel. We fit (red curve) the theoretical calculations to A cosp(φ) where A and p = 4.017 are fitting parameters. will contribute strongly to the CP signal for the ensem- ble since (i) the (38,0) tube has a diameter in the center of the measured diameter distribution and (ii) mod-2 zigzag tubes tend to have very strong intrinsic CP signals.18,22,23 Our theoretical results for the (38,0) nanotube are shown in Fig. 6. Our computed electronic π band struc- ture for the (38,0) tube is shown in Fig. 6(a). The bands are doubly degenerate with two distinct cutting lines cor- responding to each band. The π valence bands have neg- ative energy, and the π∗ conduction bands have posi- tive energy. For z-polarized light, the allowed E44 and E77 transitions (selection rule ∆µ = 0) giving rise to the strongest CP signals are indicated by vertical arrows. The absorption coefficient for (38,0) nanotubes as a function of linearly-polarized photon energy is shown in Fig. 6(b) for polarization angles varying from 0◦ (parallel to tube) to 90◦ (perpendicular to tube). For the (38,0) nanotube, the RBM phonons have a computed energy ω = 10.91 meV and an oscillation period of 379 fs. We simulate the generation of RBM coherent phonons by pumping with a 50 fs pump pulse, which is much shorter than the RBM oscillation period. Polarization-dependent CP spectra for RBM coherent phonons are shown in Fig. 6(c) for pump polarization angles φ ranging from 0◦ to 90◦. In these simulations, the probe polarization is kept fixed parallel to the tube axis (Type II geometry). As can be seen in the figure, the CP signal is maximum when the pump is polarized paral- lel to the tube. As the pump photon energy is varied, we excite RBM coherent phonons by successively photoex- citing carriers in different bands. The strongest RBM CP signals are obtained by pumping at the E44 and E77 transitions. As the pump polarization angle increases, the RBM CP signal decreases until it is finally quenched when the pump polarization is perpendicular to the tube. By taking the area under the RBM CP signal curves in the vicinity of the computed E44 transition energy near 1.48 eV, we obtain the integrated power as a function of pump polarization angle shown as black dots in Fig. 6(d). The integrated CP power can be well fit with a fitting function of the form A cosp(θ), where A and p = 4.017 are fitting parameters. The polarization dependence of the RBM CP inte- grated power in Fig. 6(d) can be understood by examin- ing Fig. 7. In Fig. 7(b) we plot the coherent RBM diame- ter oscillations as a function of the pump polarization an- gle θ. Fitting the theoretical results to A cosp(θ), we ob- tain p = 2.11 as indicated in the figure. For the RBM di- ameter oscillations we get p ∼ 2 which makes sense. The driving function S(t) is proportional to the photogener- ated carrier density as seen in Eq. (2) and this in turn is proportional to the squared optical matrix element which has a cos2(θ) angular dependence3. In Fig. 7(a) we plot the amplitude of the time-dependent differential trans- mission oscillations measured by the probe beam after background subtraction. For the differential transmission signal, we get p ∼ 2 for Type II and p ∼ 4 for Type I. This also makes sense. In Type II experiments the probe polarization is parallel to the tube axis and the pump po- larization angle θ is varied. In this case the signal should be proportional to the amplitude of the diameter oscil- lations. In Type I experiments on the other hand, the pump and probe polarizations are parallel and we should get an extra factor of cos2(φ) to account for anisotropy of the probe beam absorption. In our CP spectroscopy measurements, we extract the power spectrum of the dif- ferential transmission oscillations. The CP power spec- trum should be proportional to the square of the am- plitude of the differential transmission signal which we anticipate will give us a cos4(φ) polarization dependence 6 (a) 1.64 (b) 3.65 Type I Type II (38,0) E44 2.11 1.0 0.5 0.0 0.04 0.02 ) . b r a ( e d u t i l p m A T T / ) A ( e d u t i l p m A M B R 0.00 -90 -60 RBM 10.91 meV -30 30 0 Angle (degrees) 60 90 FIG. 7: (color online) In (b) we plot the amplitude of RBM coherent phonon oscillations as a function of the pump polar- ization angle (θ or φ) for a (38,0) nanotube photoexcited by a 50 fs pump at the theoretical E44 transition. In (a) we plot the amplitude of the resulting differential transmission signal for Type I and II experiments as a function of angles θ and φ respectively. Our results are fit to A cosp(θ) [A cosp(φ)] where numerical values of the best fit p are indicated in the figure. for Type II CP spectroscopy experiments and a cos8(θ) dependence for Type I CP spectroscopy experiments. The experimental integrated CP power is obtained by taking the Fourier transform power spectrum of the time- dependent differential transmission data shown in Fig. 3 and then computing the area under the E44 peak in the resulting spectrum. The results for our Type I and Type II experiments are shown in Fig. 8, where the experi- mental data points are plotted as downward-pointing red triangles and the corresponding theoretical predictions are shown as upward-pointing black triangles. In over- laying the experimental and theoretical data points, we subtracted a background from the experimental data to obtain complete quenching at 90◦. We then rescaled ex- perimental and theoretical data (both in arbitrary units) so that the integrated power at 0◦ is equal to unity. We then fit experimental and theoretical data to functions of the form A cosp(θ) [A cosp(φ)]. The best fit functions for our experimental and theoretical results are shown in Fig. 8 as solid red and dashed black lines, respectively. For the Type II experiments, where the probe po- larization is fixed parallel to the average tube orienta- tion, we get decent agreement between theory and ex- periment with best fit parameters p = 4.017 for theory and p = 3.630 for experiment. For the Type I exper- iments, where pump and probe polarization vectors are parallel to each other, there is a discrepancy between the- ory and experiment. As can be seen in Fig. 8(a), the best ) s t i n u . b r a ( r e w o P d e t a r g e t n I 1.0 0.8 0.6 0.4 0.2 E44 (a) Type I pump probe e b u t Type II e b u t e b o r p (b) pump 8.308 Theory (38,0) 4.435 Expt 3.630 Expt 4.017 Theory (38,0) 0.0 0 30 60 30 Angle (degrees) 0 60 90 7 1.0 0.8 0.6 0.4 0.2 0.0 0 30 Standard Deviation Angle degrees) 0 10 20 30 (a) Type I A cos8( (b) Type II A cos4( 60 0 30 Angle (degrees) 60 90 ) s t i n u . b r a ( r e w o P d e t a r g e t n I FIG. 8: (color online) Integrated RBM coherent phonon power as a function of (a) θ for Type I and (b) φ for Type II experiments pumping close to the E44 transition. Experi- mental data points are red downward pointing triangles fit by a solid red line. Theoretical data points for a (38,0) nanotube are black upward pointing triangles and are fit by a black dashed line. The fitting functions are of the form A cosp(θ) and A cosp(φ) where the numerical values of the best fit p are indicated in the figure. The experimental curves have under- gone background subtraction to obtain complete quenching at an angle of 90◦ and then rescaled so the integrated power at an angle of 0◦ is equal to unity. fit parameters are p = 8.308 for theory and p = 4.435 for experiment. The experimental fits in Fig. 8 would imply that the CP intensity scales roughly as cos4(θ) indepen- dently of the probe polarization. This seems unlikely given the anisotropy of the absorption coefficient in car- bon nanotubes. In fact, from our theory, we would expect p ∼ 4 for Type II and p ∼ 8 for Type I. We believe that the cause of this discrepancy is most likely due to misalignment effects. Our sample consists of an ensemble of nanotubes lying horizontally on a sap- phire substrate. While the tubes are highly-aligned, they are not perfectly aligned. To consider the effects of mis- alignment on the experimentally-measured integrated CP intensity in a Type I experiment, we assume that the tube alignment angles on the transferred film are described by a Gaussian distribution function with a small standard deviation ∆θ. If the angle between the pump polariza- tion vector and the ensemble averaged tube axis is θ, then the angles ϑ between the pump polarization vector and the axes of each tube in the ensemble are described by a Gaussian distribution P (ϑ, θ, ∆θ) = 1 √2π(∆θ) exp(cid:18)− (ϑ − θ)2 2(∆θ)2 (cid:19) (9) If the integrated RBM coherent phonon power for each nanotube in the ensemble is given by A cosp(ϑ), the en- semble averaged integrated power Icp(θ, ∆θ) is obtained FIG. 9: (color online) Effects of tube misalignment on inte- grated coherent phonon power for an ensemble of nanotubes with tube axis orientation angles following a Gaussian distri- bution. (a) Type I fitting function A cosp(θ) with p = 8 and (b) Type II fitting function A cosp(φ) with p = 4. by taking the ensemble average over ϑ Icp(θ, ∆θ) = A Z ∞ −∞ dϑ P (ϑ, θ, ∆θ) cosp(ϑ) (10) In Eq. (10), we extended the integration limits on ϑ to infinity in the limit of small ∆θ. If p is a positive in- teger, Icp(θ, ∆θ) can be found analytically. Otherwise, it can be evaluated numerically. In the case of Type II experiments, the ensemble averaged integrated power is obtained by replacing θ with φ in Eqs. (9) and (10). The effects of nanotube misalignment on the integrated CP power fitting function for an ensemble of Gaussian misaligned tubes is illustrated in Fig. 9, assuming p = 8 for Type I and p = 4 for Type II experiments. As the standard deviations ∆θ [∆φ] increase, quenching of the CP intensity is reduced and Icp(θ, ∆θ) [Icp(φ, ∆φ)] become flatter. Using this model, we are able to get reasonable fits to the experimentally measured CP intensity in both Type I and Type II experiments. We assume a fitting function of the form A cosp(θ + ∆θ) + B [A cosp(φ + ∆φ) + B], where A and B are background subtraction and rescaling pa- rameters and ∆θ [∆φ] is a random Gaussian distributed misalignment angle (standard deviation ∆θ [∆φ]). In our fitting procedure, we set p = 8 for Type I and p = 4 for Type II and use the same standard deviation for the tube misalignment angles in fitting both Type I and II data. The results of our fitting procedure are shown in Fig. 10, where the best fits for Type I and Type II geometries are shown as dashed black and solid red lines, respectively. Our best fit standard deviation is ∆θ = ∆φ = 18.7◦, which implies that the tube align- ment angles on the sapphire substrate are 0◦ ± 9.35◦. ) s t i n u . b r a ( r e w o P d e t a r g e t n I 1.0 0.8 0.6 0.4 0.2 0.0 0 p = 8 Type I pu m p pro b e e b u t p = 4 Type II pu m p e b o r p e b u t Type I Type II 30 E44 Experiment 90 60 120 Angle (degrees) 150 180 FIG. 10: (color online) Experimental integrated RBM coher- ent phonon power for the E44 transition as a function of θ for Type I (black upward pointing triangles) and φ for Type II (red downward pointing triangles) experiments. Type I exper- iments are fit to A cosp(θ+∆θ)+B where p = 8 (black dashed line) and Type II experiments are fit to A cosp(φ+∆φ)+B for p = 4 (solid red line). A and B are background subtraction and rescaling parameters and the standard deviations for the random tube axis misalignment ∆θ and ∆φ are restricted to be the same for both Type I and II. For small ∆θ (measured in radians) the nematic order parameter is S = exp(−2 (∆θ)2) = 0.81. Although a value of S = 0.81 indicates a strongly aligned sample, it is not perfectly aligned (S = 1). This is unlike previous work in the THz regime with these highly-aligned samples, where the nematic order param- eter was calculated to be exactly S = 1.17 It is clear that there is a wavelength dependence of the nematic order parameter, but we believe this can be explained quali- tatively: any slight misalignment in the sample would go undetected in the THz regime, as the wavelength is much larger than our visible wavelengths that can de- tect the misalignments. This could explain the differ- ence in calculated nematic order parameter between our 8 CP measurements and previous THz measurements, but nonetheless, regardless of the calculated differences with wavelength, our calculated nematic order parameter is still quite large and we can confidently confirm the high degree of alignment of the sample with our CP measure- ments. VI. CONCLUSION In summary, we investigated the polarization anisotropy of coherent phonon dynamics in highly- aligned single-walled carbon nanotubes and measured RBM coherent phonons as a function of polarization angle. We saw a very nearly complete quenching of the RBM for both geometries and extended our results to determine the degree of alignment of the sample. Comparing our results with theory, we performed simulations of polarization-dependent CP spectroscopy on a (38,0) zigzag nanotubes and also found a decrease in CP signal as optical polarization varies from parallel to perpendicular to the nanotube axis. Using those simulated results, we also theoretically determined a cos8(θ) dependence for Type I CP spectroscopy experi- ments and a cos4(φ) polarization dependence for Type II CP spectroscopy experiments. Including misalignment effects to our fitting, we finally determined the nematic order parameter of our sample to be 0.81. Acknowledgments This work was supported by the National Science Foundation under grant numbers DMR-0325474, OISE- 0530220, and DMR-0706313, the Robert A. Welch foun- dation under grant number C-1509, and the Office of Naval Research (ONR) under contract number 00075094. Y.-S. Lim and K.-J. Yee are supported by a Korea Science and Engineering Foundation (KOSEF) grant funded by the Korean Government (Most) (R01-2007-000-20651-0). We acknowledge useful discussions with Andrew Rinzler at the University of Florida. ∗ [email protected]; www.ece.rice.edu/~kono; corresponding author. 1 R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physi- cal properties of carbon nanotubes (World Scientific, Sin- gapore, 2003). 2 J.-C. Charlier, X. Blase, and S. Roche, Rev. Mod. Phys. 79, 643 (2007). M. S. S. Dantas, M. A. Pimenta, A. M. Rao, R. Saito, C. Liu, and H. M. Cheng, Phys. Rev. Lett. 85, 2617 (2000). 6 Z. M. Li, Z. K. Tang, H. J. Liu, N. Wang, C. T. Chan, R. Saito, S. Okada, G. F. Li, J. S. Chen, N. Nagasawa, et al., Phys. Rev. Lett. 87, 127401 (2001). 7 A. Hartschuh, H. N. Pedrosa, L. Novotny, and T. D. Krauss, Science 301, 1354 (2003). 3 H. H. Gommans, J. W. Alldredge, H. Tashiro, J. Park, J. Magnuson, and A. G. Rinzler, J. Appl. Phys. 88, 2509 (2000). 4 G. S. Duesberg, I. Loa, M. Burghard, K. Syassen, and S. Roth, Phys. Rev. Lett. 85, 5436 (2000). 8 M. F. Islam, D. E. Milkie, C. L. Kane, A. G. Yodh, and J. M. Kikkawa, Phys. Rev. Lett. 93, 037404 (2004). 9 H. Ajiki and T. Ando, J. Phys. Soc. Jpn. 62, 2470 (1993). 10 J. P. Lu, Phys. Rev. Lett. 74, 1123 (1995). 11 M. A. L. Marques, M. d'Avezac, and F. Mauri, Phys. Rev. 5 A. Jorio, G. Dresselhaus, M. S. Dresselhaus, M. Souza, B 73, 125433 (2006). 12 S. Zaric, G. N. Ostojic, J. Kono, J. Shaver, V. C. Moore, R. H. Hauge, R. E. Smalley, and X. Wei, Nano Lett. 4, 2219 (2004). 13 S. Zaric, G. N. Ostojic, J. Kono, J. Shaver, V. C. Moore, M. S. Strano, R. H. Hauge, R. E. Smalley, and X. Wei, Science 304, 1129 (2004). 14 M. F. Islam, D. E. Milkie, O. N. Torrens, A. G. Yodh, and J. M. Kikkawa, Phys. Rev. B 71, 201401 (2005). 15 J. Shaver, A. N. G. Parra-Vasquez, S. Hansel, O. Portugall, C. H. Mielke, M. von Ortenberg, R. H. Hauge, M. Pasquali, and J. Kono, ACS Nano 3, 131 (2009). 16 T. A. Searles, Y. Imanaka, T. Takamasu, H. Ajiki, J. A. Fagan, E. K. Hobbie, and J. Kono, arXiv (2010), 1001.0524v1. 17 L. Ren, C. L. Pint, L. G. Booshehri, W. D. Rice, X. Wang, D. J. Hilton, K. Takeya, I. Kawayama, M. Tonouchi, R. H. Hauge, et al., Nano Lett. 9, 2610 (2009). 18 Y.-S. Lim, K.-J. Yee, J.-H. Kim, E. H. H´aroz, J. Shaver, J. Kono, S. K. Doorn, R. H. Hauge, and R. E. Smalley, Nano Lett. 6, 2696 (2006). 19 A. Gambetta, C. Manzoni, E. Menna, M. Meneghetti, G. Cerullo, G. Lanzani, S. Tretiak, A. Piryatinski, A. Sax- ena, R. L. Martin, et al., Nature Physics 2, 515 (2006). 20 K. Kato, K. Ishioka, M. Kitajima, J. Tang, R. Saito, and H. Petek, Nano Lett. 8, 3102 (2008). 21 J.-H. Kim, J. Park, B. Y. Lee, D. Lee, K.-J. Yee, Y.-S. Lim, L. G. Booshehri, E. H. H´aroz, J. Kono, and S.-H. Baik, J. Appl. Phys. 105, 103506 (2009). 9 22 G. D. Sanders, C. J. Stanton, J.-H. Kim, K.-J. Yee, Y.-S. Lim, E. H. H´aroz, L. G. Booshehri, J. Kono, and R. Saito, Phys. Rev. B 79, 205434 (2009). 23 J.-H. Kim, K.-J. Han, N.-J. Kim, K.-J. Yee, Y.-S. Lim, G. D. Sanders, C. J. Stanton, L. G. Booshehri, E. H. H´aroz, and J. Kono, Phys. Rev. Lett. 102, 037402 (2009). 24 Y.-S. Lim, J.-G. Ahn, J.-H. Kim, K.-J. Yee, T. Joo, S.-H. Baik, E. H. H´aroz, L. G. Booshehri, and J. Kono, ACS Nano (2010), published online on May 13, 2010. 25 C. L. Pint, Y.-Q. Xu, M. Pasquali, and R. H. Hauge, ACS Nano 2, 1871 (2008). 26 C. L. Pint, Y.-Q. Xu, S. Moghazy, T. Cherukuri, N. T. Alvarez, E. H. H´aroz, S. Mahzooni, S. K. Doorn, J. Kono, M. Pasquali, et al., ACS Nano 4, 1131 (2010). 27 D. Porezag, T. Frauenheim, T. Kohler, G. Seifert, and R. Kaschner, Phys. Rev. B 51, 12947 (1995). 28 J.-W. Jiang, H. Tang, B.-S. Wang, and Z.-B. Su, Phys. Rev. B 73, 235434 (2006). 29 C. Lobo and J. L. Martins, Z. Phys. D 39, 159 (1997). 30 A. V. Kuznetsov and C. J. Stanton, Phys. Rev. Lett. 73, 3243 (1994). 31 S. L. Chuang, Physics of Optoelectronic Devices (Wiley, New York, 1995). 32 W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes (Cambridge University Press, New York, 1992). 

1607.04802

2

1607

Heat production and error probability relation in Landauer reset at effective temperature

The erasure of a classical bit of information is a dissipative process. The minimum heat produced during this operation has been theorized by Rolf Landauer in 1961 to be equal to $k_B T \ln 2$ and takes the name of Landauer limit, Landauer reset or Landauer principle. Despite its fundamental importance, the Landauer limit remained untested experimentally for more than fifty years until recently when it has been tested using colloidal particles and magnetic dots. Experimental measurements on different devices, like micro-mechanical systems or nano-electronic devices are still missing. Here we show the results obtained in performing the Landauer reset operation in a micro-mechanical system, operated at an effective temperature. The measured heat exchange is in accordance with the theory reaching values close to the expected limit. The data obtained for the heat production is then correlated to the probability of error in accomplishing the reset operation.

cond-mat.mes-hall

cond-mat

Heat production and error probability relation in Landauer reset at effective temperature Igor Neri1,2,*,+ and Miquel L ´opez-Su´arez1,!,+ 1NiPS Laboratory, Dipartimento di Fisica e Geologia, Universit`a degli Studi di Perugia, 06123 Perugia, Italy 2INFN Sezione di Perugia, via Pascoli, 06123 Perugia, Italy *[email protected] [email protected] +these authors contributed equally to this work ABSTRACT The erasure of a classical bit of information is a dissipative process. The minimum heat produced during this operation has been theorized by Rolf Landauer in 1961 to be equal to kBT ln 2 and takes the name of Landauer limit, Landauer reset or Landauer principle. Despite its fundamental importance, the Landauer limit remained untested experimentally for more than fifty years until recently when it has been tested using colloidal particles and magnetic dots. Experimental measurements on different devices, like micro-mechanical systems or nano-electronic devices are still missing. Here we show the results obtained in performing the Landauer reset operation in a micro-mechanical system, operated at an effective temperature. The measured heat exchange is in accordance with the theory reaching values close to the expected limit. The data obtained for the heat production is then correlated to the probability of error in accomplishing the reset operation. Introduction The minimum energy required to reset one bit of information represents one of the fundamental limits of computation arising when one bit of information is erased or destroyed. Starting with a system encoding two possible states with the same probability and finishing the procedure with only one possible state, the variation of entropy in the system is equal to the difference of entropy between the final state and the initial one, D S = kB ln 1 − kB ln 2 = −kB ln 2. On average this reduction of entropy has to be accompanied with an increase of heat in the surrounding environment in order to not violate the second law of thermodynamics, QL ≥ kBT ln 2. This limit takes the name of Landauer limit and was theorized by Rolf Landauer in the 60's1, and remained untested for over fifty years. The development of computational methods allowed to evaluate tiny amounts of heat exchanged and recent technical advances in micro- and nano-fabrication made possible to recently test the validity of the Landauer principle. In particular, the first experimental verification of the Landauer principle has been carried out by B´erut et al. using a colloidal particle trapped in optical tweezers2. More recently Jun et al. presented similar results considering a colloidal particle in a feedback trap3. Finally, the last experimental verification of the Landauer limit has been performed by Hong et al. in a nanomagnetic system4. In this case information is encoded in the magnetization state of the system while an external magnetic field is used in order to flip between two preferred magnetization states. The evaluation of the heat produced during the reset operation was demonstrated to be compatible with the Landauer limit. Even if there is no doubt nowadays on the validity of the Landauer principle, the test on micro-mechanical systems is still missing. We have recently shown that it is possible to use electro-mechanical devices to accomplish basic5 and complex logic operations6 with an arbitrarily low energy expenditure. Therefore the possibility to use this class of devices for memory storage completes the logic architecture for a complete computing device. In the following we show the measurement on heat production when performing the reset operation in a novel memory unit based on a bistable mechanical cantilever at effective temperature. The results are in good agreement with the Landauer limit. The dependence of the dissipation with the error rate is also investigated showing a trend in accordance with the theory10. Results Bistable micro-mechanical system The mechanical system used to perform the experiment is depicted in Figure 1 (a). A triangular micro-cantilever, 200µm long, is used to encode one bit on information. In order to obtain two stable states two magnets with opposite magnetization are placed on the tip of the cantilever and on a movable stage facing the cantilever. In this way, depending on the distance between the magnets, d, and the relative lateral alignment, D x, it is possible to induce bistability on the system. Figure 1 (b) shows the potential energy as a function of d reconstructed from the probability density function of the position of the cantilever at equilibrium, r (x,d) = A exp(−U(x,d)/kBT ), which implies that U(x,d) = −kBT lnr (x,d) + U0 2. When the magnets are far away the effect of the repulsive force is negligible, the system is then monostable and can be approximated to a linear system. Decreasing the distance the repulsive force between magnets tends to soften the system up to the point where two stable positions appear. The effect of reducing even more d is to enlarge the separation of the rest states and to increase the potential barrier separating these two wells. Eventually, when the distance between the magnets is small enough, the system remains trapped in one well for a period of time larger than the relaxation time of the system. Logic states are encoded in the position of the cantilever tip: logic 0 for x < 0 and logic 1 for x > 0. The proposed system presents intrinsic dissipative processes that depend on the maximum displacement of the cantilever tip6. The minimum heat produced when performing a physical transformation of the system is proportional to x2 max. In our setup it is not possible to reduce the separation between the two potential wells to a value that bounds the heat produced by intrinsic dissipation below the Landauer limit. Increasing the effective temperature increases the value for the Landauer limit making possible to have negligible intrinsic dissipation. A piezoelectric shaker is used to excite the structure with a band limited white Gaussian noise to mimic the effect of an arbitrary temperature. In the present experiment the white noise is limited to 50kHz, well above the resonance frequency of the free cantilever ( f0=5.3kHz). The dependence of the effective temperature with the root-mean-squared voltage supplied to the shaker is reported in Figure 1 (c). The red dot, corresponding to an effective temperature of Teff = 5 × 107 K, highlights the condition considered in the present case. The solid line represent the expected trend where T (cid:181) V 2 7. The effective rms temperature has been estimated computing the power spectral density (PSD) of the system at various piezoelectric noise excitation voltages. The obtained curves have been fitted with Lorenzian curves taking as reference for the calibration the one at room temperature (T =300K). The other curves have been used to extract the only varying parameter Teff, corresponding to the effective temperature of the system under external excitation. Finally two electrostatic probes, placed one on the left and the other on the right of the cantilever, are used to apply a negative and positive forces respectively. When a voltage different to zero is applied on one probe the cantilever feels an attractive electrostatic force toward the probe due to the polarization of the cantilever itself. The voltage on the probes, the distance between the magnets and their time evolution are used to specify the protocols used in order to change the bit stored in the system as described in the following subsection. Reset protocol Varying the magnets distance d the barrier separating the two stable wells varies from the minimum value B0 for d=3.65 a.u. to Bmax for d=2.8 a.u. Applying a voltage on one probe corresponds to apply a force toward the probe itself. Thus a voltage on the left probe corresponds to a negative force and a voltage on the right probe corresponds to a positive force. Details of the force calibration are presented in methods section. In Figure 2 (a) the protocol followed to reset the bit to the state 0 and 1 is presented. The procedure is similar to the ones presented in Refs. 8 and 2. Initially the barrier separating the two stable states is removed moving the magnet away (red curve in the first panel) making the system monostable. Once the barrier is removed we apply a negative (positive) force to reset the bit status to 0 (1) applying a finite voltage VL (VR) on the left (right) probe. This is represented by the magenta curve in Figure 2 (a). Once the force is applied we restore the barrier to its original value. Finally, we remove the lateral force finishing in the original parameters configuration. At the end of the operation, if there are no errors the cantilever position encodes the desired bit of information. In order to be sure to perform the operation starting from both initial states, we mimic a statistical ensemble where the initial probability is 50% to start in the left well and 50% in the right well. A trace of the cantilever tip position, x, is shown in Figure 2 (a) (black line), where the dashed blue lines represent the two stable positions once the barrier is restored. When the barrier is removed the system goes from the local prepared state to an undefined state with a free expansion, the entropy on the system thus increases in a irreversible manner8, 9. This increment is related to uncontrollable transitions from one well to the other once the barrier height is close to kBT . These large excursions of the cantilever can be seen in the time series of position. Figure 2 (b) shows a representation of the time evolution of the potential energy during the set of the bit to the logic state 1. In a first step the barrier is removed allowing the system to oscillate in a monostable potential landscape. Then the potential is slightly tiled and when the barrier is restored the system is confined in the desired state. After these stages the bit is set to the state 1. In order to have a reliable measure of the heat produced during the considered operations, reset protocols are repeated for 800 times in order to have a large enough statistic. Heat vs error probability In the optimal case the initial configuration is a mixed logical 0 and 1 where both states have the same probability while the final configuration is the selected state with a 100% probability. This corresponds to an entropy variation of D S = −kB ln(2) and a minimum heat produced of Q ≥ −TD S = QL. It has been shown that the bound QL = −kBT ln(2) applies only for 11. In symmetric potentials. Considering asymmetries on the system the minimum produced heat can be lowered below QL our setup the system is slightly asymmetric and we have evaluated the variation of entropy from the initial to the final state 2/8 from the probability density function of the tip position, r (x), being D SG = −0.61kB D SS = −0.68kB for the Gibbs and Shannon entropy respectively, both close to −kB ln(2). If we consider the possibility to commit errors during the reset operation the heat produced becomes a function of the probability of success10: Q(Ps) ≥ kBT [ln(2) + Ps ln(Ps) + (1 − Ps)ln(1 − Ps)] (1) where Ps is the probability of success or success rate. When Ps is 0.5 no reset operation is performed and thus there is no minimum heat to be produced during the operation3, 10. In Figure 3 (a) we present the average heat produced for the reset operation as function of the lateral alignment of the counter magnet D x. When the system is aligned closely to perfection (i.e., D x ≈ 0) we estimate a heat production slightly above kBT and below two times QL. Asymmetrizing the potential, by means of setting D x 6= 0, the heat produced tends to decrease reaching values this time below QL. However, in this conditions the error rate in performing the reset operation have a major role, in fact in this configuration the probability of success, Ps, decrease rapidly. This is represented by the color map of dots in Figure 3 (a), where green represents higher success rate while blue represents a higher probability of error. In Figure 3 (b) the success rate of the reset operation is reported as function of the lateral alignment. Solid violet circles represent the overall success rate while red and black symbols represent the error rate for resetting to 1 or to 0 respectively. Circles are used to report the error probability for the same initial and final state while crosses are used for 0 to 1 and 1 to 0 transitions. For instance let us consider the case where D x < 0: the counter magnet is moved towards the right and as a consequence the 0 state is more favorable respect the 1 state. From Figure 3 (b) we can see that for D x < 0 the probability of resetting toward 0 is almost 100% while the probability of resetting toward 1 decreases rapidly reaching values below 50%. The same behavior is present in the case D x > 0, where the counter magnet is moved to the left, where the state 1 is more favorable. We can now correlate the heat produced to the probability of success for resetting to 0 and 1 as presented in Figure 3 (c). Dashed lines represent the Landauer limit for a 100% of success rate (≈ 0.7kBT ). While in both cases the heat produced is above the Landauer limit, in the reset to 0 case the obtained values are very close to QL. As expected, decreasing the success rate the obtained values goes below the Landauer limit for both cases accordingly to Equation 1. As it is well known the adiabatic limit in presence of dissipation mechanisms like viscous damping can be only reached if the operation is performed slowly when these mechanisms are negligible. We increased the protocol time for the reset operation from 0.25s up to 3.5s. The results are presented in Figure 3 (d). Increasing the protocol time decreases the heat production reaching values well below the Landauer limit. Notice that in these cases where Q < QL the Ps is well below 1 since the system has more time to relax and therefore tends to thermalize before the reset operation is correctly performed. In fact for protocols lasting more than 1 s the success rate is below 75%. Discussion We have measured the intrinsic minimum energy dissipation during the reset of one bit of information in a micro-mechanical system. We have considered a completely different physical system respect to the existing literature, i.e., micro-electro- mechanical system. To achieve these results we have performed the experiment at an effective temperature of 5 × 107 K in order to make the intrinsic dissipation of the mechanical structure negligible respect to the thermodynamic contribution. In these conditions we have reached values of heat produces consistent with the Landauer limit approaching it closely. Moreover we presented experimental data relating the minimum heat produced with the probability of success of resetting one bit of information. Nowadays where there is a lot of attention on micro-electro-mechanical systems able to perform computation at arbitrary low energy, the achieved results have a significant importance in the development of new computing paradigms based on systems different from the well established CMOS technology. Methods Setup preparation and calibration The micro-cantilever used is a commercial atomic force microscopy (AFM) probe (NanoWorld PNP-TR-TL12). It is long 200µm, with a nominal stiffness k=0.08Nm−1 and a nominal resonace frequency of 17kHz. A fragment of NdFeB (neodymium) magnet is attached to the cantilever tip with bi-component epoxy resin. To set the magnetization to a known direction the sys- tem is heated up to 670K, above its Curie temperature13 in the presence of a strong external magnetic field with the desired orientation. With this additional mass the resonance frequency decreases to 5.3kHz. The quality factor of the system has been estimated from the power spectral density of the displacement, x, fitted with a Lorenzian curve, giving a value of Qf = 320. The deflection of the cantilever, x, is determined by means of an AFM-like laser optical lever. A small bend of the cantilever provokes the deflection of a laser beam incident to the cantilever tip that can be detected with a two quadrants photo detector. 3/8 The laser beam is focused on the cantilever tip with an optical lens (focal length f =50mm). For small deflections the response of the photo detector remains linear, thus x = rxD VPD, where VPD is the voltage difference generated by the two quadrants of the photo detector. In order to determine rx we look at the frequency response of the system as in6 under thermal excitation. The relation between the measured voltage and the expected displacement gives rx =1.8365 × 10−5m V−1. The system is placed in a vacuum chamber and isolated from seismic vibrations to maximize the signal-to-noise ratio. All measurements were performed at pressure P =4.7 × 10−2 mbar. Effective temperature estimation As the system is modeled as a harmonic oscillator with one degree of freedom, according to the Equipartition Theorem the thermal energy present in the system is simply related to the cantilever fluctuations as 1 2 khx2i. According to Parseval's 0 G(w )2 dw where G(w ) stands for the PSD of the system. This takes the form of a Lorenzian function 2 kBT = 1 theorem hx2i =R G(w ) =s 4kBT Qfkw 0vuut (cid:16)1 − w 2 w 2 0(cid:17)2 1 + 1 Q2 f w 2 w 2 0 where w = 2p only free parameter, T , from the measured PSDs to the expected function G(w ). f . Once the system has been calibrated at room temperature we estimated the effective temperatures fitting the Force calibration A set of two electrodes (see Fig. 1(a)) is used in order to polarize the cantilever producing a bend on the mechanical structure. Electrostatic forces depend on the voltage applied to the electrodes, i.e. VL and VR for the left and right electrode respectively. Since the restoring force of the cantilever can be expressed as Fk = −kx, the relation between applied voltage to the probe and the force acting on the cantilever has been estimated in static conditions assuming Fk = Fel. The relation between the applied voltages, VL and VR, and the electrostatic force Fel is then fitted with a 9th degree polynomial. Heat production evaluation The work performed on the system along a given trajectory x(t) is given by the integral14, 15: W =Z 0 t p M(cid:229) k=1 ¶ U(x,lll ) ¶l k ¶l k ¶ t dt (2) where t p is the protocol time duration, U(x,lll ) is the total potential energy of the system and lll is a vector containing all the M control parameters. In our case we have two controls parameter, the voltage applied to the piezoelectric stage to control the energy barrier, and the electrostatic forces. To obtain the heat produced Q we have to consider the variation in internal energy, D E. Ideally the potential energy at both the bottom wells is the same, however considering asymmetries on the system D E can be different from zero. The evaluation of the total energy variation D E is obtained from the reconstructed potential energy, U, and the variation of the kinetic energy. The latter quantity is however negligible even for the shortest t p, where the total kinetic energy variation is one order of magnitude lower than QL (1.4 × 10−17J versus 5.3 × 10−16 J). Finally the heat produced is obtained from Q = W − D E. References 1. Landauer, R. Irreversibility and heat generation in the computing process. IBM journal of research and development 5, 183 -- 1911961. 2. B´erut, A. et al. Experimental verification of Landauer's principle linking information and thermodynamics. Nature 483, 187 -- 189 (2012). 3. Jun, Y., Gavrilov, M. & Bechhoefer, J. High-precision test of Landauer's principle in a feedback trap. Physical review letters 113, 190601 (2014). 4. Hong, J., Lambson, B., Dhuey, S. & Bokor, J. Experimental test of Landauer's principle in single-bit operations on nanomagnetic memory bits. Science Advances 2 (2016). 5. Lopez-Suarez, M., Neri, I. & Gammaitoni, L. Operating micromechanical logic gates below kBT: Physical vs logical reversibility. In Energy Efficient Electronic Systems (E3S), 2015 Fourth Berkeley Symposium on, 1 -- 2 (IEEE, 2015). 6. Lopez-Suarez, M., Neri, I. & Gammaitoni, L. Sub kBT micro electromechanical irreversible logic gate. Nature Commu- nication 7 12068 (2016). 4/8 ¥ 7. Venstra, W. J., Westra, H. J. & van der Zant, H. S. Stochastic switching of cantilever motion. Nature communications 4 (2013). 8. Bennett, C. H. The thermodynamics of computation -- a review. International Journal of Theoretical Physics 21, 905 -- 940 (1982). 9. Crooks, Gavin E. Entropy production fluctuation theorem and the nonequilibrium work relation for free energy differences. Physical Review E 60.3 2721 (1999). 10. Gammaitoni, L., Chiuchi´u, D., Madami, M. & Carlotti, G. Towards zero-power ICT. Nanotechnology 26, 222001 (2015). 11. Sagawa, Takahiro. Thermodynamic and logical reversibilities revisited. Journal of Statistical Mechanics: Theory and Experiment 2014.3 P03025 (2014) 12. NanoWorld - AFM tip - PNP-TR-TL - Pyrex-Nitride. http://www.nanoworld.com/pyrex-nitride-triangular-silicon-nitride- tipless-cantilever-afm-tip-pnp-tr-tl. Accessed: 2016-05-30. 13. Ma, B. et al. Recent development in bonded NdFeB magnets. Journal of magnetism and magnetic materials 239, 418 -- 423 (2002). 14. Seifert, U. Stochastic thermodynamics, fluctuation theorems and molecular machines. Reports on Progress in Physics 75, 126001 (2012). 15. Douarche, F., Ciliberto, S., Petrosyan, A. & Rabbiosi, I. An experimental test of the Jarzynski equality in a mechanical experiment. EPL (Europhysics Letters) 70, 593 (2005). Acknowledgements The authors gratefully acknowledge useful discussion with L. Gammaitoni. The authors acknowledge financial support from the European Commission (FPVII, Grant agreement no: 318287, LANDAUER and Grant agreement no: 611004, ICT- En- ergy). Author contributions statement M.L.S. and I.N. designed the experiment, performed the measurements, analyzed the measured data and contributed to the writing of the manuscript. Data availability statement The data that support the findings of this study are available from the corresponding author upon request. Additional information Competing financial interests The authors declare no competing financial interests. 5/8 Figure 1. Schematic of the whole system and measurement setup. Lateral view of the whole system and measurement setup. Two magnets with opposite magnetic orientations are used to induce bistability in the system. Two electrodes are used to apply electrostatic forces on the mechanical structure: VL and VR to force the cantilever to bend to the left (negative x) and to the right (positive x) respectively. The magnetic interaction can be engineered by changing geometric parameters such as d and D x. (b) Color-map of the reconstructed potential energy as function of the distance between the magnets, d. The distance is expressed in arbitrary units proportional to the voltage applied to the piezoelectric stage. Decreasing the distance between the magnets the potential energy softens and eventually two stable states appear. (c) Dependence of the effective temperature, Teff, with the root mean square of the white Gaussian voltage applied to the piezoelectric shaker. The red dot represent the condition accounted for the experimental data presented. 6/8 Figure 2. Reset protocol. (a) Protocol used to perform the reset operation. In order to account for all the possible transitions we considered the reset to 1 (first two columns) and reset to 0 (last two columnss) starting from both 0 and 1 states. The first row depict the protocol used for removing the barrier. Once the barrier is removed a lateral force is applied (second row). The resulting displacement of the cantilever tip is represented in the third row. Once all the forces are removed and the barrier is restored the cantilever tip encodes the final state. (b) Schematic time evolution of the potential energy and state of the system for the case presented in the first column of panel (a). The operation starts from a double well potential and in the first step of the protocol the barrier is removed. During the next step the potential is tilted and the barrier is restored to its initial value. Finally, the lateral force is removed recovering the initial condition where the barrier between wells is at its maximum and the electrostatic forces are equal to zero. 7/8 Figure 3. Produced heat and probability of success for the reset operation. (a) Average heat produced during the reset operation as function of the lateral alignment D x. For D x < 0 the counter magnet is moved to the right and the 0 state (x < 0) is favorable. Accordingly, for D x > 0 the 1 state is more favorable. Introducing an asymmetry on the potential Q decreases, which is accounted to the probability of success, Ps, that tends to decrease (Ps is encoded in the color map). (b) Success rate of the reset operation as function of lateral alignment. Solid violet circles represent the overall success rate while black and red symbols account for the success rate resetting to 0 and 1 state respectively. The maximum overall success rate is present when the system is almost symmetric, D x ≈ 0. (c) Relation between success rate and heat dissipated. Red circles correspond to the resetting to 1 case while black ones correspond to the resetting to 0. (d) Dependence of Q with the protocol time duration, t p. As t p is increased the effects of frictional phenomena becomes negligible and the produced heat should approach the thermodynamic limit. However, for large t p the reset operation fails giving a wrong logic output. In these cases, where the error probability is high, the produced heat is clearly below the Landauer limit. Inset shows the obtained relation between error probability (1-Ps) and produced heat. The data are compatible with the minimum energy required for a given error probability as predicted by Eq. 1, represented by dashed line. 8/8 

1109.0619

1

1109

Magnetic Fields Effects on the Electronic Conduction Properties of Molecular Ring Structures

While mesoscopic conducting loops are sensitive to external magnetic fields, as seen by observations of the Aharonov-Bohm (AB) effect in such structures, the field needed to observe the AB periodicity in small molecular rings is unrealistically large. The present study aims to identify conditions under which magnetic field dependence can be observed in electronic conduction through such molecules. We consider molecular ring structures modeled both within the tight-binding (H\"uckel) model and as continuous rings. In fact, much of the observed qualitative behavior can be rationalized in terms of a much simpler two-state model. Dephasing in these models is affected by two common tools: the B\"uttiker probe method and coherence damping within a density matrix formulation. We show that current through a benzene ring can be controlled by moderate fields provided that (a) conduction must be dominated by degenerate (in the free molecule) molecular electronic resonances, associated with multiple pathways as is often the case with ring molecules; (b) molecular-leads electronic coupling must is weak so as to affect relatively distinct conduction resonances; (c) molecular binding to the leads must be asymmetric (e.g., for benzene, connection in the meta or ortho, but not para, configurations) and, (d) dephasing has to be small. Under these conditions, considerable sensitivity to an imposed magnetic field normal to the molecular ring plane is found in benzene and other aromatic molecules. Interestingly, in symmetric junctions (e.g. para connected benzene) a large sensitivity of the transmission coefficient to magnetic field is not reflected in the current-voltage characteristic. Although sensitivity to magnetic field is suppressed by dephasing, quantitative estimates indicate that magnetic field control can be observed under realistic condition.

cond-mat.mes-hall

cond-mat

Magnetic Fields Effects on the Electronic Conduction Properties of Molecular Ring Structures Dhurba Rai, Oded Hod and Abraham Nitzan School of Chemistry, Tel Aviv University, Tel Aviv 69978, Israel. Abstract While mesoscopic conducting loops are sensitive to external magnetic fields, as is pronouncedly exemplified by observations of the Aharonov-Bohm (AB) effect in such structures, the small radius of molecular rings implies that the field needed to observe the AB periodicity is unrealistically large. In this paper we study the effect of magnetic field on electronic transport in molecular conduction junctions involving ring molecules, aiming to identify conditions where magnetic field dependence can be realistically observed. We consider electronic conduction of molecular ring structures modeled both within the tight- binding (Hückel) model and as continuous rings. We also show that much of the qualitative behavior of conduction in these models can be rationalized in terms of a much simpler junction model based on a two-state molecular bridge. Dephasing in these models is affected by two common tools: the Büttiker probe method and coherence damping within a density matrix formulation. We show that current through benzene ring can be controlled by moderate fields provided that several conditions are satisfied: (a) conduction must be dominated by degenerate (in the free molecule) molecular electronic resonances, associated with multiple pathways as is often the case with ring molecules; (b) molecular-leads electronic coupling must be weak so as to affect relatively distinct conduction resonances; (c) molecular binding to the leads must be asymmetric (e.g., for benzene, connection in the meta or ortho, but not para, configurations) and, (d) dephasing has to be small. When these conditions are satisfied, considerable sensitivity to an imposed magnetic field normal to the molecular ring plane is found in benzene and other aromatic molecules. Interestingly, in symmetric junctions (e.g. para connected benzene) the transmission coefficient can show sensitivity to magnetic field that is not reflected in the current-voltage characteristic. The analog of this behavior is also found in the continuous ring and the two level models. 1 Although sensitivity to magnetic field is suppressed by dephasing, quantitative estimates indicate that magnetic field control can be observed in suitable molecular conduction junctions. 2 1. Introduction that comprise junctions through molecular transmission Controlling electron molecular ring structures by magnetic fields is considered challenging because the required field strengths are believed to be unrealistically high, of the order of the Aharonov-Bohm 410 Tesla for typical molecular rings.1, 2. In contrast, it was demonstrated period of ~ theoretically3-6 that the conductance of a nano-size ring can be significantly modulated by relatively moderate magnetic fields (< 50 Tesla). This large sensitivity to an external magnetic field results from the presence of sharp resonances that are possible only for low coupling between the molecular-bridge and the metal-contacts. Recent studies of electronic conduction through molecular ring structures such as benzene, biphenyl, azulene, naphthalene and anthracene as well as carbon nanotubes, by us7 and others8-11 have shown that although the net current through these molecules at low metal- molecule coupling is relatively small, induced circular currents can be considerable. The magnitude of such a voltage driven ring current depends significantly on the metal-molecule coupling strength, while its very existence depends on junction geometry, specifically on the location and symmetry of the molecule-metal contact along the circumference of the ring. The magnetic field associated with such a circular current at the center of a molecular ring can be quite significant, for instance, we have found ~ 0.23 Tesla at 2 Volt bias in a tight- binding model of a meta-connected benzene bridge.7 Possible exploitations of such high local magnetic fields at the molecular level could come through the realization of carbon nanotubes as molecular solenoids,11-13 or by controlling the alignment of spin orientation of magnetic ions embedded in the bridge as suggested in Ref. 14. Voltage driven circular currents in molecular ring structures are similar in nature to the persistent currents induced in isolated mesoscopic rings that are threaded by magnetic fluxes. Both phenomena originate from splitting of degenerate ring electronic states characterized by opposite orbital angular momenta, either by the magnetic fields or by the voltage bias even in the absence of external magnetic field.15-19 20 Theoretical studies have shown that magnetic flux induced persistent currents can be controlled by external radiation.21, 22 Also, theory indicates that ring currents can be induced by polarized light,23-25 twisted light,26 and other optical coherent control methodologies,27-29 Such optically induced circular currents can be effectively controlled by externally applied magnetic fields,30 or, conversely, can be used to control the local magnetic field at the ring,31-33 thereby opening the possibility to control the orientation of an impurity spin at the ring center. 3 In a recent preliminary study,34 we have reconsidered the possibility of controlling electrical conduction characteristics of molecular ring structures by static uniform moderate magnetic fields, following the lead of Refs. 3-6, which indicate that large sensitivity to magnetic fields may be found in junctions where (a) electronic state degeneracy leads to interference that can be suitably tuned by the magnetic field, and (b) weak molecule-metal coupling results in sharp transmission resonances. We have shown that for certain geometries and under specified conditions magnetic field effects on molecular ring conduction can become observable. In the present study we analyze in detail the role of molecular geometry and dephasing on the transport properties of molecular rings such as benzene, biphenyl and anthracene, subject to external magnetic fields by means of tight-binding (Hückel) molecular ring models as well as a scattering matrix approach to transmission through continuum loops. The close similarity between the results obtained from these different models indicates their generic nature, and provides evidence to the integrity of results obtained for the effect of a magnetic field in the limited-basis tight binding model (the London approximation). In addition we show that the essential physics underlying the observed magnetic field dependence of conduction can be obtained already from a suitably constructed two-state model. The important role of interference processes implies that the system behavior will strongly depend on the junction geometry that determines the transmission pathways and on the effect of dephasing (decoherence) processes resulting from thermal motions in the junction. In this regard we note that an earlier study35 indicates that increasing electron- vibration coupling in the junction may lead to sharper resonance structure in the current dependence on the magnetic field, and therefore to larger sensitivity of conduction to such field. This however is not simply related to pure dephasing as the effect discusses by Ref. 35 increases at lower temperatures. The structure of the paper is as follows: in section 2, we present three different models for electron transmission under the influence of an external magnetic field: (i) A tight-binding (Hückel) model, which is analyzed using the steady state approach described in our earlier work7, 36, 37 and (ii) continuum ring model analyzed using scattering theory6. We also describe (iii) a simple 2-level model that captures the essence of the observed behavior. Section 3 presents results from our model calculations that describe (a) the magnetic field effect on the transmission probability and the current voltage characteristics of simple molecular ring junctions of various symmetries; (b) comparison between the tight-binding, continuum and two-level models and (c) the effect of structure and geometry on the dependence of junction 4 transport properties on the applied magnetic field. The effect of dephasing processes on these behaviors is discussed in Section 4. Section 5 summarizes our main results and discusses their implications for further study. 2. Models and Methods In this section we describe the models used in this work to analyze magnetic field effects on electron transport through molecular rings. The tight-binding (Hückel) model seems to be the most suitable for a simple description of molecular transport, however we will see in the following sections that the main characteristics of the transport behavior are found also in the continuous ring model. In fact, much of the physics is already contained in an ever simpler model based on a two-level bridge. These models are described below. (a) The tight-Binding Model: Scattering theory on a 1-d lattice We consider a molecular junction formed by a ring molecule bridging the metal leads (L, R) through two chosen sites on the ring. The molecule is described by a tight-binding (Hückel) model with on-site energies Mα and nearest-neighbor interactions Mβ . The metal contacts (L, R) are modeled as infinite 1-dimensional tight-binding periodic arrays of atoms with lattice constant a and on-site energies and nearest-neighbor coupling matrix elements Kα and R LK β K , respectively. A sketch of such a molecular junction is shown in Fig. ) , (  1 Fig. 1. A tight-binding model for current conduction through a molecular ring structure connected to  two 1-dimensional metal leads L and R, at bias voltage V. A static uniform magnetic field B is applied perpendicular to the molecular plane. In the site representation, this TB Hamiltonian is given as HH H H V V ,  LM RM    L  R M (1) 5 where K 1  2 m 1  2 m H K 2 p   β K p   α (2) n  1 e A nn nn 1   M R LK n ; , ,  , where m and e are the electron mass and charge. We    K n K n     V Mm R LKn nm mn β ; ; , , (3)     KM LM where  n is an orthogonal set of atomic orbitals centered at atomic sites n ). The indices K = L, R and M correspond to the subspaces of the left (L), M R LK n ( , , right (R) and molecule (M), respectively.  The model (1)-(3) is supplemented by an additional magnetic field B applied to the junction, so that the kinetic energy operator of an electron is modified according to  2     assume that the field is uniform, applied in the bridge region only and, for the planar ring- molecules considered, perpendicular to the molecular plane. In the standard London approximation38 one (a) represents this field by a vector potential in the symmetric gauge,     , and (b) modifies the finite basis of field-free atomic orbitals  ,   1 ( )A r B r n r   n 2 where r is measured from the center of atom n, so as to account for the phase difference between wavefunctions centered on different atomic sites,       A r i e ) n ( /        r r r   . This leads to a tight binding (Hückel) Hamiltonian with coupling between    A r A n n atomic sites given by            i e A A r /    *       dr r e r V n m (5)    n mn m For nearest neighbor interactions, r in the exponent is approximately replaced by     / 2 r r , leading to n m where (4)   n   e  n  n   M e      M mn i  mn n m M ; ,  , where  mn     A A  m n   e   B   r m   r n  r m      r n   e  4   r r  m n 2  r   n   r m   B e  2 and the molecular Hamiltonian is now given by  (6) (7) 6 )    2 2 ) K .  K H M  mn  i  mn e )    r m ( E )   e  ( E (8) ( E)  ( E )   K  i  mn n n 4  K m n n m (i 2)  Note that position vectors 2   B 0 (       M M n m M n M ,    is twice the area of the triangle spanned by the vectors, so r n B is the magnetic flux through the triangle spanned by the , where   mn r r /h e , is the flux quantum. , and 0   To evaluate the transmission coefficient associated with this setup one could use the non-equilibrium Green function method, but here we follow the scattering method used in Refs. 36 and 37 that gives an easier access to bond currents and, in its density matrix version, can be generalized to account (approximately) for dephasing processes. In this approach, a wire or a network of wires described by a tight-binding Hamiltonian is made to carry current by injecting electrons at one of its sites (source site). At the same time absorption is affected at the “end sites” of other wires by adding the self-energy term E ( ) i   K 2 2 to the site energy. This “absorption” represents the infinite extent of the wire and makes it possible to describe dynamics in an infinite system by a finite system calculation. For n  electrons injected at energy E, The steady state wavefunctions are written in the site   representation in the form      C E t n , n n The orbital coefficients or amplitudes   t n nC E obtained by solving the Schrödinger equation ( ) for steady state situation provide the particle current between any two nearest-neighbor sites (n, m) on wire K as 2  K  which, in the presence of magnetic field (applied in the molecular region), becomes 2  K  The transmission probability through any exit segment or bond is calculated as the ratio of inter-site current (bond current) to the incoming particle current, i.e., K J E ( ) nm E J ( ) In C C Im( n )  n m ( , ) )  n m ( , ) K M    C C n m C n n (12) (11) (10) (13) Im( e (9)   i E /  K ( E ) J K nm J K nm  K nm ( E )  ( E )  E t ,  e . ,     m nmi  . K 7 . j  n j I n j (14) N segments of nj bonds ( ) on which the current has been determined to be Ij, the Under certain conditions, a bond current in a biased molecular ring structure may exceed the cI has developed in the ring. In a net transport current – a signature that a circular current previous paper,7 we have defined the circular current as the sole source of (bias induced) magnetic flux through the ring. For a ring comprising n identical bonds, divided by nodes into N  j n 1  circular current in a ring is given by7 1   n j It was also useful to define the circular transmission coefficient7 as the ratio of circular current to the incoming current as EJ ( ) c J E ( ) In note that this number can be larger than 1. For a finite bias voltage, the net bond current between any two nearest neighbor sites in a two-terminal junction is obtained from the Landauer formula     )(E f K RLK μ K where and are the Fermi functions and chemical potential of the left ) , (  and right leads, respectively. In the calculations reported below, unless otherwise stated, we have taken the leads temperature to be zero, assumed that the potential bias falls on the metal- eV FE and molecule interfaces and considered symmetric potential drop, that is / 2    L FE e     R )) dE ( E ) ( f E ( R f E ( L ) (16) (15) V ( )  / 2 . ( E )   eV I mn  c , , I c nm  (b) Scattering in the continuous ring model Next, we consider the continuous ring model, in which the junction comprises a one- dimensional (1D) conducting ring connecting between two 1D leads (Fig. 2a). The symmetry of the scattering process is imposed through the angle γ between the leads, 39 and the properties of the ring-lead contact are embedded in the imposed junction scattering amplitudes (Fig. 2b) as detailed below. 8 (b) (a) y  U2 U1  L1  D1 L2 D2 R1 R2 x Fig. 2: Schematic representation of the scattering model. Panel (a): assignment of the wave amplitudes on the different parts of the system. Panel (b): junction scattering amplitudes. We assume that the electrons can be represented by plane waves traveling along the ikl ikl A e  A e where l is the electron path length (on the ring l R , ring with the form  2 1 where  is the angle traversed by the electron, and R – the ring radius) and 1,2A are the amplitudes of the respective waves (In Fig. 2a these amplitudes are denoted L, U, D and R in different segments of the ring and the leads). We use the standard notation where negative(positive) k values represent (counter-)clockwise propagating waves. The ring-lead coupling is modeled by assigning scattering amplitudes at the ring-leads junctions40-46 as 2c is the probability of an electron approaching the junction from the lead shown in Fig. 2b. to be back scattered into the lead,  is the probability of an electron approaching the junction 2a is the probability of an electron approaching the junction from the lead to mount the ring, from one of the arms of the ring to be back scattered into the same arm, and 2b is the probability of an electron approaching the junction from one of the arms of the ring to be transmitted into the other arm. Based on current conservation considerations the scattering matrix can be shown to be unitary such that all scattering amplitudes (taken to be real42, 47, 48) can be folded into a single parameter which we choose to be such that: 1 1 2 2 1 2   (17)   c   1     1  c  ; b ; a c 9 It is now possible (Appendix A) to write scattering equations for both junctions taking into account the spatial and magnetic phases accumulated by the electrons while traveling along the arms of the ring. Focusing on the scattering process associated with an incident wave coming on the right lead with amplitude R1, that is, taking the incoming amplitude on the left lead to vanish, L1=0, these equations can be used to relate all amplitudes in the ring segment and the leads to the incoming amplitude R1. This leads (see Appendix A) to the transmission probability in the form 2 (18a)   k  , B   L 2 R 1   1  2  1  2 4     kR 1 cos 2       cos 2   kR         2 sin kR     2      sin   kR    B cos 2     0         (18b)  2   2 c  2  1     kR 4 cos 2      2  a cos 2   kR        2  b    C    2  kR cos 2   2  a cos 2   kR        2  b   B cos 2     0   B cos 2     0                 (18c) where  S S   B S B S     B z  , S is a vector perpendicular to the ring's surface such that 2  / e R . We note that when   the above expression  and, as above, 0 2   reduces to the standard expression for symmetrically connected rings.49 (c) A two State Model In the weak leads-ring coupling limit, the width of the doubly-degenerate energy levels on the ring is considerably smaller than the inter-level spacing between states of different angular momentum. Therefore, at low bias voltages, we can safely assume that electronic transport takes place mainly through a couple of degenerate levels close to the Fermi energy of the leads and model the transport physics using the simplified two-level model shown in Fig. 3. To assign the relevant model parameters, consider an external   zB threading a molecular ring of radius R that lies in the XY plane. magnetic field B ,0,0 The energy levels of an electron moving otherwise freely on the ring are given by (see Appendix B): 10 2 2 2  1 R mE    B m      0 ,2,1,0 is the angular where we use atomic units unless otherwise stated. Here, m quantum number (for the isolated ring, the quantum number m relates to the wave vector k of the previous section via ) and the flux quantum is . In the basis of the 2  kR m  0 corresponding eigenstates of the isolated ring, the molecular ring Hamiltonian MH is given m ) ( MH , that correspond to states whose by a repeated sequence of diagonal 2x2 blocks, (19) degeneracy is split by the field: ) 2 2 2 2 0    m H       m ( M         1 R  B  0 m B z 2                         The full transport problem of a molecular ring connecting between two metal leads can thus be replaced by the simplified model involving only the two levels characterized by a given ,j KV K L R j m , shown in Fig. 3. Here , 1, 2 ; denotes the coupling between molecular ;  m B z 2 . (20)      B  0 0 E 2 E 1 0 1 R          m    2 0 2 2 2 level j and the lead K, and E 1/ 2  1 R 2 2 2 m          B  0    2      m B z 2 V2L E2 V2R (21) V1L V1R E1 Figure 3: Schematic representation of the two-level model. The leads are represented in this figure by tight binding chains. transmission coefficient The corresponding           a r   E G E E G E   M M where the broadening function of lead K (=L,R) is given by       a r   K K the Landauer is given by formula  i    (23a) (22)   Tr E E E E      K R L 11 the retarded Greens function of the molecule is calculated as 1 † E          a r    G E G E    M M And the self-energy of lead K is represented as EI H      E r R   r L M   (23b) † 2 K K      kk         r K a K E E (23c)      r 0 G E K V K 1, * V V K 2, 1, * V V K K 1, 2, 2 V 2,             0r KG E being the retarded Green's function of the isolated lead K. Eq. (23c) is written under the assumption of short range interaction between ring and lead, whereby the ring is coupled to the nearest neighbor lead site denoted by the index k. We could have used here the explicit Newns-Anderson model for a 1-dimensional tight binding lead for which  is given by Eqs. (9), however, because we will be using the 2-level model as a generic simple model to gain physical insight into the nature of our results it is enough to make the simplest wide-band     r r 0 0 G E G E R L         kk kk electronic states and we assume identical leads. On the other hand, the magnetic field dependent transport properties are determined by the choice of lead-ring coupling elements V  that enter Eq. (23c). Aiming to capture this dependence, we assume that lead ring where ρ is the density of lead i   approximation for which        im  Ve reflects the phase of the wave function on the ring at the positions of the leads-ring junction. Referring to Fig. 2a and setting the angle at which the right lead is attached to the ring to be 0  , the left lead is attached at the respective angle γ (γ is π, 2π/3 and π/3 for the para, 0 meta and ortho configurations, respectively). Hence we take im im im      Ve Ve V V Ve V V V V ; ; ; 0 0 R R L L 1, 2, 1, 2, where V is the coupling strength. Consequently, Eq. (23c) yields the retarded self-energies associated with the right and left leads in the forms 2 1 1   1 1          where m is related to the imposed magnetic field and to the electron energy through Eq.(21).   /L R E It is now possible to obtain explicit expressions for the broadening matrices and   EGEG   the retarded and advanced molecular Green's functions from which the r a , M M transmission probability can be calculated using Eq. (22). A long but straightforward calculation (see Appendix B for a detailed derivation) leads to 1 im 2   i V    i V    r R E (25) (24) ;    im  E   r L 1 e e    2 2 12     zE B  ,  1  2  1 2   E E  1 2      2  E E  1   E E  2   cos 2  m    E E  2 2    (26a) (26b)  2     2   E E  1   2 E E  1  E E  2 2   2  2  E E  1   E E  2  2 cos   m    E E  2 2       sin   m  4   (26c)   2 2 V  where the dependence on 3. Results and discussion zB originates from Eq.(21). (26d) As discussed above, current conduction through ring structures is inherently associated with interfering transmission pathways50 that may be conveniently described in terms of degenerate eigenstates of the isolated ring. The corresponding degenerate states can be represented in terms of rotating, clockwise and counter-clockwise, Bloch states on the ring. Indeed, it is the tuning of relative phases of these states by magnetic field that potentially provides magnetic field control of the ring transmission properties. This implies several important aspects of the resulting behavior: First, transport will be affected by interference (and consequently most amenable to magnetic field control) in energy regimes dominated by such degenerate states. Second, strong interference effects and large sensitivity to magnetic field are expected when these states are associated with sharp transmission resonances, i.e., for sufficiently weak metal-molecule coupling. Third, the symmetry of a given junction geometry strongly affects the interference pattern and hence dictates the transport properties. Fourth, these phenomena will be strongly affected by dephasing processes. In what follows we will see different manifestations of these statements. We study single-molecule junctions consisting of molecular ring structures such as benzene, biphenyl and anthracene connecting metal leads. The results presented below focus on the response of these systems to an externally applied static uniform magnetic field in terms of modification in their electronic transport characteristics. The molecular junction is emulated by the tight-binding (Hückel) molecular Hamiltonian and 1-dimensional tight binding leads as presented above. For the latter we take zero on-site energies, i.e., K K L R eV6 ( , ) 0  and nearest neighbor coupling that corresponds to a      L R metallic band of width 24 eV. The zero bias Fermi energies of these contacts are set to 13  M  eV5.2 eV5.1 and for all α M  0FE . For the molecular structure we take nearest-neighbor atom pairs. 51 Consider first a simple benzene ring that can couple to the metal leads in para, meta and ortho configurations (Fig. 1). In the free molecule the highest occupied molecular orbitals (HOMOs) and the lowest unoccupied molecular orbitals (LUMOs) constitute pairs of doubly degenerate orbitals which, with our choice of molecular parameters and energy origin, are eV eV 4 positioned at , respectively. Upon connecting to the and 1        M M M M metal leads these levels get broadened and, more importantly, their degeneracy split. For sufficiently weak metal-molecule coupling these split levels constitute sharp transmission resonances at the corresponding energies. It should be emphasized that degeneracy splitting in benzene affected by imposing perturbations at some atomic positions does not by itself specify the nature of the new eigenstates. An important property of the resonances obtained when the ring is connected to infinite leads at the meta or ortho positions (i.e. scattering resonances characterized by scattering boundary conditions, that is, incoming in one lead and outgoing in the other(s)), is that they can be shown to maintain the character of circulating Bloch eigenstates of the isolated ring. The corresponding transmission resonances are therefore associated with considerable circular current in the benzene ring.7 Consequently, in the meta and ortho- connected configurations, large circular currents are found when bias and gate potentials are such that one of the split resonances dominates. In contrast, in the para connected ring, one of the split eigenstates turns out to have a node at one of the para positions and, consequently, does not contribute to transmission, while the other is characterized by zero net circular current as could be expected from symmetry. This is shown in the upper panel of Fig. 4, where the bias voltage is set to V = 2 V. This brings the upper Fermi energy to the vicinity of the LUMO pair: In the meta and para connected benzenes one of the split resonances is below and the other above this energy.52 The metal-molecule coupling is taken βLM = βRM = 0.05 eV, low enough so these resonances remain well separated. For these parameters the net current through the junction is of order ~ nA, while the circular current in both the meta and ortho configurations is three orders of magnitudes larger, yielding  0.23 Tesla for the induced magnetic field in the ring center in both cases. Note that the direction of the circular current and the ensuing magnetic field is opposite in the meta and ortho configurations. See Ref. 7 for more details. 14  Fig. 4. Internal current distribution in (a) para (b) meta and (c) ortho-connected benzene ring, eV under voltage bias of V  2 V. The upper panel shows the 0.05 KM  connected to leads with bond currents calculated in the absence of an external magnetic field, while the lower panel corresponds to the presence of a magnetic field, B = -2T (negative B corresponds to a field pointing down into the plane). The arrows along bonds represent bond currents with magnitudes proportional to the corresponding arrows lengths. The encircled dot (cross) in the meta (ortho) structures in the upper panel denote the directions of the magnetic field induced by the circular current: out of (into) the molecular plane. When an external magnetic field is switched on, the bond-current map changes. In these and the following calculations the external magnetic field is taken in the Z direction, perpendicular to the molecular XY (also page) plane. Positive field direction is taken to be outwards, towards the reader and a positive circular current is taken to be in the counterclockwise direction. This field generates an additional circular, so called persistent, current that can reinforce or suppress the voltage driven circular current. Thus, a negative magnetic field (direction into the paper plane) generates a current in the anticlockwise V  2 V direction that adds to the circular current in the meta connected ring (at ) and subtract from it in the ortho-connected structure, as seen in the lower panels of Fig. 4. Of course, the interplay between the voltage driven and field induced circular current depends on the voltage range considered. For example, in the meta-connected ring the voltage driven circular V  2 V current reverses its direction above and a negative field induced persistent current will add to it destructively as in the ortho case. This implies that at any finite bias I B I B ( ) ( )   c . c 15  Fig. 5. Circular current as a function of bias voltage in the range (1.994 to 2.006 Volt) in a in the presence of external magnetic field, B = 0, +/- 1T meta-connected benzene for eV05.0 KM  and +/- 5T. The inset depicts the case for for applied field B = 0, +/- 100 T. eV5.0 KM  Fig. 5 shows another aspect of this effect. Here the circular current in a meta- eV (K = L, R) is shown as a 0.05, 0.5 connected benzene ring connected to leads with KM  function of voltage for different applied magnetic fields. Again it is seen that I B ( c c in the presence of bias voltage. It is also seen (inset) that the sensitivity to magnetic field is strongly reduced when the molecule-lead coupling becomes stronger. B  I (  ) ) 16 2V  V plotted as function of Fig. 6. The circular current in a meta-connected benzene ring biased at magnetic field applied perpendicular to the ring. The three cases shown correspond to different 0.05 0.10 eV, dashed line (red): eV and molecule-lead coupling. Full line (black): KM  KM  eV (in the inset). The magnetic field induced by the voltage driven indB KM  current is practically the same in all cases, dotted line (blue): 0.5  0.23 T. It is of interest to ask, what is the external magnetic field that will annihilate the circular current in a voltage driven molecule? One may naively expect that this is just the field equal in magnitude and opposite in direction to that induced by the circular current so that the two fields annihilate each other, however Fig. 6 shows that the magnetic field needed to stop the circular current is considerably larger than the magnetic field produced by that 0.05, 0.1, 0.5 current, and generally depends on the molecule-lead coupling. For eV we KM  find this field to be 0.98, 3.92 and 96.40 tesla, respectively, at the voltage bias employed (2 V). On the other hand we find that the magnetic field induced by the circular current, indB tesla at the same voltage bias, almost independent of the molecule-lead 0.23  coupling.53 This non-trivial behavior results from the fact that the application of the external magnetic field does not simply oppose the circular-current induced magnetic field but also alters the electronic structure of the ring and strongly influences the interference pattern of coherent electrons mounting the ring thus influencing the resulting induced magnetic field itself. 17  Fig. 7 The I-V Characteristics of a junction comprising para- (upper panel) and meta- (lower panel) connected benzene coupled to the leads with coupling element 0.05 eV, evaluated for different magnetic field strengths B normal to the ring. The results obtained for different magnetic fields for the para system are essentially indistinguishable from each other. The inset in the upper panel shows a close-up on the V = 2 V neighborhood that shows the consequence of the split degeneracy in the para- connected junction. The inset in the lower panel shows the I-V behavior in the meta configuration for molecule-metal coupling 0.5 eV, which is essentially field independent in the same range of magnetic field strengths. Results for the ortho-connected molecule are qualitatively similar to those shown for the meta configuration. Of more practical implications is the question whether the junction transport properties can be affected by an externally applied magnetic field. As already mentioned,  E previous studies1, 2 seem to indicate that while the transmission may be affected by an external magnetic field, the integrated transmission that yields current-voltage characteristics, is not sensitive to this field. The main reason for this observation is that at realistic magnetic 18 the field values the splitting between the degenerate levels of the ring is of the order of meVs (see Fig. 9). At high leads-ring coupling this splitting is much smaller than the width of the  E curves. At the low coupling limit, two levels and is therefore hardly seen even in the  E curve clearly shows the magnetic field induced level splitting but in order to observe the magnetic field effect in the I(V) curves, the Fermi integration window (see Eq. (16)) should include only one of the split levels. This, however, requires bias and gate voltage precision smaller than the level splitting, as well as very low temperatures. One may conclude that despite the ability to control the magnetic field sensitivity of the transmission probability through molecular rings via the leads-ring coupling,5, 6 practical measurements of the I(V) curves will hardly show any magnetic-field effect. This can be clearly seen in the upper panel of Fig. 7 where the current-voltage relationship of a symmetrically (para-) connected benzene ring is found to be robust against the external field. Only when zooming into the current step region (inset of the upper panel of Fig. 7) one finds a small shoulder resulting from the level splitting at a finite magnetic field value. While this conclusion is true for the symmetric junction, in the asymmetrically connected junction a different behavior is observed. In the lower panel of Fig. 7 we present the I(V) curves of the meta-connected ring for different magnetic field intensities. Despite the fact that the level splitting is similar to that of the symmetric junction, the current-voltage characteristics show pronounced sensitivity towards the magnetic field. 19  around E=1eV through a junction comprising para- (upper Fig. 8 Transmission probability  E panel) and meta- (lower panel) connected benzene coupled to the leads with coupling matrix element 0.05 eV, evaluated for different magnetic field strengths B normal to the plane of the ring. Again, results for the ortho-connected molecule are qualitatively similar to those shown for the meta configuration. To get further insight into the origin of this behavior, we show in Fig. 8 the ( E ) transmission functions that constitute the input to the I-V results of the preceding figure. Shown are the transmission functions for para- (upper panel) and meta- (lower panel) connected benzene rings under different perpendicular magnetic fields in the vicinity of the eV doubly degenerate LUMO energy of the isolated molecule, 1 . We see that the    M M transmission function depends on the magnetic field in both the symmetric and the asymmetric junctions. Consider first the para connected junction and denote the split states in this configurations by 2 . As noted above, only one of these, say 1 and 1 , contributes to the transmission and the system is characterized by a single transmission resonance. In terms of the two counter-rotating Bloch states that are eigenfunctions of the isolated ring, this resonance is a linear superposition in which the corresponding paths add constructively. The 20 the eigenstates become (as B0) , implying that (a) the single transmission 2 of zero transmission corresponds to the superposition in which they interfere other state destructively to give a node in one of the para positions. In the presence of an applied field B 1/ 2    2    2 1 0B  splits into two peaks of equal intensities, and (b) the total area under the peak at transmission function remains unchanged. This leads to an I-V characteristics that does not depend on the magnetic field (Fig. 7a) except in the very narrow voltage region where the Fermi step goes through the split peak (Fig. 7a inset). As discussed above, such a sharp Fermi step requires very low temperatures (~1 K) to be resolved. In contrast, in the meta- and ortho- connected junctions, the asymmetric coupling to the leads results in the appearance of two transmission peaks to appear already in the absence of external magnetic field. As B is increased, this splitting reduces up to a certain magnetic field intensity (for instance, +/-2 tesla in the meta-configuration for 0.05 eV molecule-lead coupling) where it vanishes (level crossing) engendering constructive interference at this field value. This can be understood as phase adjustment of the interfering electron waves by field, causing them to interfere constructively until full resonant transmission is reached. Interestingly, in this regime of magnetic field strength not only the splitting but also the area under the transmission function is field dependent. As the field intensity increases from zero the total area under the peaks increases until the peaks become fully separated and then the area remains constant. This is clearly manifested in the field dependence of the current- voltage characteristic shown in Fig. 7b, suggesting that in the asymmetric case the I-V field dependence should be experimentally accessible in the low leads-ring coupling regime. Note that Fig. 7b shows results obtained at 0K, however the results obtained at 300K are almost indistinguishable. A more general view of this behavior is seen in Fig. 9, which shows, for the para- and junctions, meta- connected levels at the benzene energy the evolution of E eV 1 as a function of molecule-leads coupling (left side of figures) and      M M magnetic field (right) as expressed by the transmission function. It is seen that level degeneracy is lifted by the molecule lead coupling in the meta structure, but not in the para structure. Increasing the magnetic field for a given (0.05 eV) molecule-lead coupling splits the levels in the para case, but brings them together first in the meta case, as discussed above. 21 E plane at B = 0 (left of the vertical K L R ,  Fig. 9. Transmission probability maps in the   KM dashed line) and in the E B plane for 0.05 eV KM  junctions comprising para- (upper panel) and meta-connected (lower panel) benzene molecules. Color ) to red ( ). code varies from deep blue ( 1 0  (right of the vertical dashed line) for Two additional observations should be pointed out. First, another regime of field dependence takes place at very high fields, where shift of energy levels makes more levels to appear within the Fermi window between μL and μR. This happened at unrealistically large fields   ( Ec ~ 1000T . Second, in contrast to the circular transmission coefficient , the ) BI BI ( E ( ) ) ) ( total transmission coefficient is not affected by the field direction,   . Qualitatively similar results as described above are obtained for other ring-containing molecular systems. For example, the isolated biphenyl molecule, which comprises two coupled benzene rings, possesses two 2-fold degenerate orbitals, viz., HOMO-1 and eV and 4 LUMO+1 that for our choice of parameters are positioned at     M M , respectively. A biphenyl molecule connected to leads at positions 6,10 (see   M M Fig. 10) can be considered as two coupled benzene rings in para configuration, and consequently we expect that its I-V characteristic is insensitive to an imposed weak magnetic field. Indeed, a recent work54 finds that to affect transport in this configuration by magnetic 0.5 which, as discussed, is unrealistic for such a small field requires flux of order 0 molecular structure. On the other hand, the diagonally connected biphenyl shown in Fig. 10a eV 1  22 can be considered as two coupled benzene rings, each in meta configuration. For such a structure weakly coupled to leads, we find again high sensitivity of the transmission (Fig. 10b) and the I-V curve (Fig. 10c) to the magnetic field. It should be noted that sensitivity to magnetic field is manifested when the molecular levels at 1eV (degenerate in the free molecule) enter the Fermi window at voltage bias 2V. The current rise at and V52.0V  3.58V is due to resonant transmission through non-degenerate energy levels at ~ 0.26 eV (LUMO) and 1.79 eV (LUMO+2), respectively, and is not affected by the field. Fig. 10. (a) Field free internal current distribution in a diagonally connected biphenyl molecule at a bias voltage of 2V showing the circular currents in the absence of external magnetic field for metal- molecule coupling of 0.005 eV. (b) The transmission probability displayed against the electron energy  around 1 eV at different field strengths. (c) Magnetic field effects on the I-V characteristic for B in the range 0…15 Tesla. 23  Fig. 11. (a) Field free internal current distribution in a diagonally connected anthracene molecule at a bias voltage of 2V showing the circular currents in absence of external magnetic field for metal- molecule coupling of 0.005 eV. (b) The transmission probability displayed against the electron energy around 1 eV at different magnetic field strengths. (c) Magnetic field effects on the I-V characteristic  for Tesla. B  20 Similar results for a junction with diagonally connected anthracene bridge are shown in Fig. 11. In the voltage range shown, sensitivity to a weak magnetic field is associated with eV (LUMO+1) and 1 levels at the doubly degenerate anthracene    M M 2.03 eV  (LUMO+2) (responsible for the current rise at V = 2 and 4.06 eV, 2   M M respectively). The other current steps seen in Fig. 11c are due to non-degenerate levels and are not sensitive to the field. As in the previous cases discussed, this behavior depends on the symmetry of the molecular junction. For example, in agreement with Ref. 55, no sensitivity of the I-V characteristic to weak fields is found for contacts placed in the (2,6) positions (see Fig. 11a) although the transmission function itself does depend on the field in a way reminiscent to the para-connected benzene junction. Finally, we note that naphthalene does not have orbital degeneracy and indeed no I-V sensitivity to weak magnetic field is found (although circular currents are induced) in junction models based on this structure. Conduction through this molecule is found to be affected only by unrealistic strong fields of order ~ 1000 tesla. These observations can be summarized by stating that sensitivity of the I-V behavior to relatively weak external magnetic fields is a generic phenomenon in many ring molecular structures characterized by 24 weak molecule-lead coupling, however its manifestation depends on details of the electronic structure of the molecule and on the junction geometry. ) for (a)   0.0005   4  . 0/ ~ 10 B  in a continuum model ( Fig. 12. Transmission probability around 1kR 2 / 3   (meta) at very small fractions of the flux quantum (para) and (b) It is interesting to compare the results shown above for the field affected electronic transmission through molecular structures to the equivalent transmission problem in the continuum ring model described in Section 2b. Fig. 12 depicts the transmission probability as function of the dimensionless parameter kR calculated from Eq. (18), using the reflection 0/B  of the order of 10-4. The cases   parameter at different flux ratios, 0.0005  2 / 3 and correspond to the para and meta connected rings, respectively. Estimating the   kR  correspond to a free electron energy of benzene ring radius as R~0.13 nm, we find that 1 4  0/ B  of the results displayed in Fig. 12 and those of Fig. 8 is obvious, showing that the complex corresponds to B of order 10 tesla. The close similarity the order ~ 2eV, while  10 25 nature of the magnetic field dependent transmission probability through junctions involving rings of various geometries is intimately related to the symmetry of the wavefunctions obtained by a simple model of a particle on a ring. This further justifies the use of the simplified two-level model introduced in Section 2c which relates the generic nature of the magnetic field dependence described above to the phase dependent coupling coefficients and magnetically controlled wave interferences on the ring. In Fig. 13, we plot the transmission probability as function of energy obtained by the two-level model expression (Eq. (26)) for two system geometries under various magnetic field intensities. 0 0 180 120     Fig. 13. Transmission probability as a function of energy obtained using the two-level model, Eq. (26 ), for meta (left panel) and para (right panel) connected rings. We use m=2, ring's radius= 1.0 nm, V=0.1 eV, =0.05 eV-1 and let the x-axis origin follow the average position of the two levels in the presence of the magnetic field. The ortho-connected ring results are identical to those obtained for the meta configuration. As can be seen, the two-level model fully captures all the features appearing in the transmission probability curves obtained by both the tight-binding Hamiltonian and the continuum scattering description. Here, as well, for the symmetrically connected ring the magnetic field serves to split the energy levels while conserving the total area under the transmission peaks, whereas for the para and ortho configurations the integrated transmission probability grows with the magnetic field. This equivalence between the three approaches (tight binding Hamiltonian, scattering model, and two-level model) sheds light on the origin of the different behavior of the transmission probability between the three system geometries. 26 As suggested by the two-level model, the differences between the three geometries enter via the coupling integrals between the ring and the leads taken to be proportional to the phase of the wave function at the locations of the junctions. Coupling of one of the leads to a nodal point of the wave function will cause destructive interferences which may be lifted by the application of the external magnetic field. Since different geometries couple the leads with different phases the response of the transmission probability towards the magnetic field is altered. 4. Effect of dephasing Since much of the effects discussed above result from interference between transmission pathways, it is expected that dephasing processes will have a strong effect on these observations. Such effects were studied by several authors in this context using the Büttiker probe method56 whose application predominates the field of junction transport. Because this phenomenological method is based on a rather artificial process of replacing coherently transmitted electrons by electrons with indeterminate phase, we chose to compare such results with those obtained from another phenomenological model in which the dynamics of the density matrix of the molecular bridge incorporates damping of non-diagonal elements as done in the Bloch or Redfield equations describing relaxation in a multilevel system. In both the Büttiker probe and the density matrix approaches it is possible to affect dephasing locally, i.e. at any given site of the tight-binding bridge that represents our molecule, using the following procedures: In the Büttiker probe method56, 57 the dephasing rate at such a site, j, is determined by its coupling to an external thermal electron reservoir, J, with chemical potential set such that no net current flows through the corresponding contact. We describe the probe by the same tight-binding metal model, Eq. (2), and the same energetic parameters (site energy and intersite coupling) as our source and drain leads. The dephasing rate is determined by the ,j J between molecular site j and the site of the probe J that is coupled to it. In the coupling calculations reported below we take all these coupling parameters between molecular sites , for all j and J. Within the model, one calculates the and probes to be the same,   ,j J MB   K K E ' , probes) and obtains the effective transmission function between source and drain in the form between any two leads ( transmission functions for N  1, ... N L R J , ,  K K , '  27    eff RL   RL N   J J , ' 1   RJ W JJ '  J L ' , where W 1     JJ ' JJ '  1   JJ '    1   JJ   JJ ' (δ is the Kronecker delta function) and R JJ    1  K R L J , ,  '  J   JK is the reflection function in   ,  m .     n ( E )   ( E ) (1  (27)   nm ]  nm n and where )   nm nm probe J. In the density matrix description36, 37 one looks, for a given injection current in the source wire, for the steady state solution of the Liouville equation for the density matrix of the inner system i H 1 1 [  nm  2 2 n is the self-energy, Eq. (9), representing the effect of an infinite lead coupled to site   , where nM is 1 if site n is on the molecule and is zero (1 / 2)       nm mM nM otherwise. Here, the parameter η (taken to be the same for all ring sites) represents the dephasing rate. The resulting steady state solution is then used to evaluate the current on any bond segment as well as the transmission coefficient.36, 37 Figures 14 compares results from these calculations in the absence of a magnetic field. Here dephasing is affected on all sites of the molecular (benzene) ring and the transmission coefficient is plotted against electron energy for the para, meta and ortho connected benzene molecules for different values of the dephasing parameters. We note in passing that our calculations using the Büttiker probe method are practically identical to those obtained by Dey et al.58 when the same junction parameters are used. On this level of presentation the main effect of dephasing is seen to be broadening of the transmission peaks. We note that the two different phenomenological models of dephasing give qualitatively similar results. 28  Fig. 14. Transmission probability as a function of energy in the presence of dephasing: Top, middle and bottom figures correspond to para-, meta- and ortho-connected benzene molecules. Left: Results obtained using the Büttiker probe model with the indicated coupling parameter BM . Right: results obtained by the density-matrix calculation with the indicated dephasing rate η. The molecule-leads coupling is 1 eV. The other model parameters are as given in the second paragraph of Section 3 (parameters of the probe leads are the same as for the source and drain leads). Figures 15 (using the Büttiker probe method) and 16 (density matrix model) focus on 1E  eV. The broadening effect caused by dephasing can be the transmission resonance near clearly seen both in the strong (upper panels) and weak (middle panels) leads-ring coupling regimes. In addition the effect of dephasing on eliminating interference characteristics is apparent. This is most pronouncedly manifested in the integrated transmission (lower panels). For low bias and temperature, broadening alone makes the integral smaller at any finite E 29 because part of the integrand exits the narrow integration window. Furthermore, in the para connected molecule, the integrated transmission goes down also because of the destruction of constructive interference. Conversely, in the meta configuration it goes up (Fig. 15) or considerably more weakly down (Fig. 16) because the destructive interference is eliminated. Fig. 15. Dependence of the transmission resonance at 1 eV on dephasing calculated with the Büttiker probe method. The line style and color representing different values of BM and given in the framed inset in panel 1a correspond to all panels. Panels 1a,b,c show results for para-connected benzene while panels 2a,b,c correspond to the meta-connected molecule. In the a and b panels the molecule- K L R 1 , KM  eV and 0.05 eV, respectively (  source/drain couplings are ). In the weak molecule- 0 BM  (no dephasing) lines are scaled down by a source/drain coupling case (panels 1b, 2b) the factor of 100 (i.e. multiplied by 1/100) in the para case and by a factor of 2 in the meta case, in order E    a to fit on the given scale. The c panels show the integral about the resonance, for the dE  E 30 weak coupling (0.05 eV; as in panel s b) case plotted against E, where the 1eV resonance but well above the lower transmission resonance. a e 1 V is placed well below Fig. 16. Same as Fig. 15, where the effect of dephasing is obtained from the density matrix approach. The line style and color representing different values of  and given in the framed inset in panel 1a 0 lines are scaled down by the multiplicative factor correspond to all panels. In the b panels, the 2 10 4 10 2 4 in the para case, and of in the meta connected molecule in order to fit into the 4 10 4 , i.e. the peak transmission in scale used here. Note that the vertical scale itself goes up to the para-connected molecule without dephasing is 1. In Figures 7 and 8 above we have demonstrated the possibility of magnetic field control of the I-V characteristic in the case of meta (and ortho) connected benzene. Figures 17-18 show the effect of dephasing on this dependence. Figures 17a is the analog of Fig. 7b, 31  ) meta-connected junction with dephasing implemented by the Büttiker probe Fig. 17. (a) Magnetic field dependence of the current near the 2V step (associated with the E  1eV ; analog of Fig. 7b) calculated for a weakly coupled transmission resonance at 0.05 eV KM  0.1eV BM  method (       I I B 0 0   plotted respectively against the dephasing parameters BM ( Büttiker probe   method) and  (density matrix method). The inset in 17c displays the results of the main figure on a different scale, emphasizing the observation (also seen in 17a) of deviations from B symmetry at intermediate dephasing rates. ). (b) and (c) The same magnetic field dependence expressed by the ratio (  I 32  where the I-V characteristic for the meta-connected benzene under different magnetic fields normal to the ring plane is shown, now in the presence dephasing (Büttiker probe method, 0.1eV ). Similar results are obtained when dephasing is introduced in the density BM  matrix method. We see that the field dependence of the current step strongly diminishes in the presence of dephasing. This is shown more explicitly in Figures 17b,c , where the difference between the current evaluated at V = 2.2 V for 0B  and B   is plotted 15 tesla against the dephasing parameter. More insight about the origin of this behavior is obtained from Fig. 18, which is the analog of Fig. 8 calculated in the presence of dephasing. The primary effect of dephasing in the range considered is seen to be the destruction of interference effects, which strongly diminishes the difference between the transmission properties of the para and meta connected junctions as well as the dependence on an imposed magnetic field. Fig. 18. Transmission probability for para (1a, b) and meta (2a, b) weakly connected 0.05 eV KM  ( ) benzene as a function of electron energy in the presence of dephasing affected through Büttiker probes (panels a) ( eV 001.0 BM  ) at B = 0, -5 and -15 Tesla.  ( eV1.0 ) and density matrix method (panels b) 33 The significance of the dynamic destruction of phase in the quick erasure of the magnetic field effect on the conduction properties of asymmetrically connected benzene molecules can be gauged against other effects that can potentially affect this sensitivity. First, our calculations have disregarded the implications of the voltage distribution across the molecular junction. On the simplest level of description this can appear in the voltage division between the two contact, expressed by a factor ξ and a voltage V such that   and     V V V V are the potential drops on the two molecule-lead contacts 1       while V is the potential drop on the molecule itself. All the calculations described above 0V  . We have established that our results are not sensitive and where done with 0.5 to the choice of . An example of the effect of V is shown in Fig. 19, which extends the calculations displayed in Fig. 8b (meta-connected benzene) to the case shown in panel (a) where some of the site energies are changed,       . We see that the magnetic field effect on the transmission decreases with increasing  and practically disappears for eV. This emphasizes the need to carry such experiments under low bias 0.05  conditions, implying that need to align the molecular spectrum with respect to the lead Fermi energy with a gate potential. 34 Fig. 19. The transmission function for the meta-connected Benzene, same as Fig. 8b,  E with some site energies shifted as shown in panel (a). In panels (b), (c), (d)  is taken 0.01, 0.02 and 0.05 eV, respectively. Secondly, keeping in mind that the required slow dephasing implies low temperature, it is important to note that the effect seen in Figs. 17, 18 arises from the dynamic destruction of phase, and is not reproduced merely by raising the electronic temperature of the leads. This is seen in Fig. 20, which shows the effect of leads electronic temperature on the observed magnetic field dependence of the conduction through a meta-connected benzene junction near the molecular resonance at 1 eV (bias voltage 2 V) in the absence of dephasing. Fig. 20a is similar to Fig. 7b except that the Fermi distributions in the electrodes were taken at 300K. Fig. 20b displays the current at V = 2.2 V through this junction, plotted against the electrodes temperature for B = 0 and 15 tesla. Only weak electrode temperature effect is seen at the realistic temperature range considered. Fig. 20. (a) Current vs. bias voltage for a weakly ( 0.05 eV KM  different imposed magnetic fields perpendicular to the molecular plane (same as Fig. 7b) calculated at T = 300K. (b) The current through the same junction at V = 2.2 V, displayed as a function of temperature for B = 0 and 15 T. ) meta-connected benzene for Finally, It is interesting to note that in the presence of dephasing, deviations from Onsager symmetry under reversal of field direction, B B  , are observed (as seen in Fig. 17). Such deviations were discussed in previous work in different contexts, including coupling to a thermal environment.59-62 We defer further discussion of this issue to a later publication. 35  5. Concluding Remarks In this paper we have addressed the issue of magnetic field effect on electronic transport in molecular conduction junctions. Observations of the Aharonov-Bohm effect in mesoscopic conducting loops could suggest that molecular junctions comprising molecular ring structures as bridges will show similar effects, however the small radius of molecular rings implies that the field needed to observe the AB periodicity is unrealistically large. Still, we have found that strong magnetic field effects can be seen under the following conditions: (a) The molecular resonance associated with its conduction behavior is at least doubly degenerate, as is often the case in molecular ring structures; (b) the molecule - lead coupling is weak, implying relatively distinct conduction resonances, (c) asymmetric junction structure and (d) small dephasing (implying low temperature) so as maintain coherence between multiple conduction pathways. Interestingly, in weakly connected symmetric junctions (e.g. para connected benzene) the transmission coefficient can show sensitivity to magnetic field, however it is found that the integrated transmission is field independent, so that this sensitivity is not reflected in the current-voltage characteristic. For the organic structures we have used the within the tight-binding (Hückel) model modified for the presence of magnetic field using the London approximation. Qualitatively similar results were obtained from the analog model of a continuous ring, showing that the qualitative effect studied depends mostly on the strength and symmetry of the molecule-lead coupling. We have also shown that much of the qualitative behavior of conduction in these models can be rationalized in terms of a much simpler junction model based on a two-state molecular bridge. When the conditions outlined above are satisfied, strong dependence of conduction on the imposed magnetic field can be found. The effect of dephasing processes on this observation are studied using two different phenomenological models: the Büttiker probe and phenomenological coherence damping imposed on the Liouville equation for the molecular density matrix. Both treatments are approximate: The approximate nature of the density matrix approach stems from the fact that dephasing was affected by damping non-diagonal elements of the density matrix in the local site representation while assuming that the transmission energy remained well defined.36, 37 The Büttiker probe approach is limited to linear response and cannot be rigorously applied to threshold phenomena in the current- 36 voltage dependence. Still, the fact that these two different approaches gave qualitatively similar results in all cases studied, provide some assurance about their validity. Both models show strong suppression of the sensitivity of conduction to the imposed magnetic field. Next, consider the implications of the conditions outlined above to experimental considerations. Conditions (a) and (c) can be met by making a proper choice of molecular bridge and the positions of linker groups. Condition (b) of weak molecule-lead coupling does not imply weak molecule-lead bonding, only that the resonance states that involve multiple pathways through the ring (or counter propagating wavefunctions in the ring) are weakly coupled to the metal electrodes. This can be achieved by connecting molecular ring to leads via saturated alkane chains. Condition (d), the requirement for small dephasing, is inherent in all experiments trying to observe interference phenomena in molecular junctions, and implies the need to work at relatively low temperatures. The Büttiker-probe procedure is not related directly to a physical process, so it is hard to assess the experimental implication of the coupling MNV . The equivalent analysis in terms of the coherence damping rate η does provides an estimate: For the reasonable molecular parameters chosen in our calculations, Fig. 17c shows that magnetic field effects are suppressed when this damping rate exceeds ~ 0.001 eV, that is, dephasing times of the order of 1 ps. Recent observations in different systems63 have shown that molecular electronic coherence can persist on such timescales even at room temperatures. This suggests that the magnetic field effects discussed in this paper may be observables. Finally, we have observed magnetic asymmetry (under reversal of field direction) in the presence of dephasing. This observation and its repercussions will be discussed elsewhere. Acknowledgment We thank Joe Imry for helpful discussions. The research of A.N. is supported by the Israel Science Foundation, the Israel-US Binational Science Foundation, the European Science Council (FP7 /ERC grant no. 226628) and the Israel – Niedersachsen Research Fund. O.H. acknowledges the support of the Israel Science Foundation under Grant No. 1313/08, the support of the Center for Nanoscience and Nanotechnology at Tel-Aviv University, and the Lise Meitner-Minerva Center for Computational Quantum Chemistry. D. R. Acknowledges a Fellowship received from the Tel Aviv University nanotechnology Center. 37  Appendix A: Scattering model for asymmetric nanoscale junctions containing molecular rings and magnetic fields. We consider the setup presented in Fig. 2a, and write the electron wavefunction in each ikl ikl A e where l denotes a propagation distance, i.e. l R A e in segment of the ring as 1 2 ring segments;  is the angle traversed by the electron, and R the radius of the ring. We use the standard notation where positive k represents counter-clockwise propagating waves and negative values represent clockwise moving waves. In order to calculate the transmission probability and the circular current as a function of magnetic field we assign scattering amplitudes as shown in Fig. 2b and explained in the main text. For simplicity, we assume in what follows that all scattering amplitudes are real. This assumption implies that no rigid phase shifts occur upon scattering at the junctions consistent with the tight-binding model which conserves the continuity of the wave-functions across the junctions.42, 47, 48 Focusing on junction I (Fig. 2a) we may write the following scattering equation relating the incoming amplitudes to the outgoing amplitudes:           c   R 2 D 2 U 1       a b        a b   Current conservation implies that the scattering amplitude matrix must be unitary, i.e. 2 R 1 D 1 U           2 (A.1)  I (A.2) c       a b    a b        where I is a 33  unit matrix. This provides three independent equations for the four scattering amplitudes, leading to 1 1 2 2 Having characterized the contacts we turn back to the circular setup in Fig. 2a. For a given wavenumber k we can write a scattering equation similar to Eq. (A.1) for contacts I and II, taking into account the spatial phase accumulated by the electrons while traveling along the arms of the ring: 1 2 ;   (A.3)    1  1     a b  ; c  c c 38  R 2 D 2 U 1                  c         a b       a b   R 1  i k R  2    i k R  D e 1 U e 2 (A.4)       2 L 1  i k R  A i k R         1 2  2    (A.5) c   L 2 D 1 U   r B    D e 2 U e 1               a b              a b      To model the influence of an external magnetic field threading the ring we assume that the   zB such magnetic field is homogeneous and perpendicular to the plane of the ring B ,0,0 that the vector potential may be written as: x y z z x y B 0 0 z On the circle defining the ring this becomes  1   A B R 2 z The magnetic phase accumulated by an electron traveling along the ring is thus given by:     2 2   A dl    1 1  (A.8) (A.7) (A.6)     1 2 y x , , 0 d   B R z      sin yB z ,  xB z  sin , cos    , 0    , 0    , 0 , cos sin    , 0   B z      , cos     m   1 2   2 e  1 2        e  1 2   1 2  B  0 e . In the presence of such magnetic field, Eqs. 2 , S R , and  zB S where B 2     0 (A.4) and (A.5) are modified as follows: R 1  B   0  i k R   D e 1 R 2 D 2 U 1                  c          a b      a b      i k R    B   0     U e 2  2                L 2 D 1 U 2                  c          a b      a b     L 1  B   0  i k R     2       D e 2  i k R    B   0     U e 1         (A.9) (A.10) 39 Focusing on a scattering process with incoming electron coming from the right, we set L1 = 0, that is, take zero incoming amplitude on the left lead. Eqs. (A.9)-(A.10) then lead to be 0 2 1   2     2        be ae  2     B   0              B   0  B   0  B   0  i k R    i k R    i k R    i k R    i k R   0  B   0                               Inverting (A.11) yields the wavefunctions amplitudes D and U on the ring segments as a function of the incoming amplitude R1. In particular, the results for D2 and U1 can be used in (A.10) to yield L2 and therefore the transmission probability               0 R  1 R  1 0 D 1 D 2 U 1 U 1   B   0 (A.11)  i k R    i k R    i k R           B   0  B   0         ae be  2        1  0 1  ae ae be     0   k ,  B   2  L 2 R 1  2 R 1 D e 2  i k R    B   0      2     i k R    B   0     2  U e 1 (A.12) which finally results in Eq. (18) Appendix B: The two-level transport model Here we construct a two-level transport model that, for weak molecule-lead coupling, captures the main features observed in the magnetic field dependence of electron transmission through a molecular ring. The validity of a two-level model stems for the fact that in this coupling limit only pairs of molecular levels, degenerate in the limit of zero coupling, are coupled through their mutual interaction with the leads. The Hamiltonian for system two-level the Hamiltonian, considering by obtained be can this      2  for a charged particle moving in a magnetic field. The   P qA V r  Hamiltonian describing a free ( e    ; atomic units are used 1 0V  ) electron (  1 / 2 H   q throughout) moving on a circular ring of radius R lying in the XY plane under a uniform   B B magnetic field oriented in the Z direction, can be written by setting 0, 0, z    r B A   (B.1) y x ,   , 0    1 2 zB 2  This leads to 40  P  A 2     A P P A     2 B z 8 2 2 2 i x y y x            B z 2 H H 1 2 (B.2) 1 2     2         y x     where the central term in the last stage may be identified as the angular momentum component along the Z direction. In circular coordinates with the origin placed at the center of the ring Eq. (B.2) becomes 2 2 2 R B B 1   z z 2 2 8 R 2     in which the last term is an additive constant. The eigenstates of this Hamiltonian can be written in the form 1 2  with the corresponding energy eigenvalues 2 m     0, 1, 2,     (B.3) (B.4) ime    ;    i 2 2  mE  B  0 1 R 2  B S    m     2  Here, B and 0   respectively. Using relation (B.4) the wave functions can be rewritten as are the magnetic flux threading the ring and the flux quantum, (B.5)  i     B   0     2 E R m e      cw 1 2  Noting that Eq. (B.5) may be written in the form 2   ;   ccw i    e 1 2   2 E R m  B   0     , (B.6) 2 2 m 2    , m       E  1 R  B  0 m B z 2         the Hamiltonian of the isolated ring in the subspace of these two levels is given by Eq. (20). When coupled to leads as in Fig. 3, the self-energy terms appearing in Eq. (23c) are given by         12 11             r 0  G L R /    r 0  G L R /  * V  L R 1, / * V  2, 0 0 0 0 V  L R 2, / 0 0  V  L R 1, / 0 0  r 0 G L R / r 0 G L R /     (B.7) r L R /      L R / E E E E E 21 22                                    * V V L R 2, 1, / 2 V 1, L R / r 0 G L R /  E       11      * V 2, V L R / 1, L R / V 2, L R / L R / 2      41 (B.8) Here, is the retarded (r) Green's function of the isolated (0) left/right (L/R) lead and  E G r 0 RL / we assume short range interaction between the ring and the leads, whereby the ring is coupled to the nearest neighbor leads sites. The advanced self-energy is given by  †     r  RL / The coupling matrix elements that appear in expression (B.8) K L R ,  1, 2 ; a RL / E E  ;  . j ,j KV for the self-energy are formally calculated via V  and should therefore be ring lead proportional to the phase of the wave function on the ring. In Eq. (24) we take this phase dependence into account where the specific symmetry of the system (ortho, meta, or para) is taken explicitly into account via the angular separation between the leads. Using the coupling matrix elements given in Eq. (24) we obtain explicit matrix representations for the retarded (and advanced) self-energies in the forms  r  R E  i V   2 1 1   1 1     ;  r L  E  i V    2 1 im 2       e 2 im  e 1     (B.9) And  a  E  r     E  †   , i.e.    i V a  R E 2 1 1   1 1          The broadening matrices Γ, Eq. (23a), and the ring Green's functions, Eq. (23b) are then obtained in the forms 1 im 2   (B.10) i V  ;    E a L   1 e e 2   2 im    R E   V 2 2 1 1   1 1     ;  L  E   V 2  2 1 im 2       e 2 im  e 1     (B.11) r  G E M     EI H  m M   r L  E    r R  E 1     2      E E i V 2    1     im   2     i V e m   2 cos       where we have used Eq. (20) for the Hamiltonian of the isolated ring. Inverting the matrix in a   MG E leads to (B.12) and in the corresponding expression for   m cos  2 i V 2  2 i V e 2  E E  2 im   1  (B.12) 42 1 i V 2   2 E E  2 i V 2    2 i V 2   im    cos  m  2  im   2 2 V 4     m cos  2 i V 2   4 2 cos   m      (B.13)  2 i V e 2  E E   1 r G M  E    E E  1      And the retarded counterpart is † E E  2 2 i V e 2   a G M  E   G r M    E     E E  1  i V 2  E E  2       2   i V 2   im   cos E E  2 2  i V e 2  2  im   2 2 V 4     m cos  2 i V 2       4 2 cos   m  1 i V 2  2  m  2 i V e 2  E E  1  (B.14) With the broadening and Green's functions matrix these explicit expression for representations, evaluating the transmission coefficient (22) becomes a lengthy but straightforward calculation leading to the final result (26). 43 References 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 K. Walczak, Cent. Eur. J. Chem 2, 524 (2004). S. K. Maiti, Chemical Physics 331, 254 (2007). O. Hod, R. Baer, and E. Rabani, The Journal of Physical Chemistry B 108, 14807 (2004). O. Hod, R. Baer, and E. Rabani, Journal of the American Chemical Society 127, 1648 (2005). O. Hod, E. Rabani, and R. Baer, Accounts of Chemical Research 39, 109 (2006). O. Hod, R. Baer, and E. Rabani, Journal of Physics: Condensed Matter 20, 383201 (2008), and references therein. D. Rai, O. Hod, and A. Nitzan, Journal of Physical Chemistry C 114, 20583 (2010), and references therein. S. Nakanishi and M. Tsukada, Jpn. J. Appl. Phys. 37 (1998) pp. , L1400 (1998). S. Nakanishi and M. Tsukada, Physical Review Letters 87, 126801 (2001). M. Ernzerhof, H. Bahmann, F. Goyer, M. Zhuang, and P. Rocheleau, J. Chem. Theory Comput. 2, 1291 (2006). N. Tsuji, S. Takajo, and H. Aoki, Phys. Rev. B 75, 153406 (2007). K. Tagami, M. Tsukada, W. Yasuo, T. Iwasaki, and H. Nishide, Journal of Chemical Physics 119, 7491 (2003). B. Wang, R. Chu, J. Wang, and H. Guo, Physical Review B 80, 235430 (2009). K. Tagami and M. Tsukada, Current Applied Physics 3, 439 (2003). A. M. Jayannavar and P. Singha Deo, Physical Review B 51, 10175 (1995). K. Tagami and M. Tsukada, e-Journal of Surface Science and Nanotechnology 2, 205 (2004). J.-L. Xiao, Y.-L. Huang, and C.-L. Li, Journal of Luminescence 119-120, 513 (2006). Y.-l. Huang and J.-l. Xiao, Superlattices and Microstructures 41, 17 (2007). L. G. Wang, Physica B: Condensed Matter 404, 143 (2009). Note that splitting take place already due to the asymmetric coupling. However, without magnetic field or externally induced current, the split states do not carry net angular momenta. O. Entin-Wohlman, Y. Imry, and A. Aharony, Physical Review Letters 91, 046802 (2003). O. Entin-Wohlman, Y. Imry, and A. Aharony, Physical Review B 70, 075301 (2004). I. Barth, J. Manz, Y. Shigeta, and K. Yagi, J. Am. Chem. Soc. 128, 7043 (2006). I. Barth and J. Manz, Angewandte Chemie-International Edition 45, 2962 (2006). K. Nobusada and K. Yabana, Phys. Rev. A 75, 032518 (2007). G. F. Quinteiro and J. Berakdar, Opt. Express 17, 20465 (2009). A. Matos-Abiague and J. Berakdar, Phys. Rev. Let. 94, 166801 (2005). 44 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 S. S. Gylfadottir, M. Nita, V. Gudmundsson, and A. Manolescu, Physica E: Low- dimensional Systems and Nanostructures 27, 278 (2005). A. S. Moskalenko and J. Berakdar, Physical Review B 80, 193407 (2009). A. S. Moskalenko and et al., EPL (Europhysics Letters) 78, 57001 (2007). Y. V. Pershin and C. Piermarocchi, Physical Review B 72, 245331 (2005). E. Räsänen, A. Castro, J. Werschnik, A. Rubio, and E. K. U. Gross, Physical Review Letters 98, 157404 (2007). E. Rasanen, A. Castro, J. Werschnik, A. Rubio, and E. K. U. Gross, Physical Review B (Condensed Matter and Materials Physics) 77, 085324 (2008). D. Rai, O. Hod, and A. Nitzan, The Journal of Physical Chemistry Letters, 2118 (2011). O. Hod, R. Baer, and E. Rabani, Phys. Rev. Lett. 97, 266803 (2006). V. Ben-Moshe, A. Nitzan, S. S. Skourtis, and D. Beratan, J. Phys. Chem. C 114 8005 (2010). V. Ben-Moshe, D. Rai, A. Nitzan, and S. S. Skourtis, J. Chem. Phys. 133, 054105 (2010). F. London, J. Phys. Radium 8, 397 (1937). V. A. Geyler, V. V. Demidov, and V. A. Margulis, Technical Physics 48, 661 (2003). S. Datta, Electric transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1995). Y. Imry, Introduction to Mesoscopic Physics (Oxford University Press, Oxford, 2008). M. Buttiker, Y. Imry, and M. Y. Azbel, Physical Review A 30, 1982 (1984). Y. Gefen, Y. Imry, and M. Y. Azbel, Physical Review Letters 52, 129 (1984). M. Cahay, S. Bandyopadhyay, and H. L. Grubin, Physical Review B 39, 12989 (1989). C. Benjamin, S. Bandopadhyay, and A. M. Jayannavar, Solid State Communications 124, 331 (2002). A. M. Jayannavar and C. Benjamin, Pramana-J. Phys. 59, 385 (2002). B. Kubala and J. König, Physical Review B 67, 205303 (2003). I. A. Ryzhkin, Phys. Solid State 41, 1901 (1999). A. Aharony, O. Entin-Wohlman, B. I. Halperin, and Y. Imry, Physical Review B 66, 115311 (2002). T. Hansen, G. C. Solomon, D. Q. Andrews, and M. A. Ratner, J Chem Phys 131, 194704 (2009). the actual parameters vary slightly for the different molecules used in our calculations. We have checked that these variations do not affect our qualitative observations. In the para case, one of the levels actually remains at that energy. 45 This statement holds for the voltage used (2 eV) for which the circular current goes through a maximum. The dependence of the circular current on the molecule-leads coupling is more pronounced at other voltages. S. K. Maiti, Journal of Computational and Theoretical Nanoscience 6, 1561 (2009). S. Nakanishi, R. Tamura, and M. Tsukada, Jpn. J. Appl. Phys 37, 3805 (1998). M. Buttiker, Phys. Rev. B 33, 3020 (1986). J. L. D’Amato and H. M. Pastawski, Phys. Rev. B 41, 7411 (1990). M. Dey, S. K. Maiti, and S. N. Karmakar, Organic Electronics 12, 1017 (2011). B. Szafran, M. R. Poniedziałek, and F. M. Peeters, Europhys. Let. 87, 47002 (2009). D. Sánchez and K. Kang, Physical Review Letters 100, 036806 (2008). C. A. Marlow, R. P. Taylor, M. Fairbanks, I. Shorubalko, and H. Linke, Physical Review Letters 96, 116801 (2006). G. L. J. A. Rikken and P. Wyder, Physical Review Letters 94, 016601 (2005). A. Ishizaki, T. R. Calhoun, G. S. Schlau-Cohen, and G. R. Fleming, Physical Chemistry Chemical Physics 12, 7319 (2010). 53 54 55 56 57 58 59 60 61 62 63 46 

1601.03861

1

1601

Decoherence and Decay of Two-level Systems due to Non-equilibrium Quasiparticles

It is frequently observed that even at very low temperatures the number of quasiparticles in superconducting materials is higher than predicted by standard BCS-theory. These quasiparticles can interact with two-level systems, such as superconducting qubits or two-level systems (TLS) in the amorphous oxide layer of a Josephson junction. This interaction leads to decay and decoherence of the TLS, with specific results, such as the time dependence, depending on the distribution of quasiparticles and the form of the interaction. We study the resulting decay laws for different experimentally relevant protocols.

cond-mat.mes-hall

cond-mat

Decoherence and Decay of Two-level Systems due to Non-equilibrium Quasiparticles 1Institut fur Theoretische Festkorperphysik, Karlsruhe Institute of Technology, D-76128 Karlsruhe, Germany Sebastian Zanker,1 Michael Marthaler,1 and Gerd Schon1 It is frequently observed that even at very low temperatures the number of quasiparticles in superconducting materials is higher than predicted by standard BCS-theory. These quasiparticles can interact with two-level systems, such as superconducting qubits or two-level systems (TLS) in the amorphous oxide layer of a Josephson junction. This interaction leads to decay and decoherence of the TLS, with specific results, such as the time dependence, depending on the distribution of quasiparticles and the form of the interaction. We study the resulting decay laws for different experimentally relevant protocols. INTRODUCTION Superconducting quantum devices have a wide range of applications. Due to the weak dissipation in the su- perconducting state they are promising candidates for building large scale quantum information systems [1, 2]. They are easily controlled and measured by electromag- netic fields, but for the same reason they couple rather strongly to the environment and are prone to decoher- ence [3]. Much effort has been put into understanding and minimizing various noise sources. One ubiquitous source of decoherence arises from two level systems (TLS), such as bistable defects residing in dielectric substrates, disordered interfaces, surface ox- ides, or inside the barriers of Josephson junctions [4]. While originally introduced to explain anomalous prop- erties of glasses at low temperatures [5] there is evidence that they are also an important source of decoherence for superconducting qubits [1] or superconducting res- onators [6]. A bath of TLS can explain the 1/f noise which limits the performance of many devices [7], while fluctuations of very slow TLS induce long-time parame- ter shifts. Still, the microscopic origin of those TLS re- mains unclear. Some potential sources are small groups of atoms that tunnel between two stable positions, or dangling bonds, or hydrogen defects. A better experi- mental as well as theoretical understanding of those TLS has been the focus of much recent work [8–12]. Recent experiments demonstrated the coherent control of TLS, residing inside the amorphous layer of a phase qubit’s Josephson junction, with help of this qubit [13]. It was possible to carry out typical coherence experiments as used in magnetic resonance or other qubit experiments [14]. The aim of those experiments is a better under- standing of the microscopic nature of individual TLS as well as their respective environment responsible for TLS state fluctuations, with the ultimate goal to reduce their detrimental effects. In this paper we analyze the decoherence of charged TLS residing inside the amorphous layer of a Josephson junction [4, 15] due to scattering and tunneling of non- equilibrium quasiparticles in the superconducting leads. Experiments suggest that even at low temperatures the number of quasiparticles is large as compared to the pre- dictions of equilibrium BCS theory [16]. Similar as ob- served for superconducting qubits [17] or the dynamics of Andreev bound states [18] the scattering with quasi- particles provides an intrinsic noise source also for the TLS. We investigate the TLS decoherence properties due to the coupling between the TLS and the quasiparticles. We find characteristic differences for quasiparticles which scatter back to the same superconducting electrode and those which tunnel across the junction. The effect of the latter depends on the phase difference. Our results ap- ply both for the TLS, which are the main focus of the present paper, as well as for qubit decoherence [19]. In fact for the latter the effect of tunneling electrons is more pronounced. THE MODEL We consider the scattering of quasiparticles from a two- level system located inside the amorphous barrier of an aluminum oxide Josephson junction. The model Hamil- tonian reads as H = HTLS + Hqp + HC, (1) where HTLS is the TLS Hamiltonian, Hqp are the free quasiparticle Hamiltonians of the left and right lead and HC is the coupling between both subsystems. Since the microscopic nature of TLS remains unclear, we use the phenomenological TLS standard model for HTLS of Ref. [20]. It describes the TLS as an effective charged particle trapped in a double well potential with asymme- try
between the two potential minima and tunneling amplitude ∆0,
2 ∆0 2 σz + σx. HTLS = 2 ETLS σz with ETLS = (cid:112)∆2 In the TLS eigenbasis the Hamiltonian reduces to HTLS = 1 0 +
2. Electrons in nearby leads couple to the TLS’ electric dipole mo- ment and induce an interaction that is well established in the context oft metallic glasses [21]: (2) HC = σz V = σz † gkk(cid:48)c kck(cid:48) + h.c. . (3) (cid:88) (cid:16) kk(cid:48) (cid:17) 6 1 0 2 n a J 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 6 8 3 0 . 1 0 6 1 : v i X r a Here, gkk(cid:48) is the coupling strength between electrons and the TLS dipole and ck is an electron annihilation opera- tor with multi-index k = {(cid:126)k, σ, α} that includes electron momentum (cid:126)k, spin σ, and the index α = l, r of the lead (left or right), where the electron resides. In general we can distinguish two processes: Electrons that tunnel through the junction (α (cid:54)= α(cid:48)) while interacting with the TLS and electrons, that scatter back into their original lead (α = α(cid:48)). Because of the exponentially decaying electron wave function inside the junction and the local- ized character of the TLS wave function, the interaction rapidly decreases for TLS away from the junction edges. We expect that for most TLS scattering electrons are the main source of TLS decoherence, while the influence of tunneling electrons is insignificant. This is quite differ- ent for the decoherence of a qubit, where only tunneling electrons couple to the qubit and induce decoherence. Furthermore, electrons that contribute to TLS decoher- ence have energies close to the Fermi energy with mo- mentum (cid:126)k = kF . Hence, we take the direction average and introduce the direction-averaged coupling constant αα(cid:48) ≡ (cid:104)g2 kk(cid:48)(cid:105). Since (cid:126)k ≈ kF the averaged coupling con- g2 stant does not depend on energy. The free particles in the leads are Bogoliubov quasi- particles with mixed electron- and hole-like nature and † creation operators a k and Hamiltonian Hqp = Their energy is Ek =(cid:112)ξ2 bcs with ξk being the elec- tron energy in the normal state. Rewriting the coupling Eq. (3) in terms of quasiparticle operators we find k + ∆2 V = gαα(cid:48) eiϕ/2ukuk(cid:48) − e−iϕ/2vkvk(cid:48) † a kak(cid:48) + h.c. (cid:16) (cid:88) kk(cid:48) (5) with coherence factors u2 2 (1 + ξk/Ek). For tunneling quasiparticles ϕ is the superconducting phase difference across the junction. The phase difference ϕ vanishes for scattering quasiparticles. k = 1 − v2 k = 1 An important quantity in the context of decoherence is the noise spectral density SV (ω) = (cid:104) V (t) V (0)(cid:105)eiωt (6) where the average is over the quasiparticle states. With Eq. (5) we find SV (ω) =4N 2 0 g2 dEdE(cid:48) ρ(E)ρ(E(cid:48)) ∆bcs 1 − ∆2 BCS EE(cid:48) cos ϕ × {fα(E)[1 − fα(cid:48)(E(cid:48))]δ(E − E(cid:48) + ω) + fα(cid:48)(E(cid:48))[1 − fα(E)]δ(E(cid:48) − E + ω)} (7) with the density of states at the Fermi energy of the normal state N0, the BCS density of states ρ(E) = 2π (cid:90) dt αα(cid:48)(cid:82) ∞ ×(cid:16) ∆bcs (cid:82) ∞ (cid:88) k † kak Eka (4) UI (t, t0) = Texp −i dt(cid:48) HC,I (t(cid:48)) , (10) where UI (t, t0) is the time evolution operator (cid:20) (cid:90) t (cid:21) 2 E/(cid:112)E2 − ∆2 BCS and the quasiparticle distribution func- † tion f (Ek) = (cid:104)a kak(cid:105). We assume that the quasiparticles can be described with equilibrium BCS gap and density of states, but with a non-equilibrium distribution func- tion f (Ek). Due to the square root singularity of the BCS density of states, the noise spectral density is log- divergent at low frequencies for tunneling quasiparticles, e.g. ϕ (cid:54)= 0. Scattering quasiparticles, e.g. ϕ = 0 have a finite spectral density at low frequencies. TLS DECOHERENCE The time evolution of the TLS in the presence of the quasiparticle reservoirs is best described with the help of the reduced density matrix ρ(t) = Tr [
(t)]qp ≡ (8) (cid:18) ρ0 ρ01 (cid:19) ρ10 ρ1 that is obtained from the full density matrix after tracing out quasiparticle degrees of freedom. It evolves according to ρ(t) = e−iHTLSt Trqp UI (t)
(0)U † I (t) eiHTLSt (9) (cid:104) (cid:105) t0 and HC,I (t(cid:48)) the coupling in the interaction picture. It is assumed that the initial density matrix,
(t0) ≡ ρ(t0)ρqp(t0), factorizes into a quasiparticle and TLS com- ponent and that initial correlations are irrelevant on ex- perimental time scales. We can distinguish two effects due to the quasiparticles: Decay and decoherence. The former describes exponential decay of diagonal elements of the TLS density matrix to their stationary state val- ues, while the latter concerns the decay of off-diagonal elements. Transforming the coupling (3) into the TLS energy basis we find two contributions σz →
/ETLS σz + ∆0/ETLS σx. The off-diagonal term ∼ σx induces transi- tions and is responsible for the decay rate Γ1. It can be calculated in first-order perturbation theory [22], Γ1 = ∆2 0 E2 TLS [SV (ETLS) + SV (−ETLS)] . (11) The diagonal coupling ∼ σz generates pure dephasing, determined by the low-energy part of the spectral den- sity. Due to the strong energy dependence in this energy range pure dephasing does not lead to a simple exponen- tial decay law. Rather the off-diagonal elements of the density matrix take the form ρ10/01(t) = e±iETLSte− 1 2 Γ1te−h(t) (12) (cid:17) (cid:17) where h(t) describes the deviations from the simple ex- ponential decay. It reduces to a linear time-dependence only for flat spectral densities and long times. This form of the density matrix follows from Eq. (9) and the fact that the TLS–quasiparticle coupling is diagonal in TLS space for pure dephasing. Due to the simple coupling we can pull all TLS operators through the trace and arrive at the form for the TLS density matrix given in (12) with (cid:20) (cid:26) (cid:90) t V (t(cid:48))dt(cid:48)(cid:27) (cid:21) e−h(t) ≡ Trqp TC exp i ρqp(t0) . (13) t0 The contour time-ordering operator Tc orders along a contour from t0 to t and back again. To further evaluate that expression we expand the exponential and introduce an additional approximation [19]: We assume that we can split averages over Vi operators in the form (cid:104)V (t1)V (t2) . . . V (tn)(cid:105) = (cid:104)V (ti)V (tj)(cid:105)···(cid:104)V (tk)V (tl)(cid:105) (cid:89) perm (14) With this approximation the quasiparticles behave simi- lar to a Gaussian noise source [22] and we find 3 NON–EQUILIBRIUM QUASIPARTICLES Based on the dephasing functions (16) and (15) as well as the decay rate (11) we are ready to analyze the de- phasing process due to quasiparticles. The spectral den- sity (7) depends on the quasiparticle distribution func- tion. Several experiments provide evidence that, even at low temperatures where quasiparticles should be ex- ponentially suppressed with the BCS gap, finite den- sities of quasiparticles remain, estimated to be nqp ∼ 10−6 · ∆BCSN0 [16]. Similar to the treatment of non- equilibrium quasiparticles for qubit decoherence, e.g. in Ref. [24], we assume that both, the BCS gap and the density of states are not changed, but the distribution function is of a non-equilibrium form. Although the ex- act form depends on experimental details most of the non-equilibrium quasiparticles have energies close to the superconducting gap because of scattering with phonons and among each other. We therefore assume that the distribution function has a width δ above the gap, which for a Fermi distribution is determined by temperature, δeq ∼ kBT , but here it is treated as a parameter. In the following we derive analytical forms for the different rates in the experimental relevant long-time limit t (cid:29) δ−1. hR(t) = t2 dω Sqp(ω) sin2 (ωt/2) (ωt/2)2 (15) Decay (cid:90) (cid:90) This specific form for pure dephasing is well established in the context of magnetic resonance or qubit experi- ments in a Ramsey protocol. The weighting function g(ωt) = sin2(ωt/2)/(ωt/2)2 has a pronounced peak for zero energy and decreases rapidly for larger ω. There- fore, the Ramsey-type experiments are sensitive to the spectral density at low energies. Within the Gaussian approximation we can extend our analysis to more sophisticated measurement protocols, such as spin echo or more complicated refocusing tech- niques that suppress low-frequency noise contributions. The dephasing function h(t) for those protocols looks very much like the Ramsey function but with different filter functions depending on the particular pulse proto- col [23] h(t) = t2 dω Sqp(ω) g(ωt). (16) The relevant energy scale for TLS decay is the TLS energy splitting ETLS as evident from Eq. (11). The TLS which can be probed by a qubit have energy splittings close to that of the qubit and thus fulfill ETLS (cid:29) δ. In order to evaluate the spectral density in this limit we introduce the normalized quasiparticle density (cid:90) ∞ 1 ∆bcs ∆BCS xqp = dE ρ(E)f (E). (18) For typical TLS energies ∆BCS (cid:29) ETLS (cid:29) δ we can ap- ply the ’low-energy’ approximation to evaluate the spec- tral density at the TLS energy [17, 24]. In this limit all quasiparticle energies in Eq. (7) can be set to ∆BCS. The only exception is the quasiparticle energy in the divergent BCS density of states together with the corresponding distribution function f (E). They enter in the quasipar- ticle density (18) and the spectral density, describing the decay, reads as E.g. for spin echo, which is the ’first order’ improvement to the Ramsey experiment, the filter function is ge(ωt) = sin4 (ωt/4) / (ωt/4)2 (17) with a maximum slightly shifted to higher frequencies. For typical experimental times in the range of microsec- onds the filter function measures the spectral density at energy equivalents of several MHz. SV (ETLS) = 4N 2 0 g2 αα(cid:48)∆BCS(xqp,α + xqp,α(cid:48)) × ρ(ETLS + ∆BCS) 1 − ∆BCS ∆BCS + ETLS cos ϕ , (19) while ∼ S(−ETLS) and the resulting excitation rate is much smaller. This form for the high-energy spectral density and thus decay rate Γ1 is well established in the context of qubit decay due to quasiparticles [16, 25]. (cid:18) (cid:19) Ramsey and Spin Echo Dephasing To calculate the Ramsey dephasing rate (15) we need an approximation for the low-energy spectral density. We proceed as in Ref. [19] and split the spectral density into a regular and a divergent part. For low energies, the former is flat and can be considered constant, while the latter is log divergent, Sdiv ∼ (1 − cos ϕ) log(δ/ω). We find the Ramsey dephasing function hR(t) = Γ∗ 2 t + πN 2 0 g2 αα(cid:48)[fα(∆BCS) + fα(cid:48)(∆BCS)] ×[1 − cos(ϕ)] [γe − 1 + log(4δ · t)] t (20) with the Euler constant γe and the pure dephasing rate Γ∗ 2 = 8πN 2 0 g2 αα(cid:48) [fα(E) + fα(cid:48)(E)]dE . (21) (cid:90) ∞ ∆BCS For scattering quasiparticles we have cos ϕ = 1 and the divergent contribution vanishes. Thus, scattering parti- cles induce simple exponential dephasing ∼ e−Γ∗ 2 t with rate Γ∗ 2. On the other hand, the dephasing effect of tun- neling quasiparticles is dominated by the second term in (20) stemming from the divergent contribution. The spin echo protocol filters out low energies, but we observe that the relevant energy scales are still much smaller than the width of the quasiparticles. Thus the calculation proceeds similar to the calculation for Ram- sey dephasing, with the result he(t) = Γ∗ 0 g2 2 t + πN 2 ×(1 − cos(ϕ)) αα(cid:48)[fα(∆BCS) + fα(cid:48)(∆BCS)] [γe − 1 + log(δ · t)] t. (22) 1 2 Since the non-divergent part of the spectral density is almost flat for the relevant frequency scales, spin echo does not improve coherence and, similar to white-noise- induced dephasing, the pure dephasing rate Γ∗ 2 is the same for both protocols, spin echo and Ramsey. Thus, for scattering electrons there is no measurable difference between both experimental protocols.aeternus20!0 On the other hand, for tunneling quasiparticles the second term comes into play and dominates dephasing. In this limit the ratio between spin echo and Ramsey is lim t→∞ he(t) hr(t) = 1 2 . (23) This improvement in dephasing time due to spin echo is typical for noise with divergent spectral density at small energies and could be measured in an experiment. CONCLUSION We analyzed the decoherence of TLS located in disor- dered systems in vicinity to superconducting leads, es- pecially inside the amorphous layer of a Josephson junc- tion. We distinguish in our analysis between scattering 4 quasiparticles, that cause decoherence for all the TLS mentioned above, and quasiparticles that tunnel through the junction. If there exists a phase difference between the superconducting electrodes, the latter exhibit a log- divergent spectral density for low energies leading to in- creased and time-dependent dephasing rates while the difference is negligible for the TLS decay. We further showed that the spin echo technique reduces the TLS de- coherence rate due to tunneling particles, while it has little effect for scattering quasiparticles. The results ob- tained for tunneling quasiparticles, arising form a diver- gent spectral density, apply to single-junction qubits, and they are sensitive to refocusing techniques. This opens possibilities to analyze the quasiparticle-environment of a qubit. ACKNOWLEDGMENTS We thank A. Bilmes, J. Lisenfeld and A. Shnirman for many useful discussions during the work on this paper. This work was supported by the German-Israeli Founda- tion for Scientific Research and Development (GIF). [1] Barends R., Kelly J., Megrant A., Veitia A., Sank D., Jef- frey E., White T. C., Mutus J., Fowler A. G., Campbell B., Chen Y., Chen Z., Chiaro B., Dunsworth A., Neill C., O’Malley P., Roushan P., Vainsencher A., Wenner J., Korotkov A. N., Cleland A. N., and Martinis John M., Nature 508, 500 (2014). [2] M. H. Devoret and R. J. Schoelkopf, Science 339, 1169 (2013). [3] A. Shnirman, Y. Makhlin, and G. Schon, Physica Scripta 2002, 147 (2002). [4] R. W. Simmonds, K. M. Lang, D. A. Hite, S. Nam, D. P. Pappas, and J. M. Martinis, Phys. Rev. Lett. 93, 077003 (2004). [5] P. Esquinazi, Tunneling Systems in Amorphous and Crystalline Solids (Springer, 1998). [6] L. Faoro and L. B. Ioffe, Phys. Rev. Lett. 109, 157005 (2012). [7] J. Schriefl, Y. Makhlin, A. Shnirman, and G. Schon, New Journal of Physics 8, 1 (2006). [8] T. C. DuBois, S. P. Russo, and J. H. Cole, New Journal of Physics 17, 023017 (2015). [9] W. A. Phillips, Reports on Progress in Physics 50, 1657 (1987). [10] A. M. Holder, K. D. Osborn, C. J. Lobb, and C. B. Musgrave, Phys. Rev. Lett. 111, 065901 (2013). [11] D. Gunnarsson, J.-M. Pirkkalainen, J. Li, G. S. Paraoanu, P. Hakonen, M. Sillanp, and M. Prunnila, Su- perconductor Science and Technology 26, 085010 (2013). [12] J. M. Martinis, K. B. Cooper, R. McDermott, M. Stef- fen, M. Ansmann, K. D. Osborn, K. Cicak, S. Oh, D. P. Pappas, R. W. Simmonds, and C. C. Yu, Phys. Rev. Lett. 95, 210503 (2005). 5 [13] J. Lisenfeld, C. Muller, J. H. Cole, P. Bushev, A. Lukashenko, A. Shnirman, and A. V. Ustinov, Phys. Rev. Lett. 105, 230504 (2010). [14] J. Lisenfeld, A. Bilmes, S. Matityahu, S. Zanker, M. Marthaler, G. Weiss, A. Shnirman, M. Schechter, and A. Ustinov, “Decoherence of individual two-level systems in dependence of their strain-tuned asymmetry energy,” (2015), unpublished. [15] G. J. Grabovskij, T. Peichl, J. Lisenfeld, G. Weiss, and A. V. Ustinov, Science 338, 232 (2012). [16] J. M. Martinis, M. Ansmann, and J. Aumentado, Phys. Rev. Lett. 103, 097002 (2009). [17] J. Leppakangas and M. Marthaler, Phys. Rev. B 85, 144503 (2012). [18] D. G. Olivares, A. L. Yeyati, L. Bretheau, i. m. c. O. and C. Urbina, Phys. Rev. B 89, Girit, H. Pothier, 104504 (2014). [19] S. Zanker and M. Marthaler, Phys. Rev. B 91, 174504 (2015). [20] W. Phillips, Journal of Low Temperature Physics 11, 757 (1973). [21] J. L. Black and B. L. Gyorffy, Phys. Rev. Lett. 41, 1595 (1978). [22] G. Ithier, E. Collin, P. Joyez, P. J. Meeson, D. Vion, D. Esteve, F. Chiarello, A. Shnirman, Y. Makhlin, J. Schriefl, and G. Schon, Phys. Rev. B 72, 134519 (2005). [23] Bylander Jonas, Gustavsson Simon, Yan Fei, Yoshi- hara Fumiki, Harrabi Khalil, Fitch George, Cory David G., Nakamura Yasunobu, Tsai Jaw-Shen, and Oliver William D., Nat Phys 7, 565 (2011). [24] G. Catelani, S. E. Nigg, S. M. Girvin, R. J. Schoelkopf, and L. I. Glazman, Phys. Rev. B 86, 184514 (2012). [25] G. Catelani, R. J. Schoelkopf, M. H. Devoret, and L. I. Glazman, Phys. Rev. B 84, 064517 (2011). 

1205.2029

1

1205

Mode bifurcation on the rythmic motion of a micro-droplet under stationary DC electric field

Accompanied by the development of microfabrication techniques, such as MEMS and micro-TAS, there has been increasing interests on the methodology to generate a desired motion on a micro object in a solution environment. It is well know that the principle to create an electric motor in a macroscopic scale is not applicable to a micro system because of the enhancement of sticky interaction and higher viscosity in micrometer sized system. On the other hand, living organisms generate various motions under isothermal condition in a well-regulated manner. Despite the past intensive studies, the underlying mechanism of the biological molecular motors has not been unveiled yet. Under such development stage of science and technology at the present, we are performing the study toward real-world modeling on the emergence of regular motion in micro system. In this paper, we report a novel experimental system on the generation of regular movement, such as periodic go-back motion and circular motion, for a micro object under DC electric field.

cond-mat.mes-hall

cond-mat

Mode bifurcation on the rythmic motion of a micro-droplet under stationary DC electric field Tomo Kurimura,1, a) Masahiro Takinoue,2 Masatoshi Ichikawa,1 and Kenichi Yoshikawa3 1)Department of Physics, Graduate School of Science, Kyoto University, Kitashirakawa-Oiwake-cho, Sakyo-ku, Kyoto 606-8502 Japan 2)Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology, 4259 Nagatsuta-cho, Midoriku, Yokohama, Kanagawa 226-8503, Japan 3)Faculty of Biological and Medical Sciences, Doshisha Univ.1-3 Tataramiyakodani, Kyotanabe, Kyoto 610-0394, Japan (Dated: May 2012) 2 1 0 2 y a M 9 ] l l a h - s e m . t a m - d n o c [ 1 v 9 2 0 2 . 5 0 2 1 : v i X r a a)Electronic mail: [email protected] 1 I. INTRODUCTION Accompanied by the development of microtechnology, such as MEMS and µTAS, there is increasing interest on the methodology to realize a desired motion of a micro object in a solution environment. It is well known that the principle to create an electric motor in a macro system is not applicable to micro system because of the enhanced sticky interaction and higher viscosity in micrometer sized system. On the other hand, living organisms generate various motions on microscopic scale under isothermal condition. Despite the past intensive studies1,2, the underlying mechanism of the biological molecular motors has not been fully unveiled yet. Under such development status of science and technology on micro-motor sat the present, we report a simple motoring system which work smoothly in a microscpic scale. Recently, we found that rhythmic motion is generated for an aqueous droplet in an oil phase under DC voltage on the order of 50 - 100 V. We have already reported some experiments and models for a w/o droplet under DC electric field.3,4 There are some reports about experiments of w/o droplets moving under electrical field5,6, bouncing and being absorbed on a surface between water and oil7, deforming and spliting8,9. Manipulating this kind of droplet, which is interesting as the model of the cell10 -- 12, the micro-sized reactor, by optical tweezers13, by micro channel14,15. And manipulating the cells or micro objects by electrical field has been attempted16, to know manipulating this kind of droplets in detail will help this in the future. In the present article, we will show that rhythmic motion on micro-droplet is induced under the DC potential on the order of several volts. We will also propose a simple mathematical model to reproduce the rhythmic motion and mode bifurcation. II. EXPERIMENTAL A schematic illustration of the experimental setup is given in FIG.1. A water droplet was suspended in mineral oil on a glass slide, and constant voltage was applied to the droplet using cone-shaped tungsten electrodes. Droplet motion was observed using an optical microscope (KEYENCE , Japan). The w/o droplet was generated using a vortex mixer as follows. We prepared mineral 2 oil including surfactant: 10µm surfactant, dioleylphosphatidylcholine (DOPC) (Japan), was solved in mineral oil (Nacalai Tesque, Japan) by 90 min sonication at 50 ◦C. 2µl ultrapure water (Millipore, Japan) was added to 100 µl of the prepared mineral oil, and then agitated by a vortex mixer for approximately 3 s. (cid:80)(cid:67)(cid:75)(cid:70)(cid:68)(cid:85)(cid:74)(cid:87)(cid:70)(cid:1)(cid:77)(cid:70)(cid:79)(cid:84)(cid:1) (cid:80)(cid:71)(cid:1)(cid:78)(cid:74)(cid:68)(cid:83)(cid:80)(cid:84)(cid:68)(cid:80)(cid:81)(cid:70) (cid:78)(cid:74)(cid:79)(cid:70)(cid:83)(cid:66)(cid:77)(cid:1)(cid:80)(cid:74)(cid:77) (cid:88)(cid:66)(cid:85)(cid:70)(cid:83)(cid:1)(cid:69)(cid:83)(cid:80)(cid:81)(cid:77)(cid:70)(cid:85) (cid:55) (cid:45) (cid:72)(cid:77)(cid:66)(cid:84)(cid:84)(cid:1)(cid:81)(cid:77)(cid:66)(cid:85)(cid:70) FIG. 1. Schematic representation of the experimental setup. Mineral oil containing water droplets was placed on a glass slide and couple of electrodes was situated inside the oil phase. V: Applied DC voltage. L: distance between the electrodes. III. RESULTS FIG.2 exemplifies the motion of a droplet under DC electric field, indecating the occur- rence of the periodic go-back motion between the electrodes accompanied by the increase of the electical potential. In the experiments, we observed two following types of behavior: oscillatory and stationary. These behaviors switch each other depending on the applied volt- age. When the distance between two electrodes was 213µm [FIG.2(a)], the droplet started the motion with the applied voltage above 16.3 V. When the distance between two electrodes was 141µm [FIG.2(b)], the droplet started moving with the applied voltage above 13.7 V. FIG.3 shows the diagram of the mode of droplet behavior depending on the applied voltage with the size of the droplet is -- -. When the distance of two electrodes is below approximately 70µm , droplets are sticked to an electrode (adhered). The diagram indecates 3 that the threshold of applied voltage is roughly propotional to the distance between two electrodes. (a) L = 213µm (b) L = 141µm V = 16.0V V = 16.3V t=0 [s] 0 1 ] s [ t 2 3 V = 13.0V V = 13.7V 0 1 2 3 100μm FIG. 2. Spatio-temporal diagram on the motion of a droplet with the diameter of 34 m at (a) L=213m, (b) L=141m.@Bifurcation from the stationary state into an oscillatory state is induced by the increase of the applied voltage. IV. DISCUSSION We propose a model to describe the oscillatory-stationary motion of w/o droplets. In an eqation of motion at a micrometer scale, a viscosity term is more dominant than inertia term because the Reynolds number, Re, is rather small; Re = ρvd/η ∼ 10−9 ≪ 1, where ρ(∼ 103kg/m3) and η(∼ 103Pas) are the density and viscosity of the mineral oil, respectively, and v(∼ 10−4m/s) and d(∼ 10−5m) are the velocity and diameter of the water droplet. Therefore, an over-damped eqation of motion under the constant electric field, E, is given 4  25 20 ] V [ V 15 10 5 0 0 Diameter of the droplet 20µm 34µm 50µm d Oscillatory Adhered 2-(b) 2-(a) Stationary 50 100 150 200 250 L [µm] FIG. 3. Phase diagram for the mode bifurcation between rhythmic motion and stationary state as obderved for the droplets with different diameter, where each point represent the threshold value on the bifurcation. The two arrows are correspond to the mode bifurcation as shown in FIG.2 by k x = qE + α∇E2 (1) where k(= 6πηd ∼ 10−7kg/s) is a coefficient of viscosity resistance, and k x represents the viscosity resistance for a moving droplet with diameter d and velocity x. qE and α∇E2 indicate an electric force and a dielectric force acting on the droplet with charge q and polarizability α (Il)17. Here we assume that the time-dependent rate od the charge, q, is described as q = −βǫx3 − q t0 (2) where β is the proportionality coefficient, and ǫ is the constrant is proportional to the magnitude of the electrical field. The first term of this means the time-dependence of charge is in proportion to the number of lines of electric force. The second term means the charge leak, and t0 is the relaxing time. We would like to consider the condition where the droplet stays on the same position between two electrodes. When the electric field can be written as E = (Ex, Ey), The force on the second term in the right hand of Eq. (1) is caused by the number of the lines of electric force penetrating the droplet. Comparing with the size of droplet, the change of Ex 5 along x axis can be neglected. By considering the symmetry of the system, we simply adapt that the change of E2 y along x axis is written as E2 y = ǫ (cid:0)−(x + 1)2(x − 1)2 + 2(cid:1) . Then the x component of eq.(1) is given as k x = ∼= qEx + α∂xE2 qEx + α∂xE2 y =qEx − 4αǫx(x + 1)(x − 1) x = Ex k q − 4αǫ k x(x + 1)(x − 1) (3) (4) For simplicity, we introduce the following parameters: Ex/k = a, 4αǫ/k = e and keβ/4α = γ. Then Eq.(2) and Eq.(4) can be written as x = − ex(x + 1)(x − 1) + αq q = − γex3 − q/t0 (5) (6) FIG.4 shows the result of numeric calculation with these equations, where the time and space scales ,T and X, are arbitrary. The change of the distance between the electorodes corresponds to the change of the magnitude of the electric field. For example, if L becomes larger, a and e become smaller. In FIG.4, a and e in (b) are larger than those in (a). The frequency of the back-and-force motion of the droplet is faster when the electric field between the electrodes becomes stronger. Thus, our numerical model reproduces the essential aspect of the rhythmic motion of a droplet under DC voltage. V. ACKNOWLEDGEMENTS REFERENCES 1Y. Hiratsuka, M. Miyata, T. Tada, and T. Q. P. Uyeda, Proc. Natl. Acad. Sci. U.S.A. 103, 13618 (2006). 2M. G. L. van den Heuvel and C. Dekker, Science 317, 333 (2007). 3Hase, et al., PRE 74, 046301(2006). 4Takinoue, et al. Appl. Phys. Lett. 96, 104105 (2010) 6 (a) 0 5 10 15 τ 20 25 30 35 40 (b) 0 5 10 15 τ 20 25 30 35 40 -0.8 -0.4 0 X 0.4 0.8 -1 -0.8 -0.4 0.4 0.8 1 0 X FIG. 4. Numerical results on the Spatio-Temporal diagram with eqs. (5) and (6), the common parameters are t0 = 0.5 and γ = 0.1. The parameters a and e are changed at T = 15. Initial state is the same in both graphs; a = 100 and e = 2. After T = 15 in (a), a = 200 and e = 4. In (b), a = 250 and e = 5. (a) L = 213µm (b) L = 141µm 0 1 t 2 3 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 x/L x/L FIG. 5. Experimental results on the Spatio-Temporal diagram 7 5T. Mochizuki, Y. Mori, and N. Kaji, AIChE J. 36, 1039 (1990). 6Y. Jung, H. Oh, and I. Kang, J. Colloid Interface Sci. 322, 617 (2008). 7W. D. Ristenpart, J. C. Bird, A. Belmonte, F. Dollar, and H. A. Stone, Nature (London) 461, 377 (2009). 8J. S. Eow, M. Ghadiri, and A. Sharif, Colloids Surf., A 225, 193 (2003). 9S. Teh, R. Lin, L. Hung, and A. Lee, Lab Chip 8, 198 (2008). 10A. V. Pietrini and P. L. Luisi, ChemBioChem 5, 1055 (2004). 11D. S. Tawfik and A. D. Griffiths, Nat. Biotechnol. 16, 652 (1998). 12M. Hase and K. Yoshikawa, J. Chem. Phys. 124, 104903 (2006). 13S. Katsura, A. Yamaguchi, H. Inami, S. Matsuura, K. Hirano, and A. Mizuno, Elec- trophoresis 22, 289 (2001). 14D. R. Link, E. Grasland-Mongrain, A. Duri, F. Sarrazin, Z. Cheng, G. Cristobal, M. Marquez, and D. A. Weitz, Angew. Chem. Int. Ed. 45, 2556 (2006). 15J. Atencia and D. J. Beebe, Nature (London) 437, 648 (2005). 16J. Voldman, Annu. Rev. Biomed. Eng. 8, 425 (2006). 17T. B. Jones, Electromechanics of Particles (Cambridge University Press, New York, 1995). 8 

1811.03229

6

1811

Temperature dependence of side-jump spin Hall conductivity

In the conventional paradigm of the spin Hall effect, the side-jump conductivity due to electron-phonon scattering is regarded to be temperature independent. To the contrary, we draw the distinction that, while this side-jump conductivity is temperature independent in the classical equipartition regime where the longitudinal resistivity is linear in temperature, it is temperature dependent below the equipartition regime. The mechanism resulting in this temperature dependence differs from the familiar one of the longitudinal resistivity. In the concrete example of Pt, we show that the change of the spin Hall conductivity with temperature can be as high as 50%. Experimentally accessible high-purity Pt is proposed to be suitable for observing this prominent variation below 80 K.

cond-mat.mes-hall

cond-mat

Temperature dependence of side-jump spin Hall conductivity Cong Xiao,1, ∗ Yi Liu,2 Zhe Yuan,2, † Shengyuan A. Yang,3 and Qian Niu1 1Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA 2The Center for Advanced Quantum Studies and Department of Physics, Beijing Normal University, 100875 Beijing, China 3Research Laboratory for Quantum Materials, Singapore University of Technology and Design, Singapore 487372, Singapore In the conventional paradigm of the spin Hall effect, the side-jump conductivity due to electron- phonon scattering is regarded to be temperature independent. To the contrary, we draw the distinc- tion that, while this side-jump conductivity is temperature independent in the classical equipartition regime where the longitudinal resistivity is linear in temperature, it is temperature dependent below the equipartition regime. The mechanism resulting in this temperature dependence differs from the familiar one of the longitudinal resistivity. In the concrete example of Pt, we show that the change of the spin Hall conductivity with temperature can be as high as 50%. Experimentally accessible high-purity Pt is proposed to be suitable for observing this prominent variation below 80 K. The spin Hall effect refers to a transverse spin current in response to an external electric field [1]. In strongly spin-orbit-coupled electronic systems such as 4d and 5d transition metals [2 -- 10] and Weyl semimetals [11], the spin Hall conductivity due solely to the geometry of Bloch bands, the so-called spin Berry curvature, has attracted much attention. Besides, there is a scattering induced mechanism called side-jump, whose contribution to the spin Hall conductivity turns out to be of zeroth order of scattering time and independent of the density of a given type of impurities [1]. Furthermore, the side-jump spin Hall conductivity arising from the electron-phonon scattering is conventionally regarded to be temperature (T ) independent although the phonon density varies with T [12, 13]. In this work we draw the distinction that, while the electron-phonon scattering induced side-jump spin Hall conductivity is T -independent in the classical equipar- tition regime where the longitudinal resistivity ρ is lin- ear in T , it is T -dependent at temperatures below the equipartition regime. This character distinguishes side- jump from the geometric contribution, and provides a new mechanism for T -dependent spin Hall conductivi- ties in high-purity experimental samples. An intuitive picture is proposed for the T -dependence of the side- jump conductivity, which differs from ρ that is always T -dependent. Moreover, our first-principles calculation demonstrates a prominent T -variation of the spin Hall conductivity in experimentally accessible high-purity Pt below 80 K. We consider strongly spin-orbit coupled multiband sys- tems. The Fermi energy and the interband-splitting around the Fermi level are assumed to be much larger than the room temperature, thus the thermal smearing of Fermi surface is negligible. Aiming to provide semi- quantitative and intuitive understanding, the electron- phonon scattering is approximated by a single-electron elastic process, which can be called the "quasi-static approximation". In calculating the resistivity resulting from phonon scattering, this approximation produces not only the correct low-T power law (ρ ∼ T 5 for three- dimensional isotropic single-Fermi-surface systems) [14] but also the values that are quantitatively comparable with experimental data [15]. When applied to the side- jump transport, the high-T and low-T asymptotic be- haviors are grasped in this approximation. Quantita- tive deviations appearing in the intermediate tempera- ture regime are not essential for the present purpose. The side-jump was originally proposed as the side-way shift in opposite transverse directions for the carriers with different spins, when they are scattered by spin-orbit ac- tive impurities [12, 16]. This picture works well in sys- tems with weak spin-orbit coupling [17 -- 19], where the spin-orbit-induced band splitting is smeared by disorder broadening [1]. Whereas in strongly spin-orbit-coupled Bloch bands of current interest, the side-jump contribu- tion arises microscopically from the scattering-induced band-off-diagonal elements of the out-of-equilibrium den- sity matrix [20 -- 23]. This corresponds to in the Boltz- mann transport formalism the dressing of Bloch states by interband virtual scattering processes involving off- shell states away from the Fermi surface [24]. Transport formalism involving off-shell states. -- In weakly disordered crystals perturbed by an weak external electric field E, the expectation value of an observable A (assumed to be a vector without loss of generality) reads (cid:88) (cid:104)A(cid:105) = Alfl (1) l in the Boltzmann transport formalism, where fl is the oc- cupation function of the carrier state marked by l = (η, k) with η the band index and k the crystal momentum, Al is the quantum mechanical average on state l. The car- rier state is the Bloch state dressed by interband virtual processes induced by both the electric field and scattering [24]. In the linear response and weak scattering regime, these two dressing effects are independent [24]: Al = A0 l + Abc l + Asj l , (2) 9 1 0 2 l u J 2 ] l l a h - s e m . t a m - d n o c [ 6 v 9 2 2 3 0 . 1 1 8 1 : v i X r a where (cid:17) (cid:16) Abc l e  EαΩA αβ (ηk) = β (3) 2 arises from the electric-field induced dressing, with (cid:88) η(cid:48)(cid:48)(cid:54)=η vηη(cid:48)(cid:48) α (k) Aη(cid:48)(cid:48)η (cid:0)
ηk −
η(cid:48)(cid:48)k β (cid:1)2 (k) , (4) αβ (ηk) = −22 Im ΩA (cid:17) and(cid:16) Asj l β η(cid:48)k(cid:48) = −2π (cid:88) Wkk(cid:48)δ(cid:0)
ηk −
η(cid:48)k(cid:48)(cid:1)  (cid:88) − (cid:88)
η(cid:48)k(cid:48) −
η(cid:48)(cid:48)k(cid:48) η(cid:48)(cid:48)(cid:54)=η(cid:48) (cid:104)uη(cid:48)(cid:48)kuη(cid:48)k(cid:48)(cid:105)(cid:104)uη(cid:48)k(cid:48)uηk(cid:105)Aηη(cid:48)(cid:48) × Im β
ηk −
η(cid:48)(cid:48)k η(cid:48)(cid:48)(cid:54)=η (cid:104)uηkuη(cid:48)k(cid:48)(cid:105)(cid:104)uη(cid:48)(cid:48)k(cid:48)uηk(cid:105)Aη(cid:48)η(cid:48)(cid:48) β (k) (cid:0)k(cid:48)(cid:1)  . (5) originates from the scattering-induced dressing. Equa- tion (5) is diagrammatically represented in Fig. 1. The summation over repeated spatial indices α, β is implied hereafter. Here Aη(cid:48)(cid:48)η (k) ≡ (cid:104)uη(cid:48)(cid:48)kAβuηk(cid:105) with uηk(cid:105) the periodic part of the Bloch state. For impurities , with ni the impurity density and V o Wkk(cid:48) = ni kk(cid:48) the plane-wave part of the matrix element of the impurity potential. For electron-phonon scattering (cid:12)(cid:12)V o kk(cid:48)(cid:12)(cid:12)2 β (cid:12)(cid:12)U o kk(cid:48)(cid:12)(cid:12)2 Wkk(cid:48) = 2Nq V , (6) where U o k(cid:48)k is the plane-wave part of the electron-phonon matrix element, Nq is the Bose occupation function of phonons (q is the wave-vector of phonons with energy ωq), V is the volume (area in two-dimension) of the system, and the factor 2 accounts for the absorption and emission of phonons. When calculating the electric current A = ev, ΩA αβ and vsj l are the Berry curvature and "side-jump velocity" [22, 24, 25], respectively. When calculating the spin current A = j, ΩA αβ is the so-called spin Berry curvature [11], whereas Asj l provides the spin-current counterpart of the side-jump velocity [24, 26]. The occupation function of the carrier states is de- composed, around the Fermi distribution f 0 l , into fl = f 0 l + g2s Its out-of-equilibrium part satisfies the linearized steady-state Boltzmann equations eE · l ∂f 0 v0 l /∂
l = −(cid:80) ll(cid:48)(cid:0)g2s l + ga l . l(cid:48) w2s l(cid:48) (cid:1) and (cid:88) l − g2s eE · vsj l ∂f 0 l /∂
l = w2s ll(cid:48) (ga l − ga l(cid:48)) (7) l(cid:48) Wkk(cid:48)(cid:12)(cid:12)(cid:104)ulul(cid:48)(cid:105)(cid:12)(cid:12)2 tering rate, v0 in the presence of weak scalar disorder [22]. w2s ll(cid:48) = 2π δ (
l −
l(cid:48)) is the lowest-Born-order scat- l is the usual band velocity. FIG. 1. Graphical representation of Eq. (5), where the off- shell states away from the Fermi surface are marked by red arrows. Wkk(cid:48) is represented by the disorder line connected with two interaction vertexes. (cid:17) (cid:16) (cid:88) SH = (cid:80) l Collecting the above ingredients, the spin Hall current is [27, 28] jSH = jbc SH + jsj SH + jad SH. The first two terms jbc SH = jbc l l , and jsj f 0 SH = jsj l g2s l (8) (cid:17) (cid:88) (cid:16) l l j0 l ga l AH and jad l . Whereas jad arise from off-shell-states induced corrections to the semi- classical value of j0 incorpo- rates the nonequilibrium occupation function modified by off-shell states, since ga l appears as a response to the generation term proportional to the "side-jump velocity" vsj [25]. In calculating the anomalous Hall (AH) current l [21] A = ev, jsj AH are related to the transverse (side-way) and longitudinal components of vsj l , respec- tively [25]. Thereby their sum is also often referred to as the side-jump contribution in the literature on the anomalous Hall effect [21, 22, 25]. Given this conven- tion, we also include the jad SH contribution, although jad AH has nothing to do with the original concept of side-jump, and the microscopic theory [20, 23] indeed shows that ga l is not related to the interband elements of the out-of- equilibrium density matrix. In fact, in two-dimensional nonmagnetic models for the spin Hall effect, such as the two-dimensional electronic systems with Rashba, cubic Rashba and Dresselhaus spin-orbit couplings [29 -- 34], the spin current operator (for out-of-plane spin component) has only interband matrix elements, i.e., j0 l = 0, thus jad SH does not appear at all [27, 28]. Phonon-induced T-dependence of spin Hall conductiv- ity. -- In order to show the T -dependence of the phonon- induced side-jump spin Hall conductivity, we prove that its values in the low-T and high-T limits can be different. In the low-T limit, Wkk(cid:48) for phonon scattering is highly peaked at vanishing scattering angle, and the on-shell scattering can only be the intraband transition, hence in Eq. (5) η(cid:48) = η, and k(cid:48) is very close to k. We then ex- pand the integrand up to the first order of(cid:0)k(cid:48) − k(cid:1), get- (cid:16) (cid:17) = (cid:80) ting Asj l β k(cid:48) w2s l(cid:48)lΩA αβ (ηk)(cid:0)k(cid:48) − k(cid:1) α, with w2s l(cid:48)l = 𝐴𝛽𝜂′𝜂′′η′𝐤′η𝐤η′′𝐤′η𝐤𝐴𝛽𝜂′′𝜂′η′′𝐤′η′𝐤′η𝐤η𝐤𝐴𝛽𝜂′′𝜂η′′𝐤η𝐤η𝐤η𝐤𝐴𝛽𝜂𝜂′′η′′𝐤η′𝐤′η′𝐤′η𝐤k(cid:48) w2s curvature) yields ga scattering-angle limit. Concurrently, vsj 2π Wkk(cid:48)δ(cid:0)
ηk −
ηk(cid:48)(cid:1) the scattering rate in the small- ×(cid:0)k(cid:48) − k(cid:1) (Ω is the vector form of the ordinary Berry has [28](cid:16) (cid:17) and(cid:16) l =(cid:80) l = eE·(cid:2)k × Ω (l)(cid:3) ∂f 0 (cid:88) (cid:17) (cid:16) E ·(cid:2)k × Ω (l)(cid:3) ∂f 0 αβ (l) E · v0 l /∂
l. Thus one (cid:88) l(cid:48)lΩ (l) l /∂
l, kαΩj l ∂f 0 (cid:17) jsj SH (10) = e = e (9) β l l /∂
l. jad SH j0 l β β l Thus σSH = (jSH)x /Ey is a T -independent constant in the low-T limit. It is clear that this constant equals that contributed by scalar-impurities in the long-range limit, whose Wkk(cid:48) is also highly concentrated around vanishing scattering angle. than kBT indicating Wkk(cid:48) = 2kBT V−1(cid:12)(cid:12)U o In the high-T limit, the phonon energy is much smaller /ωq, l ∼ T 0, and l ∼ T , and ga l ∼ T −1, jsj (cid:12)(cid:12)2 k(cid:48)k then we have g2s consequently, ρ ∼ T, σSH ∼ T 0. (11) Accordingly, in practice the high-T limit is identified as the equipartition regime with linear-in-T resistivity [14]. This regime is usually marked qualitatively by T > TD in textbooks, with TD the Debye temperature, but can extend practically to about T > TD/3 in Pt, Cu and Au [15, 35], and to about T > TD/5 in Al [35]. Besides, it is apparent that the T -independent σSH in the equipartition regime can be different from that in the low-T limit. k(cid:48)k (cid:12)(cid:12)2 troduced as [36, 37] λ2 = 2V−1(cid:12)(cid:12)U o To acquire a more transparent picture, we assume any large-angle electron-phonon scattering can occur via normal processes, and take the approximation of the deformation-potential electron-acoustic phonon coupling, for which a electron-phonon coupling constant can be in- /ωq, hence arriv- ing at Wkk(cid:48) = λ2kBT in the high-T regime. This Wkk(cid:48) is uniformly distributed on the Fermi surface, just simi- lar to Wkk(cid:48) = niV 2 i contributed by randomly distributed zero-range scalar impurities, with Vi the strength. There- fore, we infer that the σSH due to phonons in the high-T equipartition regime takes the same value as that due to zero-range scalar impurities. This speculation can be verified by noticing that g2s,ep niV 2 i , Asj = ga,ei , l where the superscripts "ep" and "ei" mean the contribu- tions due to electron-phonon scattering and zero-range scalar impurities, respectively. λ2kBT = g2s,ei i and ga,ep l /niV 2 /λ2kBT = (cid:17)ep (cid:17)ei (cid:16) (cid:16) Asj l l l l According to the above results, σSH induced by electron-acoustic phonon scattering is T -dependent pro- vided that the σSH induced by scalar impurity scatter- ing in the long-range and zero-range limits are different. 3 This unexpected relation in turn provides a qualitative picture for comprehending the T -dependence of phonon- induced side-jump, by analogy with the recently revealed sensitivity of the side-jump conductivity to the scatter- ing range of impurities [38]. The accessible phase-space of the electron-phonon scattering changes with temper- ature, thus implies a T -dependent average momentum transfer, i.e., effective range, of this scattering. Note that this mechanism differs from that for the T - dependent ρ. To directly see this point, one need just consider the fact that in the equipartition regime σSH ∼ T 0 although ρ ∼ T is still T -dependent. The above revealed relation facilitates judging whether a model system has a T -dependent phonon-induced side- jump conductivity based on the familiar knowledge about the impurity-induced side-jump. There are models which possess different side-jump conductivities induced by scalar impurity-scattering in the long-range and zero- range limits, such as the Luttinger model describing p- type semiconductor [39] and the k-cubic Rashba model for the two-dimensional heavy-hole gas in confined quan- tum wells [30]. In these systems the phonon-induced side- jump conductivities are thus T -dependent. In the k-cubic Rashba model [33, 34], the Hamiltonian reads H = 2k2 2m + i αR 2 (cid:16) (cid:17) σ+k3− − σ−k3 + , (12) 1 where k = k (cos φ, sin φ) is the two-dimensional wave- vector, k± = kx ± iky, σ's are Pauli matrices with σ± = σx ± iσy, αR is the spin-orbit coupling coefficient that can be tuned to very large values by the gate voltage 2 {σz, vx} [34]. The spin current operator [31, 32] x = 3 SH + jsj has only interband components, hence jSH = jbc SH. Letting η = ± labels the two Rashba bands, the spin Berry curvature is Ωj 4mαRk3 , thus σbc SH = x /Ey = 9e2 [30, 31], with kη the Fermi wave-number of band η. The side-jump spin Hall conductivity due to electron-phonon scattering in the (cid:16) low-T limit is given by Eq. (9) as (cid:80) yx (ηk) = −η 93 sin2 φ η ηk−1 (cid:17) (cid:0)jbc 16πmαR (cid:1) SH η 2 1 4 σbc SH. (13) σsj SH = When mαR/2 (cid:28) 1/ In the high-T regime, we have jsj jsj SH √ x /Ey = πn [31], one has σsj SH = 9e/32π. l = 0 [28], leading to σsj SH = 0. (14) Since the side-jump conductivities in the high-T and low- T limits are different, there must be a crossover in the intermediate regime resulting in the T -dependent behav- ior. Temperature -- dependent spin Hall conductivity in pure Platinum. -- To show the applicability of our theoreti- cal ideas in real materials, we perform a first-principles calculation to the spin Hall conductivity of pure Pt in the range 20 -- 300 K. The minimal interband split- ting around the Fermi level of Pt is much larger than 300 K [9], thus the spin Berry-curvature contribution should be T -independent up to 300 K. The tempera- ture is modeled by populating the calculated phonon spectra of Pt into a large supercell with its length L along fcc [111] and 5 × 5 unit cells in the lateral di- mensions [15, 40]. Then the transport calculation is car- ried out using the above disordered (finite-temperature) supercell sandwiched by two perfectly crystalline (zero- temperature) Pt electrodes. The scattering matrix is obtained using the so-called "wave function matching" technique within the Landauer-Buttiker formalism [40]. The calculated total resistance of the scattering geome- try is found to be linearly dependent on L following the Ohm's law. By varying L in the range of 5 -- 60 nm, we extract the resistivity at every temperature using a linear least squares fitting for the calculated resistances. For each L, at least 10 random configurations have been considered to ensure both average value and standard deviation well converged with respect to the number of configurations. The calculated resistivity ρ is plotted in Fig. 2(a) as a function of temperature. The spin-Hall angle ΘSH is computed by examining the ratio of trans- verse spin current density and longitudinal charge current density [10]. At every temperature, we use more than 20 random configurations, each of which contains 60 nm- long disordered Pt. Then the spin Hall conductivity is obtained as σSH = (/e)ΘSH/ρ, shown in Fig. 2(b). 4 low the equipartition regime, in agreement with our the- oretical prediction. Finally, we discuss the possibility of observing the pre- dicted effect in experiments. In high-purity metals, the electron-electron scattering dominates over the electron- phonon scattering in determining transport behaviors at very low temperature. To observe our prediction, lower characteristic temperature Tt marking the crossover from the electron-electron dominated regime to the electron- phonon dominated one is required, such that the inter- mediate range from Tt to the high-T equipartition regime is wide enough. In experimentally accessible high-purity Pt samples with residual resistivity as small as 10−3 -- 10−2 µΩ cm [41, 42], Tt can be as low as 10K, and at T = 20 K the phonon-induced ρ is nearly one order of magnitude larger than that contributed by the electron- electron scattering and the residual resistivity [42]. Be- cause in high-purity Pt, the T -linear scaling of ρ emerges at T (cid:38) 80 K [15], the suitable range for observing the first-principles predicted T -dependence of σSH [Fig. 2(b)] is 20 K (cid:46) T (cid:46) 80 K. Very recently experimentalists have been developing new techniques, with which the spin cur- rent is generated and detected in a single transition-metal sample, thus avoiding all the complications associated with the interfaces and shunting effect [7, 8]. The pre- dicted effect is expected to be observed as the quality of Pt samples in such measurements is improved. We thank Ming Xie, Tianlei Chai and Haodi Liu for helpful discussions. Q.N. is supported by DOE (DE- FG03-02ER45958, Division of Materials Science and En- gineering) on the transport formulation in this work. C.X. is supported by NSF (EFMA-1641101) and Welch Foundation (F-1255). S.A.Y. is supported by Singapore Ministry of Education AcRF Tier 2 (MOE2017-T2-2- 108). Y.L. and Z.Y are supported by the National Natu- ral Science Foundation of China (Grants No. 61604013, No. 61774018, and No. 11734004), the Recruitment Pro- gram of Global Youth Experts, and the Fundamental Re- search Funds for the Central Universities (Grants No. 2016NT10 and No. 2018EYT03). FIG. 2. Calculated longitudinal resistivity ρ (a) and spin Hall conductivity σSH (b) of pure Pt as a function of temperature. The black dashed line in panel (a) illustrates the linear de- pendence. For T (cid:38) 80 K, a linear-in-T ρ is obtained, and σSH is approximately a constant of 1.6 × 105 /e (Ω m)−1 [10]. Below the equipartition regime, ρ deviates from the lin- ear T -dependence [illustrated by the black dashed line in Fig. 2(a)], concurrently the calculated σSH increases with decreasing temperature. At T = 20 K, σSH reaches 2.3 × 105 /e (Ω m)−1. Compared to the value at 80 K, the T -variation of σSH is as large as 50%. This T - dependence just begins when the temperature drops be- ∗ [email protected] † [email protected] [1] J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Rev. Mod. Phys. 87, 1213 (2015). [2] A. Hoffmann, IEEE Trans. Magn. 49, 5172 (2013). [3] T. Seki, Y. Hasegawa, S. Mitani, S. Takahashi, H. Ima- mura, S. Maekawa, J. Nitta, and K. Takanashi, Nat. Mater. 7, 125 (2008). [4] O. Mosendz, J. E. Pearson, F. Y. Fradin, G. E.W. Bauer, S. D. Bader, and A. Hoffmann, Phys. Rev. Lett. 104, 046601 (2010). [5] L. Liu, C.-F. Pai, Y. Li, H.W. Tseng, D. C. Ralph, and R. A. Buhrman, Science 336, 555 (2012). [6] E. Sagasta, Y. Omori, M. Isasa, M. Gradhand, L. E. 0.1110ρ (µΩ cm)20305070100200300Temperature (K)12σSH (105 h_/e Ω-1m-1)(a)(b)5 Hueso, Y. Niimi, Y. C. Otani, and F. Casanova, Phys. Rev. B 94, 060412(R) (2016). [7] C. Stamm, C. Murer, M. Berritta, J. Feng, M. Gabureac, P. M. Oppeneer, and P. Gambardella, Phys. Rev. Lett. 119, 087203 (2017). [8] C. Chen, D. Tian, H. Zhou, D. Hou, and X. Jin, Phys. Rev. Lett. 122, 016804 (2019). [9] T. Tanaka, H. Kontani, M. Naito, T. Naito, D. S. Hi- rashima, K. Yamada, and J. Inoue, Phys. Rev. B 77, 165117 (2008). [10] L. Wang, R. J. H. Wesselink, Y. Liu, Z. Yuan, K. Xia, and P. J. Kelly, Phys. Rev. Lett. 116, 196602 (2016). [11] Y. Sun, Y. Zhang, C. Felser, and B. Yan, Phys. Rev. Lett. 117, 146403 (2016). [12] L. Berger, Phys. Rev. B 2, 4559 (1970). [13] A. Crepieux and P. Bruno, Phys. Rev. B 64, 014416 [27] For our purpose of showing the T -dependence of the spin Hall conductivity induced by electron-phonon scattering, the results for jsj SH are sufficient. For complete- ness, we present in Ref. [28] the other spin Hall contribu- tion of order (Wkk(cid:48) )0 from the antisymmetric part of the third-Born-order scattering rate. This contribution also vanishes when j0 SH and jad l = 0. [28] See Supplemental Material for the transport formalism for the spin Hall conductivities of order (Wkk(cid:48) )0 and cal- culation details of the spin Hall conductivity in the k- cubic Rashba model, as well as discussions on the low- T limit in the presence of both electron-phonon and electron-impurity scattering. [29] O. V. Dimitrova, Phys. Rev. B 71, 245327 (2005). [30] S. Y. Liu and X. L. Lei, Phys. Rev. B 72, 155314 (2005). [31] J. Schliemann and D. Loss, Phys. Rev. B 71, 085308 (2001). (2005). [14] J. M. Ziman, Principles of the Theory of Solids (Cam- [32] B. A. Bernevig and S. C. Zhang, Phys. Rev. Lett. 95, bridge University Press, Cambridge, 1972). 016801 (2005). [15] Y. Liu, Z. Yuan, R. J. H. Wesselink, A. A. Starikov, M. van Schilfgaarde, and P. J. Kelly, Phys. Rev. B 91, 220405(R) (2015). [16] S. K. Lyo and T. Holstein, Phys. Rev. Lett. 29, 423 [33] R. Moriya, K. Sawano, Y. Hoshi, S. Masubuchi, Y. Shi- raki, A. Wild, C. Neumann, G. Abstreiter, D. Bougeard, T. Koga, and T. Machida, Phys. Rev. Lett. 113, 086601 (2014). (1972). [17] H.-A. Engel, B. I. Halperin, and E. I. Rashba, Phys. Rev. Lett. 95, 166605 (2005). [18] W. K. Tse and S. Das Sarma, Phys. Rev. Lett. 96, 056601 (2006). [19] E. M. Hankiewicz and G. Vignale, Phys. Rev. B 73, 115339 (2006). [20] W. Kohn and J. M. Luttinger, Phys. Rev. 108, 590 (1957); J. M. Luttinger, Phys. Rev. 112, 739 (1958). [21] N. A. Sinitsyn, J. Phys.: Condens. Matter 20, 023201 (2008). [22] N. A. Sinitsyn, Q. Niu, and A. H. MacDonald, Phys. Rev. B 73, 075318 (2006). [23] C. Xiao, Z. Z. Du, and Q. Niu, arXiv: 1907.00577 [24] C. Xiao and Q. Niu, Phys. Rev. B 96, 045428 (2017). [25] N. A. Sinitsyn, A. H. MacDonald, T. Jungwirth, V. K. Dugaev, and J. Sinova, Phys. Rev. B 75, 045315 (2007). [26] C. Xiao, Front. Phys. 13, 137202 (2018). [34] H. Liu, E. Marcellina, A. R. Hamilton, and D. Culcer, Phys. Rev. Lett. 121, 087701 (2018). [35] J. M. Ziman, Electrons and Phonons (Clarendon, Oxford, 1960). [36] A. A. Abrikosov, L. P. Gor'kov, and I. E. Dzyaloshin- skii, Quantum Field Theoretical Methods in Statistical Physics (Pergamon Press, New York, 1965). [37] J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986). [38] I. A. Ado, I. A. Dmitriev, P. M. Ostrovsky, and M. Titov, Phys. Rev. B 96, 235148 (2017). [39] S. Murakami, Phys. Rev. B 69, 241202(R) (2004). [40] Anton A. Starikov, Yi Liu, Zhe Yuan and Paul J. Kelly, Phys. Rev. B 97, 214415 (2018). [41] G. K. White and S. B. Woods, Phil. Trans. Roy. Soc. (London) A251, 273 (1958). [42] R. H. Freeman, F. J. Blatt, and J. Bass, Phys. Kondens. Materie 9, 271 (1969). 

1106.3697

1

1106

Electronic charge and spin density distribution in a quantum ring with spin-orbit and Coulomb interactions

Charge and spin density distributions are studied within a nano-ring structure endowed with Rashba and Dresselhaus spin orbit coupling (SOI). For a small number of interacting electrons, in the presence of an external magnetic field, the energy spectrum of the system is calculated through an exact numerical diagonalization procedure. The eigenstates thus determined are used to estimate the charge and spin densities around the ring. We find that when more than two electrons are considered, the charge-density deformations induced by SOI are dramatically flattened by the Coulomb repulsion, while the spin density ones are amplified.

cond-mat.mes-hall

cond-mat

Electronic charge and spin density distribution in a quantum ring with spin-orbit and Coulomb interactions Csaba Daday,1, 2 Andrei Manolescu,1 D. C. Marinescu,3 and Vidar Gudmundsson2 1School of Science and Engineering, Reykjavik University, Menntavegur 1, IS-101 Reykjavik, Iceland 2Science Institute, University of Iceland, Dunhaga 3, IS-107 Reykjavik, Iceland 3Department of Physics and Astronomy, Clemson University, Clemson, SC 29621, USA Charge and spin density distributions are studied within a nano-ring structure endowed with Rashba and Dresselhaus spin orbit coupling (SOI). For a small number of interacting electrons, in the presence of an external magnetic field, the energy spectrum of the system is calculated through an exact numerical diagonalization procedure. The eigenstates thus determined are used to estimate the charge and spin densities around the ring. We find that when more than two electrons are considered, the charge-density deformations induced by SOI are dramatically flattened by the Coulomb repulsion, while the spin density ones are amplified. PACS numbers: 71.70.Ej, 73.21.Hb, 71.45.Lr I. INTRODUCTION The possibility of controlling the flow of the electron spins in semiconductor structures by external electric means through spin-orbit interaction (SOI) has domi- nated the recent past of spintronics research. This fun- damental principle, first explored in the Datta-Das spin transistor configuration,1 has been guiding a sustained effort in understanding all the phenomenological impli- cations of this interactions on systems of electrons. The coupling between spin and orbital motion results either from the two-dimensional confinement (Rashba)2 or from the inversion asymmetry of the bulk crystal structure (Dresselhaus).3 The usual expression of the the spin-orbit Hamiltonian HSO retains only the linear terms in the electron momentum p = (px, py) and is given by HSO = α  (σxpy − σypx) + β  (σxpx − σypy) . (1) The Rashba and Dresselhaus coupling constants are α and β, respectively, while σx,y,z are the Pauli matrices. In general, the two interactions are simultaneously present and often have comparable strengths. While α, the cou- pling constant of the Rashba interaction, can be modi- fied by external electric fields induced by external gates, the strength of the Dresselhaus SOI, β, is fixed by the crystal structure and by the thickness of the quasi two- dimensional electron system.4,5 In many situations of in- terest, an additional energy scale is introduced by the Zeeman interaction of the electron spin with an external magnetic field, proportional to the effective gyromagnetic factor, g∗, which depends on the material energy-band structure. While g∗ = −0.44 is very small in GaAs, it can be more that 100 times larger in InSb. The interplay between the two types of SOI, which have competing effects on the precession of the electron spin as they rotate it in opposite directions, and the Zeeman splitting, which minimizes the energy by align- ing the spin parallel to the external field, determines the ground state polarization of the electron system and the characteristics of spin transport. The investigation of such problems in mesoscopic rings has been pursued intensively by several authors.6 -- 10 In particular, it was noticed that, in the absence of the Coulomb interac- tion among the electrons, the interference between the Rashba and Dresselhaus precessions relative to the or- bital motion, leads to the creation of an inhomogeneous spin and charge distribution around the ring.8 The charge inhomogeneity created in this situation has a symmetric structure with two maxima and two minima around the ring and will be called here a charge-density deformation (CDD). The effect of the Coulomb interaction on this type of charge distribution has been considered for two electrons. It was obtained that, on account of the elec- trostatic repulsion, the two electrons become even more localized in the potential minima associated to the CDD, leading to an amplitude increase.9,10 In this work we obtain an estimate of the effect of the Coulomb interaction on the charge and spin distribution associated with N = 2, 3 and 4 electrons in a ring with SOI coupling by an exact diagonalization procedure that uses the configuration interaction method. Our results indicate that when the number of electrons increases, the mutual repulsion leads to more uniform charge distribu- tion around the ring, generating a dramatically flattened CDD. In contrast, the spin-density deformation (SDD) is amplified by the Coulomb effects. This can be explained by the appearance of a stronger repulsion between same spin electrons, leading to more favorable spin orienta- tions. II. THE RING MODEL The system of interest in our problem is a two- dimensional quantum ring of exterior and interior radii, Rext and Rint respectively. The ring is placed in a per- pendicular magnetic field B associated in the symmetric gauge with a vector potential A = B/2(−y, x, 0). The single-particle Hamiltonian of an electron of momentum 2 In the same basis, the Zeeman Hamiltonian is simply diagonal in the spatial coordinates, hkjσHZ k′j′σ′i = 1 2 T tBγ(σz)σσ′ δkk′ δjj ′ , (4) where γ = g∗m∗/(2me) is the ratio between the Zeeman gap and the cyclotron energy, me being the free electron mass. For the spin-orbit Hamiltonian we obtain: hkjσHSOk′j′σ′i = 1 2 T tα"tB rk 4Rext (σj r)σσ′ δkk′ δjj ′ + it1/2 ϕ (σj r + σj+1 r 2 )σσ′ δkk′ δjj ′ +1 − it1/2 r (σj ϕ)σσ′ δkk′ +1δjj ′# + T tβXk,j "σj r → (σj ϕ)∗ and σj ϕ → −(σj r)∗# + h.c. , (5) where tα = α/(RextT ) and tβ = β/(RextT ) are the two types of spin-orbit relative energies, while σr(ϕ) = σx cos ϕ + σy sin ϕ and σϕ(ϕ) = −σx sin ϕ + σy cos ϕ are the radial and angular Pauli matrices, respectively. We used the slightly shorter notations σj r = σr(ϕj ) and σj ϕ = σϕ(ϕj) for the matrices at the particular angles on our lattice. The Rashba spin-orbit terms are all in- cluded in the first square bracket. The Dresselhaus terms are very similar to the Rashba ones, being given by the substitutions indicated in the second square bracket. The single particle states of the noninteracting Hamil- tonian (2), Hψa = ǫaψa, are computed as eigenvalues and eigenvectors of the matrices (3)-(5), ψa(rk, ϕj) = Pσ Ψa,σ(k, j)σi. where Ψa,σ(k, j) are c−numbers. In the basis provided by {ψa} the interacting many- body Hamiltonian is written in the second quantization as ǫac† aca + H =Xa 1 2 Xabcd Vabcdc† ac† bcdcc , (6) where c† a and ca are the creation and annihilation opera- tors on the single-particle state a. The matrix elements of the Coulomb potential V (r − r′) = e2/(κr − r′), κ be- ing the dielectric constant of the material, are in general give by 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 t x e R / y -1.0 -1.0 -0.5 0.0 x/Rext 0.5 1.0 FIG. 1: (Color online) The discretized ring with Rint = 0.8Rext, and 10 radial × 50 angular sites. The sites are shown with circular points. The thin dotted lines connection sites are for guiding the eye. p = −i∇ + eA and effective mass m∗ is written as the sum of an orbital term HO = p2/2m∗, a Zeeman contri- bution HZ = (1/2)g∗µBBσz and the spin-orbit coupling given in Eq. (1). The ensuing expression, H = HO + HZ + HSO , (2) is discretized in a standard manner6,7,11 on a grid12 defined by Nr radial and Nϕ angular sites, as shown in Fig. 1. The radial coordinate of each site is rk = Rext − (k − 1)δr, with k = 1, 2, ..., Nr, while δr = (Rext − Rint)/(Nr − 1) is the distance between adja- cent sites with the same angle. Similarly, the angular coordinate is ϕj = (j − 1)δϕ, where j = 1, 2, ..., Nϕ and δϕ = 2π/Nϕ is the angle between consecutive sites with the same radius. The Hilbert space is spanned by the ket-vectors kjσi, where the first integer, k, stands for the radial coordinate, the second one, j, for the angular coordinate, and σ = ±1 denotes the spin projection in the z direction. In this basis {kjσi}, the matrix elements of the orbital Hamiltonian are given by: hkjσHOk′j′σ′i = T δσσ′("tϕ + tr + −(cid:20)tϕ + tB i 1 2 t2 B(cid:18) rk 4Rext(cid:19)2# δkk′ δjj ′ 4δϕ(cid:21) δkk′ δjj ′ +1 + trδkk′ +1δjj ′) + h.c. . (3) Vabcd = hψa(r)ψb(r′)V(r − r′)ψc(r)ψd(r′)i . (7) In the present discrete model the double scalar product is in fact a double summation over all the lattice sites and spin labels T = 2/(2m∗R2 ext) is the energy unit, while Rext is the length unit. In T units, we obtain tϕ = [Rext/(rkδϕ)]2 the angular hopping energy, tr = (Rext/δr)2 the radial hopping energy, and tB = eB/(m∗T ) the magnetic cy- clotron energy. (h.c. denotes the Hermitian conjugate.) Vabcd = T tC Xkjσ k′j ′σ′ Ψ∗ a,σ(kj)Ψ∗ b,σ′ (k′j′) Rext rjk − rj ′k′ (8) × Ψc,σ(kj)Ψd,σ′(k′j′) . (a) (b) (c) T / y g r e n E 174 172 170 168 166 164 1 z S 0 -1 2 1 0 0 1 2 3 4 3 GS ES1 ES2 ES3 6 7 8 9 10 5 tB (9) 0 0 0 1 x c ∆ The new energy parameter introduced by the Coulomb repulsion is tC = e2/(κRextT ). In the above summation over the sites, the contact terms (k = k′, j = j′) are avoided, as their contribution vanishes in the continuous limit. The many-body states Φµ are found by solving the eigenvalue problem for the Hamiltonian (6), HΦµ = EµΦµ . A potential solution of the equation is written in the con- figuration interaction representation13 -- 15 as a linear com- bination of the non-interacting system eigenstates (Slater determinants), cµααi , Φµ =Xα 1 , iα 2 , ..., iα Ki} where iα havePa iα with {αi = iα a = 0, 1 is the occu- pation number of the single-particle state ψa and K is the number of single-particle states considered. The oc- cupation numbers iα K are listed in the increasing energy order, so ǫK is the highest energy of the single-particle state included in the many-body basis. For any αi we a = N , which is the number of electrons in the ring. It is straightforward to derive the matrix elements of Hαα′ using the action of the creation and annihila- tion operators on the many-body basis. In practice Eq. (9) is convergent with K for a sufficiently small number of electrons, and sufficiently small ratio of Coulomb to confinement energy, tC . This procedure, also known as "exact diagonalization", does not rely on any mean field description of the Coulomb effects, like Hartree, Hartree- Fock, or DFT.16 To be able to carry the numerical calculations in a rea- sonable amount of time, we choose a small ring of radii Rext = 50 nm and Rint = 0.8Rext, containing N ≤ 4 electrons. The discretization grid has 10 radial and 50 an- gular points (500 sites), as shown in Fig. 1. Two common semiconductor materials used in the experimental spin- tronics are used for the selection of the material constants needed: InAs with m∗ = 0.023me, g∗ = −14.9, κ = 14.6, and estimated (or possible) values for the spin-orbit in- teractions α ≈ 20 and β ≈ 3 meVnm; InSb with m∗ = 0.014me, g∗ = −51.6, κ = 17.9, and α ≈ 50, β ≈ 30 meVnm.4,5 The relative energies which we defined are: for InAs tα = 0.60, tβ = 0.09, tC = 2.9, γ = −0.17; for InSb tα = 0.92, tβ = 0.55, tC = 1.5, γ = −0.36. In our calculations we have considered material parameters somewhere in between these two sets: tα = 0.7, tβ = 0.3, tC = 2.2, γ = −0.2 III. RESULTS A. Single particle calculations In the absence of the SOI, (α = β = 0), the single particle Hamiltonian (2) shares its eigenstates ψa with FIG. 2: (Color online) (a) The lowest 12 energies of the single particle states vs. the magnetic energy tB. The solid (red) and the dashed (green) lines show the states with positive and negative parity, respectively. (b) The expected value of the spin projection along the z direction Sz, in units of /2, for the first four states on the energy scale. (c) The standard deviation ∆c of the charge distribution around the circle with radial site index k = 6, for the first four energy states: ground state (GS), first, second, and third exited states (ES1, ES2, ES3). The same association of line types with states is used in panel (b). the z components of the angular momentum Lz and spin Sz = σz/2. In the presence of only one type of SOI, ei- ther α 6= 0 or β 6= 0, the Hamiltonian commutes with the z component of the total angular momentum, Lz + Sz, which is conserved. When both α 6= 0 and β 6= 0, the angular momentum is no longer conserved. How- ever, ψa continue to be eigenstates of the parity operator P = Πσz, Π being the (three dimensional) spatial in- version operator. Indeed, the general Hamiltonian (2) commutes with P, which can be easily verified by using Πp = −pΠ and the commutation rules of the Pauli ma- trices. So in general Pψa = sψa, and thus the parity s = ±1 of any state a is conserved, i. e. it is indepen- dent on the magnetic field. In particular, when α = β and g∗ = 0, all states become parity-degenerate at any magnetic field.8,10,17 We identify the parity of the single particle states calculated on our discrete ring model by looking at the relation ψa,σ(k, j) = sσψa,σ(k, ¯)) where (k, j) and (k, ¯) are diametrically opposed sites, with an- gular coordinates ϕ¯ = ϕj + π. In Fig. 2 we show the single particle states energy for 0 < tB < 10 (units of T ), which corresponds to a mag- netic field strength between 0 and 1.32 Tesla. Further increment of the magnetic field requires an augmentation of the number of sites on the ring in order to maintain the discrete model as a reasonable approximation of a physi- cally continuous ring. At zero magnetic field all states are parity degenerate, which is just the ordinary spin degen- eracy. The degeneracy is in general lifted by a finite mag- netic field. There are, however, some particular values of tB where the degeneracy persist. This situation is repre- sented in Fig. 2(a) by all intersection points of two lines corresponding to the two possible parities. Such intersec- tions do not occur between states with the same parity. Due to the spin-orbit coupling, the orbital momentum of one state depends on the spin of the other state and vice versa, and, consequently, states of same parity do in fact interact and thus avoid intersections.10 Although in Fig. 2(a) many states represented by the same line type apparently cross each other, in reality there are al- ways tiny gaps between them, similar to those visible at tB ≈ 2 between the first and the second excited states or at tB ≈ 5.5 between the first, second, and third ex- cited states. The magnitude of these gaps depends on the g−factor, reducing in size for a smaller g∗ parameter. In Fig. 2(b) the evolution of the expected spin in the z direction, Sz = hψaσzψai/2 is presented for the first four states in the energy order. One can see how the spin flips for states avoiding the crossing, like those with negative parity at tB ≈ 2 (dashed and dotted lines in Fig. 2(b)). Only one type of SOI, either Rashba or Dresselhaus, is sufficient to avoid the crossing of states with the same parity, but in this case the charge and the spin densi- ties are uniform around the ring. When both SOI types are present the charge and spin densities become nonuni- form. This situation is equivalent with the presence of a potential with two maxima and two minima around the ring, having reflection symmetry relative to the axes y = x (or ϕ = π/4, corresponding to the crystal direction [110]) and y = −x (or ϕ = −π/4, corresponding to the crystal direction [1¯10]).8,10 The amplitude of the CDD is illustrated in Fig. 2(c) where the standard deviation (in the statistical sense) ∆c of the charge density cal- culated around one circle on the ring, close to the mean radius, with radial site index k = 6, is plotted for the low- est four energy states. The density deformation occurs on account of the two combined SOI types which lead to spin interference and additional interaction between states with the same parity. Consequently, the ampli- tude of the CDD for a certain state is maximum at those magnetic fields where the parity degeneracy is lifted (the state avoids a crossing with another state of the same parity). In Fig. 2(c) this is clearly seen at tB ≈ 2, for the excited state. In this example the CDD in the ground state is very weak. The sharp peak at tB ≈ 0.4 indicates the existence of a narrow gap between the 4-th and 5-th energy states that avoid crossing. 4 T / y g r e n E 331 330 329 328 327 494 493 492 491 659 658 657 656 655 333 332 331 330 329 501 500 499 498 673 672 671 670 669 (a) N=2 tC=0 (c) N=3 tC=0 (e) N=4 tC=0 0 2 4 6 8 10 tB (b) N=2 tC=2.2 (d) N=3 tC=2.2 (f) N=4 tC=2.2 0 2 4 tB 6 8 10 FIG. 3: (Color online) Energy spectra of the first 12 states for N = 2, 3, and 4 electrons without Coulomb interaction, tC = 0 in panels (a),(c),(e), and with Coulomb interaction, tC = 2.2, in panels (b),(d),(f). The solid (red) and the dashed (green) lines show the states with positive and negative parity, respectively. B. Many particle calculations In the following considerations, we will include more than one electron. In Fig. 3 we compare the energy spec- tra for the first 12 states vs. the magnetic energy for N = 2, 3, and 4 electrons, without and with Coulomb interaction. Since the Coulomb interaction is invariant at spatial inversion (and independent on spin) the par- ity s is also conserved in the many-body states. Spectra drawn for tC = 0 and tC = 2.2 have similar features. The interacting spectrum presents a shift to higher ener- gies, on account of the additional Coulomb energy, and a slight increase of the gaps at high magnetic fields. More- over, the crossings and the anti-crossings (points where the crossings were avoided) of the energy levels have a tendency to shift slightly to higher magnetic fields. Sim- ilarly, the gap between the ground state and the excited states increases at high tB. The total spin Sz for each of the first three energy states is shown in Fig. 4. At zero magnetic field, for an even number of electrons, here N = 2 or N = 4, the ground state is non-degenerate and has total spin Sz = 0, i. e. the spin-up and spin-down states of individual electrons compensate. When the field is applied, the first spin flip in the interacting ground state, as well as the spin saturation, occur at lower magnetic fields than in the absence of the Coulomb repulsion. This is a result of the mixing of spin states with the same parity produced  z S 2 1 0 -1 -2 2 1 0 -1 3 2 1 0 -1 2 1 0 -1 GS ES1 ES2 (a) N=2 tC=0 0 2 4 6 8 10 0 2 4 6 8 10 (b) N=2 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 -2 2 1 0 (c) N=3 tC=0 0 2 4 6 8 10 0 2 4 6 8 10 (d) N=3 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 -1 3 2 1 0 (e) N=4 tC=0 0 2 4 6 8 10 0 2 4 6 8 10 tB -1 (f) N=4 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 tB FIG. 4: (Color online) The total spin projection in the z direction, in units of /2, for the many body states with N = 2, 3, 4 electrons. Without interaction, i. e. tC = 0 in panels (a),(c),(e), and with interaction, with tC = 2.2, in panels (b),(d),(f). Only the first three states are shown here, the ground state (GS), the 1-st excited states (ES1), and the 2-nd excited state (ES2). The magnetic energy tB varies between 0 and 10 and the lines showing the excited states are shifted to the right, for clarity. by the interaction. For N = 3 the ground state is spin (double) degenerate at zero field. In the presence of the Coulomb interaction, the total spin in the ground state and the higher state is reversed. Similar to the case of one electron (N = 1, Fig. 2), the charge deformation of each state is maximized for those magnetic fields where the state has an anti-crossing (or repulsion) with another state of the same parity. The charge deformation parameter ∆c is shown in Fig. 5. For N = 2 the amplitude of the CDD increases with the Coulomb interaction. There is a simple reason for that: the potential associated with the charge deformation has two minima diametrically opposite on the ring and each of the two electrons tends be localized in one of these min- ima. The mutual Coulomb repulsion fixes the electrons in those places better.9,10 The situation changes, however, for N > 2. The Coulomb forces spread the electrons differently, more or less uniformly, such that the charge deformation created by the SOI is drastically reduced. In other words, the associated potential is strongly screened even by one extra electron above N = 2. This effect can be clearly seen in Fig. 5, comparing panels (c) with (d) and (e) with (f). The vertical scale of panels (d) and (e) has been magnified three times, for visibility. 0 0 0 1 x c ∆ 3.0 2.0 1.0 0.0 3.0 2.0 1.0 0.0 3.0 2.0 1.0 0.0 (a) N=2 tC=0 GS ES1 ES2 0 2 4 6 8 10 0 2 4 6 8 10 (c) N=3 tC=0 0 2 4 6 8 10 0 2 4 6 8 10 (e) N=4 tC=0 0 2 4 6 8 10 0 2 4 6 8 10 tB 3.0 2.0 1.0 0.0 0.8 0.6 0.4 0.2 0.0 0.8 0.6 0.4 0.2 0.0 5 (b) N=2 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 (d) N=3 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 (f) N=4 tC=2.2 0 2 4 6 8 10 0 2 4 6 8 10 tB FIG. 5: (Color online) The standard deviation ∆c of the charge on the circle k = 6 around the ring, as a measure of the amplitude of the charge deformation. Shown are the results for the ground state (GS), the 1-st excited states (ES1), and the 2-nd excited state (ES2), for N = 2, 3, 4 electrons with- out without (tC = 0), and with interaction (tC = 2.2). The amplitude of the CDD's is strongly reduced by the Coulomb effects for N = 3, 4; notice the different scales used in the paired panels (c),(d) and (e),(f). The magnetic energy tB varies between 0 and 10 and the lines corresponding to the excited states are shifted to the right, for clarity. only generates the specific effective potential which de- termines the CDD. In the absence of SOI (α = β = 0), we checked that a similar screening effect occurs in the presence of a potential that induces a charge deformation comparable to that obtained with the SOI. It is, however, surprising that by adding only one extra electron such a drastic effect ensues. Next, we investigate the effect of the Coulomb inter- action on the spin distribution around the ring. The standard deviation of the spin density projected along the z direction, ∆z, is plotted in Fig. 6 where we show the results calculated as before for the circle correspond- ing to the sites with radial coordinate k = 6. The spin density deformation (SDD) is actually amplified by the Coulomb interaction for all N = 2, 3, 4. As the CDD's, the SDD's reach their maximum at the magnetic fields where level repulsion occurs and remains prominent even when the gaps are very small. The Coulomb enhance- ment is a result of the mixing of states with the same parity, but with different spin orientation produced by the Coulomb potential. Consequently the SDD's have in general a richer structure than the CDD's. In principle, the screening of the charge deformation is not particularly related to the spin-orbit effects. SOI Finally, in Fig. 7 we display an example of CDD and SDD, obtained for N = 4 interacting particles. The 6 charge and spin distributions corresponding to the sec- ond excited state and tB = 4.5 are illustrated in Figs. 5(f) and 6(f), respectively. The CDD is weak, but still it has four visible maxima. For two electrons the CDD has only two maxima which are along the directions x = y or x = −y, depending on the state and on the magnetic field, both with and without the Coulomb interaction. In particular, for a strictly one-dimensional ring model and N = 2, in the ground state, the maxima are always along the line x = −y,8,9 whereas for a two-dimensional model they can also be along x = y.10 But in general, for N > 2 electrons, screening effects may distribute the charge in more complicated configurations. Similar profiles with multiple local oscillation may be obtained for the spin density, eventually becoming spin-density waves around the ring. IV. CONCLUSIONS We calculated the many-body states of a system of N = 2, 3, and 4 interacting electrons located in a ring of finite width with Rashba and Dresselhaus spin-orbit cou- pling, in the presence of a magnetic field perpendicular on the surface of the ring. The Coulomb effects are fully in- cluded in the calculation via the "exact diagonalization" method. We obtained inhomogeneous charge densities, or CDD's, around the ring due to the combined effect of the two types of SOI. When the Coulomb interaction is included the charge deformation is amplified for N = 2, as also shown by other authors.9,10 For N > 2 we find that the CDD is dramatically flattened out in the pres- ence of the Coulomb interaction. We interpret the result as a screening effect. On the contrary, the spin inhomo- geneities, or SDD's, are amplified by Coulomb effects for all N > 1. 2 (a) N=2 tC=0 GS ES1 ES2 2 (b) N=2 tC=2.2 1 1 0 0 0 1 x z ∆ 0 2 1 0 2 1 0 0 2 4 6 8 10 0 2 4 6 8 10 (c) N=3 tC=0 0 2 0 2 4 6 8 10 0 2 4 6 8 10 (d) N=3 tC=2.2 1 0 2 4 6 8 10 0 2 4 6 8 10 (e) N=4 tC=0 0 2 0 2 4 6 8 10 0 2 4 6 8 10 (f) N=4 tC=2.2 1 0 2 4 6 8 10 0 2 4 6 8 10 tB 0 0 2 4 6 8 10 0 2 4 6 8 10 tB FIG. 6: (Color online) The standard deviation ∆z of the spin projection along the z direction on the circle k = 6 around the ring, as a measure of the amplitude of the spin-density wave. Like in the previous figures GS, ES1, and ES2 in the legend indicate the ground state, the 1-st excited states, and the 2-nd excited states, respectively. The results are shown for N = 2, 3, 4 electrons without (tC = 0), and with interaction (tC = 2.2). Unlike the CDD's, the SDD's are amplified by the Coulomb interactions for all N . The magnetic energy tB varies between 0 and 10 and the lines corresponding to the excited states are shifted to the right, for clarity. (a) (b) y t i s n e d e g r a h C -1.0 -0.5 0.0 x/Rext 0.5 1.0 1.0 0.5 y/Rext -1.0 0.0 -0.5 FIG. 7: (Color online) (a) The charge density for N = 4 electrons with interaction (tC = 2.2), in the second excited state, i. e. ES2 in Fig.5(f), with magnetic energy energy tB = 4.5. (b) The corresponding total spin distribution along the ring k = 6 where the standard deviation of the z component is calculated and shown in Fig.6(f). Acknowledgments This work was supported by the Icelandic Research Fund. Valuable discussions with Sigurdur Erlingsson, Gunnar Thorgilsson, and Marian Nit¸a are cordially ac- knowledged. 1 S. Datta and B. Das, Appl. Phys. Lett. 56 (1990). 2 Y. Bychkov and E. I. Rashba, JETP Lett. 39, 78 (1984). 3 G. Dresselhaus, Phys. Rev. 100, 580 (1955). 4 R. Winkler, Spin orbit coupling effects in two-dimensional electron and hole systems (Springer-Verlag Berlin, Heidel- berg, New York, 2003). 5 T. Ihn, Semiconductor nanostructures. Quantum states and electronic transport (Oxford University Press, 2010). 6 J. Splettstoesser, M. Governale, and U. Zulicke, Phys. Rev. B 68, 165341 (2003). 12 The first and function second f (x), generic f ′(x) [f (x + h) + f (x − h) − 2f (x)] /h2, h is considered sufficiently small. ≈ are [f (x + h) − f (x − h)] /h and 7 any as ≈ derivatives of approximated f ′′(x) respectively, where 13 P. Hawrylak and D. Pfannkuche, Phys. Rev. Lett. 70, 485 (1993). 14 N. T. T. Nguyen and F. M. Peeters, Phys. Rev. B 83, 7 S. Souma and B. K. Nikoli´c, Phys. Rev. B 70, 195346 075419 (2011). (2004). 8 J. S. Sheng and K. Chang, Phys. Rev. B 74, 235315 (2006). 9 Y. Liu, F. Cheng, X. J. Li, F. M. Peeters, and K. Chang, 15 N. T. T. Nguyen and F. M. Peeters, Phys. Rev. B 78, 045321 (2008). 16 D. Pfannkuche, V. Gudmundsson, and P. Maksym, Phys. Phys. Rev. B 82, 045312 (2010). Rev. B 47, 2244 (1993). 10 M. P. Nowak and B. Szafran, Phys. Rev. B 80, 195319 17 J. Schliemann, J. C. Egues, and D. Loss, Phys. Rev. Lett. (2009). 11 F. E. Meijer, A. F. Morpurgo, and T. M. Klapwijk, Phys. Rev. B 66, 033107 (2002). 90, 146801 (2003). 

1601.07524

2

1601

Surface plasmon polaritons in topological Weyl semimetals

We consider theoretically surface plasmon polaritons in Weyl semimetals. These materials contain pairs of band touching points - Weyl nodes - with a chiral topological charge, which induces an optical anisotropy and anomalous transport through the chiral anomaly. We show that these effects, which are not present in ordinary metals, have a direct fundamental manifestation in the surface plasmon dispersion. The retarded Weyl surface plasmon dispersion depends on the separation of the Weyl nodes in energy and momentum space. For Weyl semimetals with broken time-reversal symmetry, the distance between the nodes acts as an effective applied magnetic field in momentum space, and the Weyl surface plasmon polariton dispersion is strikingly similar to magnetoplasmons in ordinary metals. In particular, this implies the existence of nonreciprocal surface modes. In addition, we obtain the nonretarded Weyl magnetoplasmon modes, which acquire an additional longitudinal magnetic-field dependence. These predicted surface plasmon results are observable manifestations of the chiral anomaly in Weyl semimetals and might have technological applications.

cond-mat.mes-hall

cond-mat

a Surface plasmon polaritons in topological Weyl semimetals Johannes Hofmann1, 2 and Sankar Das Sarma1 1Condensed Matter Theory Center and Joint Quantum Institute, Department of Physics, University of Maryland, College Park, Maryland 20742-4111 USA 2T.C.M. Group, Cavendish Laboratory, University of Cambridge, Cambridge CB3 0HE, United Kingdom (Dated: June 15, 2016) We consider theoretically surface plasmon polaritons in Weyl semimetals. These materials contain pairs of band touching points – Weyl nodes – with a chiral topological charge, which induces an optical anisotropy and anomalous transport through the chiral anomaly. We show that these effects, which are not present in ordinary metals, have a direct fundamental manifestation in the surface plasmon dispersion. The retarded Weyl surface plasmon dispersion depends on the separation of the Weyl nodes in energy and momentum space. For Weyl semimetals with broken time-reversal symmetry, the distance between the nodes acts as an effective applied magnetic field in momentum space, and the Weyl surface plasmon polariton dispersion is strikingly similar to magnetoplasmons in ordinary metals. In addition, we obtain the nonretarded Weyl magnetoplasmon modes, which acquire an additional longitudinal magnetic-field dependence. These predicted surface plasmon results are observable manifestations of the chiral anomaly in Weyl semimetals and might have technological applications. In particular, this implies the existence of nonreciprocal surface modes. PACS numbers: 73.20.Mf, 78.68.+m, 71.20.Gj, 03.65.Vf Surface plasmon polaritons (SPPs) are collective elec- tromagnetic and electron-charge excitations that are con- fined to the surface of a metal or semiconductor. They were proposed in the 1950's [1, 2] and have been ob- served via electron energy loss spectroscopy [3, 4] as well as optically via surface gratings [5] or attenuated total reflection [6]. Over the past decades, SPPs have found widespread technological applications, for example, in surface microscopy [7], for biomolecular detection [8], or lithography [9]. Because SPPs are focused to sizes smaller than the wavelength of light, they hold promise to realize miniaturized plasmon-based optoelectronic de- vices, and research in creating such plasmonic devices is flourishing [10], with the subject being dubbed "plas- monics" or "nano plasmonics," which is a huge applied physics field in its own right. In this Rapid Communication, we add a fundamen- tal physical aspect to the study of SPPs (and the field of plasmonics), and demonstrate that the surface plas- mon polaritons of recently discovered Weyl semimetals (WSMs), which possess topological properties, show a much richer (and unanticipated) structure compared to standard SPPs in ordinary metals and semiconductors. We find that due to the quantum anomalous electrody- namic response of the WSM (which is their hallmark), the retarded Weyl surface plasmon is strongly sensitive to details of the band structure. In particular, we find a geometry in which the SPP is nonreciprocal (i.e., the propagation is unidirectional), even without an applied external magnetic field. In addition, we show that the magnetoplasmon mode displays an additional longitudi- nal field dependence which is absent in ordinary metals. This can serve as a direct signature of Weyl semimetals in surface measurements. We note that the SPP physics in- troduced in this work applies to extrinsic or doped WSM materials with no requirement of fine-tuning the chem- ical potential to the band touching points, making our predictions easy to test experimentally. An important aspect of SPPs for technological appli- cations is their nonreciprocity, i.e., the SPPs can only propagate in one direction [10]. In conventional met- als, nonreciprocal modes are only possible by breaking time-reversal symmetry in an applied external magnetic field [49, 52]. This comes with great technological chal- lenges since for a sizable nonreciprocity, these magnetic fields have to be very large [10]. In this Rapid Commu- nication, we report nonreciprocal SPPs in the pristine WSMs that are induced by topological Weyl node sepa- ration without any external magnetic field. This provides an alternative route to nonreciprocal modes and could point to interesting technological applications of WSMs in nanoplasmonics. Weyl semimetals contain a valence and conduction band that touch in isolated points of the Brillouin zone near the chemical potential µ. The minimal Hamiltonian in the vicinity of such a Weyl node is [11, 12] H = χvp · σ − µ, (1) where χ = ± is the chirality, v the Fermi velocity, p the momentum, and σ are Pauli matrices. We consider the generic case of an extrinsic (doped) semimetal with positive chemical potential µ > 0 (the "Weyl metal," although we continue referring to them as WSM). The spectrum of the Hamiltonian (1) is linear with disper- sion εp = ±vp. Weyl nodes appear in pairs of oppo- site chirality [13–15], and they can be separated by a wave vector b in the first Brillouin zone or by an en- ergy offset b0 in energy. The topological properties of a Weyl semimetal are manifested in the form of a θ-term contribution to the action Sθ = e2 4πc (cid:82) dt(cid:82) d3r θ E · B with θ = 2(b · r − b0t) [16–19], where e is the electron charge, E the electric field and B the magnetic field. If the bands are degenerate with b0 = b = 0 (the so-called Dirac semimetal), the system does not possess topologi- cal properties. The θ term changes the electromagnetic response of the material in the bulk medium by altering the constitutive relation that links the displacement field and the electric field [20–26], which in frequency space reads ε∞ + 4πi ω σ E + ie2 πω (∇θ) × E + ie2 πcω θ B, (2) (cid:18) D = (cid:19) where ε∞ is the static dielectric constant of the medium and σ the conductivity. The first term in parentheses is the standard term as in normal metals, and the last two terms arise due to the chiral anomaly. The gradi- ent term in θ describes the contribution of an anoma- lous Hall current, and the last time-derivative term de- scribes the chiral magnetic effect [19]. Weyl semimetals have recently been reported for TaAs [27, 28], NbAs [29], YbMnBi2 [30], and Eu2Ir2O7 [31]. In addition, semimet- als with degenerate bands (Dirac semimetals) are re- ported for Cd3As2 [32–34], ZrTe5 [35], and Na3Bi [36]. They are parent materials from which nontrivial topo- logical behavior is induced by symmetry breaking, for example, by applying an external magnetic field [35, 37– 39]. These experimental results, which do not necessar- ily have an exclusive interpretation in terms of the chiral anomaly [39], motivate the search for direct signatures of Weyl semimetals that are of topological origin, i.e., effects that are not explained by the linear semimetal- lic Dirac dispersion and hence are not found in Dirac semimetals or small-gap semiconductors. We establish that SPP carry distinctive observable features in Weyl systems arising purely from their topological properties. Here, we show that the topological properties of Weyl semimetals affect the surface plasmon polariton disper- sion. There are two main results of this work: First, because a WSM is an optically anisotropic medium, the surface plasmon dispersion depends on the Weyl node separation. The effect is strongest in the retarded limit (i.e., small wave vector), where the magnitude of the wave vector is comparable to the bulk plasmon frequency (di- vided by c, the velocity of light). We find that for a time- reversal broken WSM without external magnetic field, the SPP dispersion resembles retarded magnetoplasmon modes in standard metals. In particular, for certain ori- entations of the surface, we predict a nonreciprocal dis- persion, i.e., a dispersion that depends on the sign of the wave vector. As the second main result of this work, we predict that the Weyl surface magnetoplasmon modes possess an anomalous longitudinal magnetic field depen- dence that is absent for standard metals. This effect is caused by the anomalous magnetic field dependence of the WSM longitudinal conductivity ("negative magne- toresistance") [25, 40, 41], a direct consequence of the 2 chiral anomaly. In the remainder of this Rapid Commu- nication, we derive both effects and discuss their exper- imental implications. All the algebraic details are pro- vided in the Supplemental Material. Surface plasmon polaritons are solutions of Maxwell's equations localized at the interface of two media. We con- sider the following geometry: A Weyl semimetal fills the positive half volume z > 0, and a vacuum for z < 0. The WSM-vacuum interface lies in the xy plane. For simplic- ity, we restrict the analysis to a single pair of Weyl nodes (although our results are obviously valid for WSM with arbitrary pairs of nodes). Since we have translational in- variance along the interface, the SSP are parametrized by the parallel wave vector q = (qx, qy). We search for electric fields of the form y, Ej z)eiqxx+iqyye−iωte−κjz, Ej = (Ej x, Ej (3) which decay exponentially away from the boundary, i.e., for which Re κ > 0, and we label j = 0 on the vacuum side and j > 0 enumerates the solutions in the WSM. The decay constants κj are determined from a solution of the wave equation ∇ × (∇ × E) = − 1 c2 ∂2 ∂t2 D, (4) z = D2 z and B1 z = B2 x/y = E2 x/y = B2 x/y and B1 iω∇ × E). Substituting the ansatz (3) in Eq. (4), where on the vacuum side, we have D = E, and for the WSM, D is given by Eq. (2) (the magnetic field in this expression is related to the electric field by Faraday's law B = c we obtain a linear system of equations. The zeros of the determinant of this system yield κj. In general, on the WSM side, it turns out that there are two solutions of Eq. (4) with exponentially decaying field. We demand the continuity of the parallel components of electric and magnetic fields (E1 x/y) and of the perpendicular components of the displacement fields (D1 z ). This gives four linearly inde- pendent conditions that determine the surface plasmon dispersion as well as the relative magnitude of the fields. In the following, we assume that the dielectric ten- sor does not depend on the wavelength or the posi- tion inside the WSM. This approximation applies if the inverse wave vector of the SPP is large compared to the Thomas-Fermi length, which in a WSM is propor- tional to the inverse Fermi wave vector [42, 43]. In this case, the diagonal component of the dielectric tensor is ε1(ω) = ε∞(1 − Ω2 3π ( µ )2 denotes the bulk plasmon frequency [44–46] with α = e2/vε∞ being the finestructure constant of the WSM. ω2 ), where Ω2 p = 4α p We first discuss the results for the SPP of a Dirac semimetal (for which b = 0 and b0 = 0). The SPP solves with κ0 = (cid:112)q2 − ω2/c2. This coincides with the con- ε1(ω) + κ0 = 0, (5) ventional SPP condition in standard metals [1, 47]. The 3 FIG. 1. Surface plasmon dispersion of a Weyl semimetal with broken time-reversal symmetry for different values of (top to bottom) (a) ωb/Ωp = 0.25, 0.5, 0.75, and 1; (b) 0.5, 1, and 1.5; and (c) 0.5. In (c), the continuous blue line denotes positive wave numbers q > 0 and the dotted blue line q < 0. In all plots, the bulk plasmon dispersion is indicated by thin red lines. The thin black lines denote the asymptotic light line and the nonretarded frequency, respectively, and the black dashed line indicates the SPP dispersion of a standard Dirac material. As discussed in the main text, the black dots mark the points where the SPP hybridizes with the bulk plasmon mode and is damped. ω =(cid:112)ε∞/(ε∞ + 1)Ωp (horizontal thin black line), which surface plasmon dispersion is indicated by a thick black dashed line in Figs. 1 (a)-(c). In the fully retarded limit cq (cid:28) Ωp, the SPP follows the light-line ω = cq [thin black line in Figs. 1(a)-1(c)] and turns over in the hydrodynamic limit cq (cid:29) Ωp to a constant value solves ε1(ω) = −1. In particular, for ε∞ = 1, this coin- √ 2 [1]. cides with the famous result by Ritchie, ω = Ωp/ However, these surface plasmon modes are different from ordinary metals since they are purely quantum with  p ∼ αn2/3 ∼ n2/3/. Fur- appearing explicitly [48]: Ω2 thermore, they show a sub-linear density dependence as opposed to a linear density-dependence for Ω2 p in ordi- nary metals. Electron interactions can introduce a loga- rithmic correction to this scaling through charge renor- malization [45]. We predict that the linear dispersion of a Dirac semimetal is manifested in a nonlinear dependence of the squared SPP mode frequency on doping density. Note that the characteristic scaling behavior may not only be probed by varying the doping density but also by finite-temperature measurements [31, 45]. We now present results for WSMs with broken time- reversal symmetry (b (cid:54)= 0) and broken parity (b0 (cid:54)= 0) for ε∞ = 13 as measured in Eu2Ir2O7 [31]. The results are shown in Fig. 1 for three relevant configurations: (a) b perpendicular to the sample surface, where the surface plasmon dispersion depends only on the magnitude of the parallel wave vector q; (b) b parallel to the surface with q parallel to b; and (c) b parallel to the surface with q perpendicular to b. The chiral anomaly induces an off- diagonal term in the dielectric tensor iε2(ω) = iε∞ωb/ω with ωb = 2e2b/πε∞. All the analytical calculational details are provided in the Supplemental Material. Fig- ure 1 (a) shows the SPP mode as a function of wave vec- tor for four values of ωb/Ωp = 0.5, 1, 1.5, and 2. The SPP deviates from the Dirac semimetal result (black dashed line) for intermediate wave vectors cq ∼ Ωp and departs from the light line at smaller wave number and energy. For comparison, we include as thin red lines the corre- sponding bulk plasmon modes for which one of the decay constants vanishes κ = 0 and the plasmon is no longer confined to the surface. As is evident from the plots, for some wave vectors, bulk and surface modes are de- generate, while in other regions (marked by black end points), the SPP vanishes. Here, a generalized SPP still exists, but with a complex wave vector q, indicating a coupling of surface and bulk modes [49]. It is interesting to note that the geometry [Fig. 1(a)] shows signs of the chiral anomaly, even though another characteristic signa- ture of WSM – topological Fermi arc surface states – are absent in this configuration. Similar features are seen in Fig. 1(b), which is shown for three different values ωb/Ωp = 0.5, 1, and 1.5. Both cases shown in Figs. 1(a) and 1(b) are reciprocal, i.e., the dispersion is indepen- dent of the sign of q. In case (c), however, the SPP dis- persion is nonreciprocal. For positive q > 0 (blue dotted line), there is a transition from the light line to an asymp- totic nonretarded constant frequency. For negative q < 0, the dispersion has a discontinuity as it merges with the bulk plasmon mode, at which point it jumps to a higher frequency. In particular, there exists a frequency range where the system supports only modes with q < 0. The nonreciprocity that we report could have interesting tech- nological applications [50]. While nonreciprocal SSP in normal metals require magnetic fields or impurities [51], nonreciprocity is a fundamental intrinsic material prop- erty of a WSM arising from its topological nature. Strikingly, the SPP with b (cid:54)= 0 resemble, on a qual- itative level, SPP of an ordinary metal in the presence of an external magnetic field [49, 52, 53], even though they have quite a different origin. Hence, the topolog- ical contribution to the dielectric tensor, which stems from an anomalous Hall displacement current, induces an "anomalous surface magnetoplasmon." This is a cen- tral result of our work. The retarded SPP dispersion of a WSM with b0 (cid:54)= 0 (and b = 0) is shown in Fig. 2 for ωb0 /Ωp = 0.5. The dispersion solves (cid:20) (cid:21) (cid:18) (cid:113) 1 (κ0 + κ1b) κ1a × 1b − q2 + κ2 1a) ω2 (cid:19) c2 + κ0(q2 − κ2 ω2 c2 ε1 + (κ1a ↔ κ1b) = 0, (6) 2 ω2 ω4 c4 ε2 4 ω8ε4 2 + c2ω6ε1ε2 c4 /ω2, with ω2 b0 2 − where the decay constants are κ1a/b = q2 − 1 c2 ε1± 1 2 and ε2 = 2e2b0Ωp/πcω2 = = 2e2b0Ωp/πcε∞. There is no de- ε∞ω2 b0 pendence on the direction of the parallel wave number q. In addition to the changed SPP dispersion, another observable effect would be a tilt of the field polarization out of the sagittal plane, as suggested for 3D topological insulators [54]. We now consider WSM surface magnetoplasmon modes, which turn out to have an unusual magnetic field dependence. As this effect is distinct from the zero-field anomalous SPP discussed so far, we restrict our atten- tion to the nonretarded limit and neglect corrections due to the separation of the Weyl nodes. The dielectric ten- sor takes the form ε1(ω) = ε∞ + 4πi ω σ. The conductiv- ity can be derived in a semiclassical framework, in which the Berry curvature modifies the semiclassical equation of motion [25, 40, 41, 55, 56]. In particular, the Berry curva- ture induces a longitudinal magnetoconductivity [40, 41] (cid:18) (cid:19) α π ω2 c Ω2 p σ(cid:107)(ω) = iΩ2 p 4πω 1 + , (7) pωc/4π(ω2 − ω2 whereas in an ordinary metal, there is only the first Drude term and no dependence on the magnetic field. Here, ωc = ev2B/µc denotes the cyclotron frequency and Eq. (7) holds for frequencies ω (cid:29) τ−1, where τ is the inelastic intranode scattering time [40]. Note that the relative strength of the Drude and the anomalous term depends on the magnetic field and the doping density. The remaining components of the conductivity tensor pω/4π(ω2 − ω2 take the standard Drude form σ⊥ = iΩ2 c ) and σxy = Ω2 c ), provided that Berry cur- vature corrections to the density of states and the in- trinsic orbital moment are neglected [25]. In the non- retarded limit, the electric field is given by E = −∇φ, where the electrostatic potential solves Poisson's equa- tion ∇2φ = 4πρ and ρ is the induced charge density. On the vacuum side, the potential solves ∇2φ = 0. Combin- ing Poisson's equation with the current and the continu- ity equation, we obtain (setting ε∞ = 1) ∇ · (σ∇φ) = 0. ∇2φ + (8) 4πi ω 4 FIG. 2. Surface plasmon dispersion of a Weyl semimetal with broken inversion symmetry for ωb0 /Ωp = 0.5, and 1. The notation is the same as in Fig. 1. We make the ansatz φ = φieiqxx+iqyye−iωte−κjz and impose the continuity of φ at the boundary. Integrating Eq. (8) across the interface, we find the second boundary condition φ(cid:48)(0+) − φ(cid:48)(0−) + [σ∇φ]z(0+) = 0, 4πi ω (9) where the prime denotes a derivative with respect to z. These conditions are of course equivalent to demanding the continuity of the parallel components of E and the perpendicular component of D. For a magnetic field parallel to the surface, we obtain the surface plasmon condition (cid:16) (cid:17) ω2 − (Ω2 p − ω2 + ω2 α c ) 1 + π p − ω2) sin2 θ = 0, −2ωωc sin θ − (Ω2 cos2 θ ω2 c Ω2 p (10) where θ is the relative angle between the magnetic field and the parallel momentum q. The term proportional to sin θ implies that the surface magnetoplasmon is non- reciprocal, i.e., the frequency depends on the sign of qy. For a dispersion perpendicular to the magnetic field (cos θ = 0), we find ω2 − ωωcsgn(q) − Ω2 p/2 = 0, which is the same form as in ordinary metals [57]. There is a modification of the plasmon dispersion for modes that propagate along the magnetic field (sin θ = 0), for which we find (cid:0)Ω2 p + ω2 c (cid:1). ω2 = 1 + α π 2 + α π ω2 c Ω2 p ω2 c Ω2 p (11) The anomalous correction to σ(cid:107) changes the magnetic field dependence of the plasmon frequency. Compar- ing with the standard surface plasmon relation ω2 = ε∞/(ε∞ +1)Ω2 p, we interpret the correction as an anoma- lous contribution to the dielectric constant of the medium p ∼ B2n−4/3, which acquires a magnetic ∆ε∞ = αω2 field dependence. If the anomalous term dominates (at c /πΩ2 small α and doping), the plasmon mode is equal to the bulk plasmon frequency, ω2 = Ω2 p. For a magnetic field that is perpendicular to the surface, we obtain the same anomalous surface plasmon dispersion as in Eq. (11). In summary, we predict a rich (and experimentally observable) structure of surface plasmon polaritons in doped Weyl semimetals. In particular, we show the fol- lowing: (a) There is a quantum surface plasmon mode with unusual density dependence; (b) for broken inver- sion symmetry, the retarded dispersion is strongly af- fected by the chemical potential imbalance; (c) for bro- ken time-reversal symmetry, the dispersion depends on the Weyl node separation, which acts similar to an inter- nal magnetic field; (d) a nonreciprocal dispersion arises naturally even without external magnetic fields; and (e) the magnetoplasmon mode acquires an additional longi- tudinal magnetic field dependence. These effects are ex- perimentally observable signatures of the chiral anomaly in Weyl semimetals and could point the way to future technological applications of these systems. This work is supported by LPS-MPO-CMTC, Mi- crosoft Q, and JQI-NSF-PFC (J.H. and S.D.S.), and Gonville and Caius College (J.H.). [1] R. H. Ritchie, Phys. Rev. 106, 874 (1957). [2] E. A. Stern and R. A. Ferrell, Phys. Rev. 120, 130 (1960). [3] C. J. Powell and J. B. Swan, Phys. Rev. 115, 869 (1959). [4] F. J. Garc´ıa de Abajo, Rev. Mod. Phys. 82, 209 (2010). [5] R. H. Ritchie, E. T. Arakawa, J. J. Cowan, and R. N. Hamm, Phys. Rev. Lett. 21, 1530 (1968). [6] N. Marschall, B. Fischer, and H. J. Queisser, Phys. Rev. Lett. 27, 95 (1971). [7] B. Rothenhausler and W. Knoll, Nature 332, 615 (1988). [8] M. Malmqvist, Nature 361, 186 (1993). [9] W. Srituravanich, N. Fang, C. Sun, Q. Luo, and X. Zhang, Nano Letters 4, 1085 (2004). [10] W. L. Barnes, A. Dereux, and T. W. Ebbesen, Nature 424, 824 (2003). [11] O. Vafek and A. Vishwanath, Annual Review of Con- densed Matter Physics 5, 83 (2014). [12] T. Wehling, A. Black-Schaffer, and A. Balatsky, Ad- vances in Physics 63, 1 (2014). [13] H. Nielsen and M. Ninomiya, Nuclear Physics B 185, 20 (1981). [14] H. Nielsen and M. Ninomiya, Nuclear Physics B 193, 173 (1981). 5 [22] P. Hosur and X.-L. Qi, Phys. Rev. B 91, 081106 (2015). [23] M. Kargarian, M. Randeria, and N. Trivedi, Scientific Reports 5, 12683 EP (2015). [24] A. A. Zyuzin and V. A. Zyuzin, Phys. Rev. B 92, 115310 (2015). [25] F. M. D. Pellegrino, M. I. Katsnelson, and M. Polini, Phys. Rev. B 92, 201407 (2015). [26] P. Goswami, G. Sharma, and S. Tewari, Phys. Rev. B 92, 161110 (2015). [27] S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.- C. Lee, S.-M. Huang, H. Zheng, J. Ma, D. S. Sanchez, B. Wang, A. Bansil, F. Chou, P. P. Shibayev, H. Lin, S. Jia, and M. Z. Hasan, Science 349, 613 (2015). [28] B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, Z. Fang, X. Dai, T. Qian, and H. Ding, Phys. Rev. X 5, 031013 (2015). [29] S.-Y. Xu, N. Alidoust, I. Belopolski, Z. Yuan, G. Bian, T.-R. Chang, H. Zheng, V. N. Strocov, D. S. Sanchez, G. Chang, C. Zhang, D. Mou, Y. Wu, L. Huang, C.-C. Lee, S.-M. Huang, B. Wang, A. Bansil, H.-T. Jeng, T. Neupert, A. Kaminski, H. Lin, S. Jia, and M. Zahid Hasan, Nat Phys 11, 748 (2015). [30] S. Borisenko, D. Evtushinsky, Q. Gibson, A. Yaresko, T. Kim, M. N. Ali, B. Buechner, M. Hoesch, and R. J. Cava, arXiv:1507.04847 (2015). [31] A. B. Sushkov, J. B. Hofmann, G. S. Jenkins, J. Ishikawa, S. Nakatsuji, S. Das Sarma, and H. D. Drew, Phys. Rev. B 92, 241108 (2015). [32] S. Borisenko, Q. Gibson, D. Evtushinsky, V. Zabolotnyy, B. Buchner, and R. J. Cava, Phys. Rev. Lett. 113, 027603 (2014). [33] T. Liang, Q. Gibson, M. N. Ali, M. Liu, R. J. Cava, and N. P. Ong, Nat Mater 14, 280 (2015). [34] M. Neupane, S.-Y. Xu, R. Sankar, N. Alidoust, G. Bian, C. Liu, I. Belopolski, T.-R. Chang, H.-T. Jeng, H. Lin, A. Bansil, F. Chou, and M. Z. Hasan, Nat Commun 5, 3786 (2014). [35] Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosic, A. V. Fedorov, R. D. Zhong, J. A. Schneeloch, G. D. Gu, and T. Valla, arXiv:1412.6543 (2014). [36] Z. K. Liu, B. Zhou, Y. Zhang, Z. J. Wang, H. M. Weng, D. Prabhakaran, S.-K. Mo, Z. X. Shen, Z. Fang, X. Dai, Z. Hussain, and Y. L. Chen, Science 343, 864 (2014). [37] H.-J. Kim, K.-S. Kim, J.-F. Wang, M. Sasaki, N. Satoh, A. Ohnishi, M. Kitaura, M. Yang, and L. Li, Phys. Rev. Lett. 111, 246603 (2013). [38] J. Xiong, S. K. Kushwaha, T. Liang, J. W. Krizan, W. Wang, R. J. Cava, and N. P. Ong, arXiv:1503.08179 (2015). [39] P. Goswami, J. H. Pixley, and S. Das Sarma, Phys. Rev. B 92, 075205 (2015). [15] H. Nielsen and M. Ninomiya, Physics Letters B 130, 389 [40] D. T. Son and B. Z. Spivak, Phys. Rev. B 88, 104412 (1983). (2013). [16] K. Fujikawa, Phys. Rev. Lett. 42, 1195 (1979). [17] A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 115133 [41] B. Z. Spivak and A. V. Andreev, arXiv:1510.01817 (2015). (2012). [42] S. Das Sarma, E. H. Hwang, and H. Min, Phys. Rev. B [18] P. Goswami and S. Tewari, Phys. Rev. B 88, 245107 91, 035201 (2015). (2013). [19] P. Hosur and X. Qi, Comptes Rendus Physique 14, 857 (2013). [20] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). [21] A. G. Grushin, Phys. Rev. D 86, 045001 (2012). [43] S. Li and A. V. Andreev, Phys. Rev. B 92, 201107 (2015). [44] M. Lv and S.-C. Zhang, International Journal of Modern Physics B 27, 1350177 (2013). [45] J. Hofmann and S. Das Sarma, Phys. Rev. B 91, 241108 (2015). [46] J. Zhou, H.-R. Chang, and D. Xiao, Phys. Rev. B 91, tions 14, 1223 (1974). 035114 (2015). [47] J. M. Pitarke, V. M. Silkin, E. V. Chulkov, and P. M. Echenique, Reports on Progress in Physics 70, 1 (2007). [48] S. Das Sarma and E. H. Hwang, Phys. Rev. Lett. 102, [52] K. Chiu and J. Quinn, Il Nuovo Cimento B 10, 1 (1972). [53] M. S. Kushwaha, Surface Science Reports 41, 1 (2001). [54] A. Karch, Phys. Rev. B 83, 245432 (2011). [55] D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, 206412 (2009). 1959 (2010). [49] R. F. Wallis, J. J. Brion, E. Burstein, and A. Hartstein, [56] R. Lundgren, P. Laurell, and G. A. Fiete, Phys. Rev. B Phys. Rev. B 9, 3424 (1974). 90, 165115 (2014). [50] R. Camley, Surface Science Reports 7, 103 (1987). [51] A. Hartstein and E. Burstein, Solid State Communica- [57] A. L. Fetter, Phys. Rev. B 32, 7676 (1985). 6 Supplemental Material: Surface plasmon polaritons in topological Weyl semimetals 1Condensed Matter Theory Center and Joint Quantum Institute, Department of Physics, University of Maryland, 2T.C.M. Group, Cavendish Laboratory, University of Cambridge, Cambridge CB3 0HE, United Kingdom College Park, Maryland 20742-4111 USA Johannes Hofmann1,2 and Sankar Das Sarma1 1 In this supplemental material, we provide details on the calculation of the retarded surface plasmon polariton dispersion. The interface between the vacuum and the Weyl semimetal (WSM) is in the x-y plane at z = 0, with the vacuum for z < 0 and the WSM for z > 0. The evanescent ansatz (3) for the electric field at the boundary solves the wave equation (4), where the displacement field D is related to the electric field E through the dielectric tensor ε(ω), Eq. (2). Hence, the electric field solves a homogeneous equation M E = 0, where q2 M = y − κ2 −qxqy ∓iqxκj ∓iqyκj q2 j −qxqy ∓iqxκj j ∓iqyκj x + q2 y x − κ2 q2  − ω2 c2 ε(ω), (S1) M = −κ2 with the positive sign on the vacuum side and negative on the WSM side. The zeros of the determinant of M determine the decay constant κj. On the vacuum side (j = 0), we find κ2 c2 for any orientation of q. On the WSM side (j = 1), we have for the various geometries shown in Fig. 1: 0 = q2 − ω2 Case (a): b = (0, 0, b), q = (0, q, 0) q2 − κ2 1 − ω2 iω2 c2 ε2(ω) c2 ε1 0 −κ2 − iω2 c2 ε2 1 − ω2 −iqκ1 0 c2 ε1 −iqκ1 q2 − ω2 c2 ε1  , 1 = q2 − ω2 κ2 c2 ε1 ± 1 ε1 (cid:114) − ω2 c2 ε2 2 (cid:16) q2 − ω2 c2 ε1 (cid:17) (S2) Case (b): b = (b, 0, 0), q = (q, 0, 0) M = c2 ε1 1 − ω2 0 −iqκ1 0 q2 − κ2 1 − ω2 iω2 c2 ε2 −iqκ1 c2 ε1 − iω2 c2 ε2 q2 − ω2 c2 ε1  , κ2 1 = q2 + (cid:19) − ε1 ± 1 2ε1 (cid:114) 4q2 ω2 c2 ε1ε2 2 + ω4 c4 ε4 2 (S3) Case (c): b = (b, 0, 0), q = (0, q, 0) 1 − ω2 0 0 q2 − κ2 M = c2 ε1 c2 ε1 −iqκ1 − iω2 q2 − ω2 c2 ε2 c2 ε1 where we define as in the main text ε1(ω) = ε∞(1− Ω2 ω2 ) and ε2(ω) = ε∞ωb/ω with ωb = 2e2b/πε∞. In Eqs. (S2)-(S4), we also note the results for the decay constant κ1 on the WSM side. For b0 (cid:54)= 0 and q = (q, 0, 0) as shown in Fig. 2, we have κ2 1 = q2 + c2 ε2 (S4) 2 ε1 , p − ε1 ω2 c2 0 −κ2 1 − ω2 −iqκ1 + iω2 0 (cid:18) ε2 (cid:19) ω2 c2 2 2ε1 (cid:18) ε2  , −κ2 M = c2 ε1(ω) 1 − ω2 − cκ1 ε2 − icq ε2 Ωp Ωp q2 − κ2 cκ1 Ωp ε2 1 − ω2 c2 ε1(ω) iω2 c2 ε2(ω) −iqκ1 ε2 q2 − ω2 icq Ωp c2 ε1(ω) 1 = q2 − ω2 κ2 c2 ε1 − ω4 2c4 ε2 2 + 1 2  , (cid:114) ω8 c8 ε4 2 + 4 ω6 c6 ε1ε2 2, (S5) /ω2 with ω2 b0 = 2e2b0Ωp/πcε∞. There are, in general, two linearly independent where we define ε2(ω) = ε∞ω2 b0 solutions for E on the vacuum and on the WSM side which correspond to modes that are localized at the interface. We impose as a boundary condition the continuity of the parallel component of the electric field E and the perpendicular component of the displacement field Dz as well as the continuity of the magnetic field B = c set of four linearly independent constraints. Setting the determinant of this constraint matrix equal to zero, we find the SPP condition for the various cases: iω∇ × E. This gives a (cid:18) 1a+κ1aκ1b+κ2 (cid:3)− ω2 1b)
1+q2(cid:2)κ0(1−
1)+(κ1a+κ1b)
1 1b)
1 c2
1 (κ1a+κ1b)(κ0+κ1a+κ1b+q2(1−
1) (cid:18) κ1a+κ1b+κ0(1−
1) (cid:19) (cid:3)− ω2 c2
1 = 0. (S7) 2 (cid:19) = 0. (S6) Case (a):
1κ0κ1aκ1b(κ1a+κ1b)+q2(cid:2)κ1aκ1b+κ0(κ1a+κ1b)+(κ2 Case (b): κ1aκ1b(κ0+κ1a+κ1b)+κ0(κ2 1a+κ1aκ1b+κ2 Case (c): ε1κ1 + κ0(ε2 (S8) Here, we denote by κ1a and κ1b the two solutions corresponding to ± in Eqs. (S2)-(S4). For b0 (cid:54)= 0, the condition is stated in Eq. (6) of the main text. In the case b (cid:54)= 0, the result is similar as for the as for a metal in a constant external magnetic field [S1, S2], and the SPP conditions agree when correcting for the difference in the dielectric tensor. The explicit form of the dielectric components in a WSM is, of course, different from the Drude form in an external magnetic field. 1 − ε2 2) − qε2 = 0. [S1] K. Chiu and J. Quinn, Il Nuovo Cimento B 10, 1 (1972). [S2] R. F. Wallis, J. J. Brion, E. Burstein, and A. Hartstein, Phys. Rev. B 9, 3424 (1974). 

1208.0548

1

1208

Probing a Single Nuclear Spin in a Silicon Single Electron Transistor

We study single electron transport across a single Bi dopant in a Silicon Nanotransistor to assess how the strong hyperfine coupling with the Bi nuclear spin $I=9/2$ affects the transport characteristics of the device. In the sequential tunneling regime we find that at, temperatures in the range of $100 mK$, $dI/dV$ curves reflect the zero field hyperfine splitting as well as its evolution under an applied magnetic field. Our non-equilibrium quantum simulations show that nuclear spins can be partially polarized parallel or antiparallel to the electronic spin just tuning the applied bias.

cond-mat.mes-hall

cond-mat

Probing a Single Nuclear Spin in a Silicon Single Electron Transistor F. Delgado(1), R. Aguado(2), and J. Fern´andez-Rossier(1,3)1 (1) International Iberian Nanotechnology Laboratory (INL), Av. Mestre Jos´e Veiga, 4715-330 Braga, Portugal (2) Instituto de Ciencia de Materiales de Madrid (ICMM-CSIC), Cantoblanco, 28049 Madrid, Spain (3) Departamento de F´ısica Aplicada, Universidad de Alicante, 03690 San Vicente del Raspeig, Spain (Dated: 25 June 2018) We study single electron transport across a single Bi dopant in a Silicon Nanotransistor to assess how the strong hyperfine coupling with the Bi nuclear spin I = 9/2 affects the transport characteristics of the device. In the sequential tunneling regime we find that at, temperatures in the range of 100mK, dI/dV curves reflect the zero field hyperfine splitting as well as its evolution under an applied magnetic field. Our non-equilibrium quantum simulations show that nuclear spins can be partially polarized parallel or antiparallel to the electronic spin just tuning the applied bias. PACS: 73.23.Hk, 31.30.Gs, 74.55.+v, 75.75.-c The amazing progress both in the silicon process- ing technologies and in the miniaturization of silicon based transistors has reached the point where single- dopant transistors have been demonstrated.1 -- 7 Whereas this progress has been fueled by the development of clas- sical computing architectures, it might also be used for quantum computing. In this regard, the electronic and nuclear spins of single donors in silicon are very promis- ing building blocks for quantum computing.8 -- 10 Progress along this direction makes it necessary to implement sin- gle spin readout schemes both for electronic and nuclear spins. Single electronic spin readout has been demon- strated, both in GaAs quantum dots as well as in P doped Silicon Nanotransistors.11,12 The readout of the quantum state of a single nuclear spin, much more challenging, has been demonstrated for NV centers in diamond taking advantage of single spin optically detected magnetic resonance afforded by the ex- traordinary properties of that system.13 Single nuclear spin readout with either optical14 or a combined electro- optical techniques15 has been proposed, but remains to be implemented. Here we explore the electrical readout of a single nuclear spin, more suitable for an indirect band-gap host like Si. A preliminary step is to construct a circuit whose transport is affected by the quantum state of the nuclear spin. There is ample experimental evidence of the mutual influence of many nuclear spins and trans- port electrons in III-V semiconductor quantum dots in the single electron transport regime.16 -- 19 In particular, Kobayashi et al. have reported hysteresis in the dI/dV upon application of magnetic fields, reflecting the real- ization of different ensemble of nuclear states coupled to the electronic spin via hyperfine coupling.19 Here we propose a device where a single nuclear spin is probed in single electron transport. We model the single electron transport in a silicon nanotransistor such that, in the active region, transport takes place through a single Bi dopant, see Fig. 1. We show that, at suffi- ciently low temperatures, the dI/dV curves of this de- vice probe the hyperfine structure of the dopant level. In FIG. 1. (color online) a) Scheme of the Si:Bi FinFET nan- otransistor. b) Trapping Coulomb potential of the Bi dopant an single energy level participating in the transport. turn, the occupations of the nuclear spin states are af- fected by the transport electrons. Whereas single dopant transistors have been demonstrated for single P, As and in Si,3,4,6,12 we choose Bi because it has a much B, larger hyperfine splitting,20 -- 22 due to both a larger nu- clear spin I = 9/2 and a larger hyperfine coupling con- stant (A ≈ 6.1µeV). The zero-field splitting of the Bi donor level is given by 5A and has been observed by electron spin resonance20 -- 22 and in photoluminescence experiments with many dopants.23 We consider the sequential transport regime, where the occupation of the donor level fluctuates between q = 0 and q = 1. In the q = 0 state, the nuclear spin interacts only with the external field. In the q = 1 state, the electron and the nuclear spin are hyperfine coupled. The Hamiltonian that describes both states reads20 -- 22,24 H = q (cid:16)ǫd + eVG + A~S · ~I + ωeSz(cid:17) + ωN Iz, (1) where ǫd is the donor energy level with respect to the Fermi energy, which we take as EF = 0, and VG denotes an external gate voltage. We assume that valley degen- eracies of the donor level are split-off and neglect the valley degree of freedom. The third term is the hyper- fine coupling, and the last two, where ωe = geµBBz and ωN = gnµN Bz, correspond to the electron and nuclear Zeeman terms, with ge (gn) the electron (nu- 2 q = 0 and q = 1 manifolds. The corresponding transition rates are be calculated using the Fermi golden rule with Htun as the perturbation:26 0 X hM Iz(m), σi2 , (3) Γη m,M = Γη σ where Iz, σi ≡ Izi ⊗ σi. In the following we take the applied bias convention µS − µD = eV , with µS = eV /2 and µD = −eV /2. For a given temperature, bias and gate voltages and Hamiltonian parameters, we obtain the steady state solution of the master equation, ignoring the effect of the fast-decaying coherences. This yields the steady state occupations Pm(V ) and PM (V ). We consider the sequential tunneling regime, in which the energy level broadening induced by coupling to the electrodes Γ0 is small, Γ0 ≪ kBT . This also justifies the markovian approximation implicit in the Bloch-Redfield master equation. In this regime, current flows when the bias enables charge fluctuations of the dopant level. The steady state current corresponding to electrons flowing from the source electrode to the dopant level is given by I = e X m,M nPm(V )fS(∆M,m)ΓS m,M −PM (V ) [1 − fS(∆M,m)] ΓS m,Mo, (4) where ∆M,m = ǫM − ǫm and fS(ǫ) = f (ǫ − µS) is the Fermi function relative to the chemical potential of the S electrode. The first term in the right hand-side of Eq.(4) represents the electrons flowing from the S electrode to the empty Bi, while the second one corresponds to elec- trons flowing from the q = 1 Bi to the S electrode. In steady state, the continuity equation ensures that cur- rent between the dopant and the drain is the same than the source-dopant current. Figure 3a) shows the differential conductance dI/dV map for zero-applied magnetic field, with I = I/(eΓ0) and Γη 0 = Γ0/2. At zero bias, the conductance is zero except at the special value of VG for which the addition energy vanishes. Far from this point, the zero-bias charge of the dopant state, hereafter denoted with q0, is either q0 = 0 or q0 = 1. The finite bias conductance has a peak whenever the bias energy, eV /2, matches the energy difference between two states with different charge, m for q = 0 and M for q = 1, that are permitted by the spin selection rule implicit in Eq.(3). The height of the peak is proportional to both the non-equilibrium occupations Pm and PM and to the quantum mechanical matrix element Γη m,M . This determines the very different spectra when the zero bias charge in the dopant is q = 0 or q = 1. The width of the dI/dV peaks is proportional to kBT , so that the dI/dV spectra can resolve the hyperfine structure provided that kBT is smaller than the splitting of the levels. The energy differences inside the F = 4 and F = 5 manifolds, see Fig. 2), are roughly proportional to A. Thus, while the zero-field splitting can be resolved at T = 0.3 K, temperature must be significantly below 50 mK to resolve the finite field structure, see Fig. 3c). FIG. 2. (color online) Scheme of the current-induced allowed transition for the a) q = 1 charged system and b) q = 0 uncharged system. It has been assumed that ωN ≪ kBT ≪ ωe . ∆0. clear) g-factors and µB (µN ) the Bohr (nuclear) mag- neton. In equilibrium, i.e., at zero bias, the occupation of the dopant level depends on the value of the addi- tion energy, which ignoring the Zeeman terms and the tiny correction due to the hyperfine coupling, is given by ε0(VG) ≡ ǫd + eVG. We denote the q = 0 eigenstates as mi. Their energies read as ǫm ≡ ωN Iz. The eigenenergies and eigenvectors of q = 1 are denoted by ǫM and M i. The q = 1 zero- field Hamiltonian A~I · ~S can be diagonalized in terms of the total angular operator F , resulting in two multiplets (F=4, F=5) with energies EF =4 = −11A/4 and EF =5 = 9A/4, and a zero-field splitting ∆0 = 5A ≈ 30µeV. At finite magnetic field, the exact eigenvalues of H can also be calculated analytically.22 The corresponding energy levels are shown in Fig. 2. The tunneling Hamiltonian between the single Bi dopant level and the source and drain electrodes reads as Htun = X λσ Vλ (cid:0)d† σcλσ + h.c(cid:1) , (2) where operator cλ,σ annihilates an electron with spin σ and orbital quantum number λ ≡ η, ~k, with wave vec- tor ~k and electrode index η = S, D, while operator dσ annihilates a spin σ electron in the dopant level. The scattering rate for the tunneling process, ignoring the hyperfine coupling, is given by Γη  Vη2ρη, where ρη is the density of states of the electrode. Our model is very similar to the one used to describe single electron transport through a quantum dot exchanged coupled to a single Mn atom.25,26 0 = 2π The dissipative dynamics of the electro-nuclear spin system under the influence of the coupling to the elec- trodes is described by a Bloch-Redfield (BR) master equation.26,27 The coupling to the reservoir, given by the tunneling Hamiltonian, involves transitions between the 3 FIG. 3. (color online) a) and b) Contour plot of the dI/dV vs. applied bias V and on-site energy ε0 at zero magnetic field (left) and B = 0.6T (right). c) and d) Conduction spectrum dI/dV as a function of applied bias for different magnetic fields at ε0 = −0.4µeV and ε0 = 0.8meV respectively. White horizontal lines in panel a) and b) marks the values of ε0 in the 2D plots c) and d). In all cases, T = 10mK and Γ0 = 0.1µeV. Let us consider first the q0 = 1 case (left panel in Fig. 3). At 10 mK only the ground state(s) is (are) oc- cupied. Thus, a single transition is seen, from the q = 1 to the q = 0 states. As the magnetic field is ramped, the energy of the transition increases, reflecting the elec- tronic Zeeman shift. In contrast, in the q0 = 0 case (right panel in Fig. 3), all the Zeeman split nuclear levels are equally populated, even down to mK temperatures. Spin conservation selection rule implicit in Eq. (2) connects these 10 quasi-degenerate states of the q = 0 manifold to the hyperfine spin-split levels of the q = 1 manifold with different energies. As a result, the dI/dV curve reveals 2 peaks at zero field, reflecting the splitting between the F = 4 and F = 5 states. At higher fields, the two zero- field peaks split in up to 10 peaks, that can be resolved at low enough temperature [see Figs. 3c) and 4b)]. Interestingly, the application of a bias to the q0 = 0 state, for which the nuclear spin states are randomized, can result in a finite average nuclear magnetic moment. We show this in Fig. 4a) for finite B. At zero bias, the charge of the dopant level is q0 = 0, and the nuclear spins are randomized. When the bias hits the addition energy a selective depopulation of a given Iz level of the q = 0 manifold starts, in favor of a q = 1 state that mixes the Iz and Iz ±1 components, resulting in a net accumulation of nuclear spin. When all the transitions to the F = 4 manifold are allowed, the nuclear spin vanishes again. Then, when the bias permits the transitions to the F = 5 manifold, the nuclear spin accumulation starts in the opposite direction. Thus, when eV /2 matches the center of the F = 4 multiplet, see Fig. 4a), the nuclear spins tend to align antiparallel to the electronic spin. Then, when eV /2 reaches the center of the F = 5 multiplet, FIG. 4. (color online) a) Average electronic occupation of the Bi, hQi/e (black line) and nuclear and electronic spins, hIzi (red line) and hSzi (blue line), respectively. b) dI/dV vs. bias for different temperatures. Same parameters as 3c) with B = 0.6T. the nuclear spins prefer aligning parallel to the electronic spin. Whereas all our results discussed so far refer to steady state conditions, it is worth pointing out that there are two very different time scales in the dynamics of the sys- tem. Whereas the charge equilibrates in the dopant level in a time scale set by 1/Γ0, the nuclear spin relaxation, dominated by many events of hyperfine exchange with the electronic spin and subsequent recharging of the Bi,28 takes place at a much longer time scale, hundreds of time larger than 1/Γ0, but still much shorter than the intrinsic T1 time of the nuclear spin. Thus, charge fluctuations in the Bi induce nuclear spin relaxation.28 We finally discuss the experimental feasibility of our proposal with state of the art techniques. First, accord- ing to our simulations, see Fig. 4b), the finite field hyper- fine splitting is resolved at 10 mK but not a 20 mK. At 40 mK the 2 humps associated to the F = 4 and F = 5 man- ifolds are clearly resolved. Keeping the transport in the sequential tunneling regime requires that Γ0 ≪ kBT , which at 10mK, translates into I ≪ 200pA. This is within reach of experimental setups.12,16,19,29,30 In conclusion, we have studied the single electron transport spectroscopy of the hyperfine structure of a Bi dopant in a silicon nanotransistor. We have shown that, at sufficiently low temperatures, and when the dopant is ionized with a gate, the dI/dV corresponding to se- quential transport can resolve the hyperfine spectrum of the electron in the donor level. In addition, the non- equilibrium transport at finite field results in a hyper polarization of the nuclear spin state, or nuclear spin ac- cumulation. These results are different from our previous work, where we considered the same system in a different transport regime, cotunneling, and we showed that in- elastic cotunneling of the dopant in the q = 1 state could also resolve the hyperfine spectrum and drive the nuclear spin states out of equilibrium.24 Future work should de- termine how, in the cotunneling regime, the appearance of the Kondo effect7 competes with the reported effect. This work has been financially supported by MEC- Spain (Grant Nos. FIS2010-21883-C02-01, FIS2009- 08744, and CONSOLIDER CSD2007-0010) as well as Generalitat Valenciana, grant Prometeo 2012-11. 1H. Sellier, G. P. Lansbergen, J. Caro, S. Rogge, N. Collaert, I. Ferain, M. Jurczak, and S. Biesemans, Phys. Rev. Lett. 97, 206805 (2006). 2M. Pierre, R. Wacquez, X. Jehl, M. Sanquer, M. Vinet, and O. Cueto, Nature nanotechnology 5, 133 (2009). 3G. P. Lansbergen, G. C. Tettamanzi, J. Verduijn, N. Collaert, S. Biesemans, M. Blaauboer, and S. Rogge, Nano Letters 10, 455 (2010). 4K. Y. Tan, K. W. Chan, M. Mottonen, A. Morello, C. Yang, J. v. Donkelaar, A. Alves, J.-M. Pirkkalainen, D. N. Jamieson, R. G. Clark, et al., Nano Letters 10, 11 (2010). 5V. N. Golovach, X. Jehl, M. Houzet, M. Pierre, B. Roche, M. San- quer, and L. I. Glazman, Phys. Rev. B 83, 075401 (2011). 6M. Fuechsle, J. A. Miwa, S. Mahapatra, H. Ryu, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Sim- mons, Nature Nanotechnology 7, 242 (2012). 7G. C. Tettamanzi, J. Verduijn, G. P. Lansbergen, M. Blaauboer, M. J. Calder´on, R. Aguado, and S. Rogge, Phys. Rev. Lett. 108, 046803 (2012). 8B. Kane, Nature 393, 133 (1998). 9D. P. DiVincenzo, D. Bacon, J. Kempe, G. Burkard, and K. B. Whaley, Nature 408, 339 (2000). 10T. D. Ladd, F. Jelezko, R. Laflamme, Y. N. Y, C. Monroe, and J. L. O'Brien, Nature 464, 45 (2010). 11J. M. Elzerman, R. Hanson, L. H. W. van Beveren, B. Witkamp, L. M. K. Vandersypen, and L. P. Kouwenhoven, Nature 430, 431 (2004). 12A. Morello, J. J. Pla, F. A. Zwanenburg, K. W. Chan, K. Y. Tan, H. Huebl, M. Mottonen, C. D. Nugroho, C. Yang, J. A. v. Donkelaar, et al., Nature 467, 687 (2010). 13P. Neumann, J. Beck, M. Steiner, F. Rempp, H. Fedder, P. R. Hemmer, J. Wrachtrup, and F. Jelezko, Science 329, 542 (2010). 4 14K.-M. C. Fu, T. D. Ladd, C. Santori, and Y. Yamamoto, Phys. Rev. B 69, 125306 (2004). 15D. Sleiter, N. Y. Kim, K. Nozawa, T. D. Ladd, M. L. W. Thewalt, and Y. Yamamoto, New Journal of Physics 12, 093028 (2010). 16J. R. Petta, J. M. Taylor, A. C. Johnson, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 100, 067601 (2008). 17D. J. Reilly, J. M. Taylor, J. R. Petta, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Science 321, 817 (2008). 18S. Foletti, H. Bluhm, D. Mahalu, V. Umansky, and A. Yacoby, Nature Physics 5, 903 (2009). 19T. Kobayashi, K. Hitachi, S. Sasaki, and K. Muraki, Phys. Rev. Lett. 107, 216802 (2011). 20R. E. George, W. Witzel, H. Riemann, N. V. Abrosimov, N. Notzel, M. L. W. Thewalt, and J. J. L. Morton, Phys. Rev. Lett. 105, 067601 (2010). 21G. W. Morley, M. Warner, A. M. Stoneham, P. T. Greenland, J. van Tol, C. W. M. Kay, and G. Aeppli, Nature Materials 9, 725 (2010). 22M. H. Mohammady, G. W. Morley, and T. S. Monteiro, Phys. Rev. Lett. 105, 067602 (2010). 23T. Sekiguchi, M. Steger, K. Saeedi, M. L. W. Thewalt, H. Rie- mann, N. V. Abrosimov, and N. Notzel, Phys. Rev. Lett. 104, 137402 (2010). 24F. Delgado and J. Fern´andez-Rossier, Phys. Rev. Lett. 107, 076804 (2011). 25A. L. Efros, E. I. Rashba, and M. Rosen, Phys. Rev. Lett. 87, 206601 (2001). 26J. Fern´andez-Rossier and R. Aguado, Phys. Rev. Lett. 98, 106805 (2007). 27C. Cohen-Tannoudji, G. Grynberg, and J. Dupont-Roc, Atom- Photon Interactions (Wiley and Sons, INC., New York, 1998). 28L. Besombes, Y. Leger, J. Bernos, H. Boukari, H. Mariette, J. P. Poizat, T. Clement, J. Fern´andez-Rossier, and R. Aguado, Phys. Rev. B 78, 125324 (2008). 29J. Baugh, Y. Kitamura, K. Ono, and S. Tarucha, Phys. Rev. Lett. 99, 096804 (2007). 30T. S. Jespersen, K. Grove-Rasmussen, J. Paaske, K. Muraki, T. F. an J. Nygard, and K. Flensberg, Nature Physics 7, 348 (2011). 

1108.5256

2

1108

Second-harmonic generation from coupled plasmon modes in a single dimer of gold nanospheres

We show that a dimer made of two gold nanospheres exhibits a remarkable efficiency for second-harmonic generation under femtosecond optical excitation. The detectable nonlinear emission for the given particle size and excitation wavelength arises when the two nanoparticles are as close as possible to contact, as in situ controlled and measured using the tip of an atomic force microscope. The excitation wavelength dependence of the second-harmonic signal supports a coupled plasmon resonance origin with radiation from the dimer gap. This nanometer-size light source might be used for high-resolution near-field optical microscopy.

cond-mat.mes-hall

cond-mat

Second-harmonic generation from coupled plasmon modes in a single dimer of gold nanospheres A. Slablab1, L. Le Xuan1,∗ M. Zielinski1, Y. de Wilde2, V. Jacques1, D. Chauvat1, and J.-F. Roch1 1Laboratoire de Photonique Quantique et Mol´eculaire, Ecole Normale Sup´erieure de Cachan and CNRS, UMR 8537, F-94235 Cachan, France and 2Institut Langevin, ESPCI ParisTech and CNRS, UMR 7587, F-75231 Paris, France (Dated: June 5, 2018) We show that a dimer made of two gold nanospheres exhibits a remarkable efficiency for second- harmonic generation under femtosecond optical excitation. The detectable nonlinear emission for the given particle size and excitation wavelength arises when the two nanoparticles are as close as possible to contact, as in situ controlled and measured using the tip of an atomic force microscope. The excitation wavelength dependence of the second-harmonic signal supports a coupled plasmon resonance origin with radiation from the dimer gap. This nanometer-size light source might be used for high-resolution near-field optical microscopy. INTRODUCTION During the last decade, nonlinear nanoparticles have been extensively studied as new light sources at the nanoscale. In particular, nanoparticles consisting of non- centrosymmetric material exhibit second-harmonic gen- eration (SHG) [1–4] which can be used for nonlinear opti- cal microscopy [5–9]. For nanoparticles made of pure no- ble metals, high electron polarisability could lead to much stronger nonlinear effects. However, inversion symmetry of the metallic crystalline structure forbids bulk second- order electric dipole response. For a metallic nanosphere, second-order polarization associated to induced surface dipole moments and volumic quadrupole moments pro- duces a weak SHG signal [10–12], with a strong depen- dence to the nanoparticle shape and size [13, 14]. Efficient SHG at the nanoscale can be obtained by plas- mon enhancement at the surface of a metallic tip [15], or by coupled plasmon modes in engineered metallic nanostructures with specific geometry, like T-shaped gold nano-dimers [16], bowtie-shaped nano-antenna [17], and gold nanowires [18]. In that context, a simple dimer structure consisting of two metallic nanospheres appears as a testbed for studying the plasmon coupling influ- ence on the nonlinear optical response. Indeed, the lin- ear scattering of this composite nanostructure exhibits a wealth of specific properties depending on its geomet- rical parameters [19–24]. The coupled plasmon modes have resonance frequencies which can be tuned over the whole visible spectrum by changing the dimer geome- try [21]. Moreover, the electromagnetic field density is greatly enhanced at the dimer gap and highly efficient four-wave mixing has been observed from such metallic dimers [25, 26]. In this Letter, we explore the second-order nonlin- ear properties of a single dimer consisting of two gold nanospheres (GNs) with controlable distance. We show that this dimer nanostructure leads to highly efficient SHG under femtosecond optical illumination in spite of its apparent centrosymmetry considered as a whole. The relative position of the two GNs is adjusted with nanome- ter accuracy using the tip of an atomic force microscope (AFM) [23]. We show that the SHG signal under the ex- perimental excitation wavelength-particle sizes condition is strongly enhanced when the two GNs are very close to contact, with a strong dependence on their mutual size ratio. The variation of the SHG intensity with the exci- tation wavelength supports the role of an near-infrared (IR) resonance resulting from the coupling of the plas- mon oscillations in the two gold nanospheres [21, 25]. EXPERIMENT AND RESULTS The principle of the experiment is shown in Fig. 1(a). The nonlinear optical response of coupled GNs is inves- tigated using a nano-optomechanical setup consisting of an AFM (Asylum Research, MFP-3D BIO) on top of a self-made inverted optical microscope. The GNs (100- nm diameter) were purchased from the British Biocell International (BBI) corporation. The colloidal solution, which is stabilized in water by citric acid, is deposited by spin coating on a standard 150-µm thick glass coverslip. Before spin coating, the glass coverslip was silanized in order to functionalize its surface with NH+ 3 groups. Since the GNs have negative charges on their surface, an elec- trostatic interaction allows to efficiently catch them on the glass surface during the spin coating, thus leading to well-dispersed GNs on the substrate. The AFM is used both to record the surface topography of the sample and to perform mechanical manipulation of the GNs. The formation of a single gold dimer is gradually achieved by pushing a GN toward another with the AFM tip, thus controlling the interparticle distance from large separa- tion to contact between the two spheres. A titanium- doped sapphire (Ti:Sa) laser emitting 100 fs pulses in the 800-1000 nm wavelength range at 80 MHz repetition rate is tightly focused onto the sample through a high nu- merical aperture microscope objective (NA = 1.4, corre- sponding to a maximum collection half angle of 68◦). The 2 from the second-order nonlinear process [27]. We at- tribute this emission line to SHG from the dimer struc- ture. Furthermore, the emission spectra clearly exhibit a maximum of SHG when the excitation laser wavelength is within the 950 to 1000 nm range (Fig. 2(b)), giving ev- idence for an infrared resonance associated to the dimer structure. This spectral behavior of the second-harmonic re- sponse is well explained by the analysis of Ref. [21] which theoretically describes the linear optical response of a gold dimer in the nearly touching regime. When the two GNs are getting close to contact, the two single nanosphere plasmon modes are coupled, ending in a new plasmon mode which resonance is rapidly shifted towards the infrared. We confirm this prediction with a finite difference time domain (FDTD) simulation of the linear optical properties of a gold dimer, performed with the Lumerical software [28]. In agreement with Ref. [21], a resonant behavior of the scattering cross-section is ob- served as well as a rapid shift of this resonance to the infrared as the gap is reduced. Choosing a 0.08 nm ef- fective gap distance for touching GNs of 100 nm size, we compute the linear scattering cross-section of the dimer, displayed in Fig. 2(c). The dimer scattering cross-section variation closely matches the measured SHG excitation FIG. 2: Second-harmonic radiation of individual gold dimers. (a)-Emission spectrum from a gold dimer recorded for an ex- citation laser wavelength at λex = 980 nm, showing a strong SHG spectral peak and a much weaker and broader two- photon excited luminescence (TPEL). (b)-Emission spectra recorded while tuning the excitation laser wavelength λex from 850 to 1010 nm. (c)-FDTD simulation of the scattering cross- section of a dimer consisting of two 100 nm GNs separated by a 0.08 nm effective gap distance. (a)-Experimental setup. AFM tip: Olympus, FIG. 1: AC160TS used as purchased; MO: oil immersion microscope objective (×100, NA = 1.4); DM: dichroic mirror; FM: switchable mirror directing the collected light either to a spec- trograph or to a silicon avalanche photodiode (APD). To- pography measurements are done in the AFM tapping mode while the contact mode is used to perform mechanical manip- ulation of the GN. Optical images are recorded by scanning the sample in x and y directions while recording the emitted photons with the APD. The MO is mounted on a piezoelec- tric transducer in order to adjust the laser beam focus on the z-axis. (b)-When two GNs are in contact (see AFM image), a strong nonlinear emission is observed (see optical image). No emission is observed for isolated single GNs. (c)-When two isolated GNs are brought in contact using the AFM tip, an associated bright emission spot appears in the optical image. light emitted by the GNs is collected with the same ob- jective, spectrally filtered from the remaining excitation light using a dichroic beamsplitter, and finally directed either to a spectrograph or to a silicon avalanche pho- todiode (APD) working in the photon counting regime. The AFM tip and the excitation laser beam are carefully aligned on the same axis. Simultaneous record of the to- pography and the optical response then allows to monitor the onset of second-order nonlinear effects as the dimer is formed. A typical realization of the experiment is depicted in Figs. 1(b) and (c). For the laser input mean power range used in the experiment (∼ 500 µW), the nonlinear optical emission is not observed for isolated GNs (see Fig. 1(b)), whereas a strong nonlinear optical signal clearly appears when two isolated GNs are brought to contact using the AFM-based nanopositioning technique (Fig. 1(c)). Emission spectra recorded from this gold dimer while tuning the excitation laser wavelength from 850 to 1000 nm are shown in Fig. 2(b). For each excitation wavelength, a strong narrow emission peak at half the excitation wavelength is observed alongside with a weak and broad two-photon excited luminescence (TPEL) background (Fig. 2(a)). In addition, the peak intensity scales quadratically with the laser power, as expected DMAFM tipSpectrographAPDFM200 nm300 nmAFM imageOptical imagefs laser(a)(b)(c)GNsMO@ 990 nmxyz(b)250020001500100050005405205004804604404201000980960940920900880Emission wavelength (nm)λem=λex2Intensity (a.u.)Excitation wavelength (nm)Intensity (a.u.)(a)300025002000150010005000560520480440Emission wavelength (nm)Emission wavelength (nm)SHGTPELExcitation wavelength (nm)Scattering cross section (a.u.) 120011001000900(c)3 allows to infer when the two GNs are nearly in contact. As one GN is approached toward the other with nano- metric steps, one could expect a gradual rise of the SHG signal. However, no SHG is detected until the two GNs are very close to contact (see inset in Fig. 4-middle panel), in which case a high count rate of SHG photons is observed (Fig. 5-left panel) at IR resonance wavelength. By pushing further one GN towards the other, the ob- served SHG signal remains unchanged, allowing to con- clude that the SHG signal only appears when the two GNs are very close to contact. This behavior is related to the sudden increase of the excitation field at the gap re- gion for very small interparticle distance. Indeed, the in- duced accumulation of opposite charges discussed above leads to a strong enhancement of the excitation electro- magnetic field at the dimer gap. Using the measured particles sizes (particle diameters of 100 nm and 80 nm) from Fig. 4 (middle panel) as parameters and consid- ering the GNs completely spherical, FDTD simulations depicted in Fig. 4 (right panel) shows that the excitation field gains about 6 orders of magnitude from an inter- FIG. 4: (Left panel) (a) to (d)-Topography images showing the assembly of a dimer by using the AFM tip to move one GN towards the other. (Central panel) A cross-section (red dashed line in (a)) is used to estimate the distance d between the two GNs. The SHG signal is observed only when the two GNs are in contact as shown in the insets of the cross-section graphs. (Right panel) FDTD simulation of the electromag- netic field intensity building up at the dimer gap as this gap is reduced. The gap values are those inferred from the fit of the cross-sections. In the case (d) of contact between the two spheres, a 0.08 nm value is taken in order to agree with the spectral behavior shown in Fig. 2. The simulation has been done with spherical particles of 100 nm and 80 nm in diame- ter, as measures topographically. Note the log-scale showing a six-order-of-magnitude increase between the two extreme values of the gap. FIG. 3: Polar diagrams showing the SHG efficiency as a func- tion of the angle of the linearly-polarized excitation laser for two different dimers (see topography image). Intense lobes along the dimer axis are observed. spectra. In particular, it exhibits a minimum close to 920 nm and a maximum close to 990 nm, as observed in Fig. 2(a). Such a correlation supports the resonant plas- monic origin of the SHG emission. To further characterize the nature of the coupled plas- mon mode, a polarization analysis is performed while ex- citing the dimer in resonance at λex = 990 nm and rotat- ing the linear excitation polarization using a half-wave plate placed before the dichroic beamsplitter (Fig. 1(a)). The polar diagrams measured for two different dimers are shown in Fig. 3. The SHG intensity vanishes when the excitation polarization is perpendicular to the dimer axis, and increases to a maximum value when the polar- ization is set along the dimer axis, as revealed by the cor- responding AFM topographic images. This dipolar like response can be explained with a simple model based on the coupling between the plasmon oscillations in the two GNs. At the dimer gap, a strong accumulation of oppo- site charges close to each other is induced. Any charge oscillation driven by external light within the gap is ac- companied by a charge redistribution over each particle in order to maintain intraparticle charge neutrality. This redistribution finally leads to a large dipole strength of the gold dimer considered as a whole. While the interac- tion of light with charges is weak for an exciting electric field perpendicular to the dimer axis, the plasmon mode coupling becomes efficient when the field is applied along this axis, as experimentally observed (see Fig. 3). Now we study the onset of the nonlinear optical re- sponse while varying the interparticle distance from large separation to contact. For that purpose, a gold dimer is gradually formed with nanometric steps by pushing a GN toward another with the AFM tip. After each dis- placement, an AFM topography image and a raster scan SHG image are jointly recorded. Four intermediate sit- uations are shown in Fig. 4(a)-(d). Despite a relatively small AFM tip radius (≈ 7 nm), a direct measurement of the gap distance d between the two nanoparticles is obviously not possible. However, this parameter can be inferred with a few nanometers accuracy from the dip ob- served in a cross section of the topographic image of the dimer (see central panel of Fig. 4), once the size of the GNs and the tip radius are known. The corresponding fit 0°45°90°135°180°225°270°315°0°45°90°135°180°225°270°315°200 nmSHGDistance (nm)0200400(b)SHGSHGd∼20nmd∼4nmd∼0nm300 nmSHG100(c)100100SHGd∼38nm100 nm(a)100-202468logE2(d)particle distance decreasing from 4.0 to 0.08 nm. SHG originates from this huge field which is confined within a nanometric volume located at the dimer gap. This localized excitation is compatible with the SHG raster- scan images (see Fig.1(c)) which show bright spots with a diffraction-limited size (≈ 320 nm FWHM), as expected for a point-like emission. DISCUSSION Second-order nonlinear optical processes require an overall noncentrosymmetry which can originate from a combination of different effects, such as intrinsic geomet- rical features of the dimer, excitation field dissymetry, or our specific detection geometry. Since a simple size difference between the two GNs would break the dimer centrosymmetry [25], we measured the SHG efficiency as a function of the size difference h between the two GNs. The experiment is performed for 19 single dimers in close contact geometry while keeping the excitation at 990 nm wavelength. As shown in Fig. 5(a), the SHG efficiency decreases by almost two orders of magnitude when h is changed from 0 to 35 nm. This result is a priori in con- tradiction with the above argument where SHG efficiency might increase with the size difference between the two GNs of the dimer. To understand this behavior, we compute a FDTD sim- ulation of the optical scattering cross section of the dimer as a function of the h parameter. As the asymmetry be- tween the two GNs is increased, the resonance resulting 4 from the coupling of the plasmon modes is blue-shifted and its amplitude decreases (Fig. 5(b)). From these sim- ulations, we can then compute the fourth power of the electric field amplitude at the gap while exciting at the fixed 990 nm wavelength. This quantity, which is rele- vant to the locally enhanced field around the gap, is found to decrease by roughly two orders of magnitude when h varies from 0 to 35 nm (see inset in Fig. 5(b)), the same factor as measured. It confirms that the decrease in SHG intensity matches the wavelength shift in resonance, and that the SHG originates from the neighborhood region around the dimer gap, rather than from an overall bro- ken symmetry corresponding to the association of two GNs with different size. In our specific case, the SHG might originate from the specific experimental geometry. The tightly focused ex- citation beam creates a field dissymmetry from one side to the other of the gap plan in the light propagation direction. This dissymmetry induces nonlocally excited electric-dipole second-order processes inside the spheres and surrounding the gap, and locally excited quadrupolar processes from the metallic surfaces [11, 12, 14]. Any re- sulting off-axis radiation can then be efficiently collected by the high numerical aperture objective. In addition, the SHG could be enhanced either by specific geomet- rical features at the gap associated to facets of the two GNs [30] or by organic layers on the GNs surface [31]. CONCLUSION AND PROSPECTS In conclusion, we have experimentally demonstrated that highly efficient SHG is obtained from two GNs when very close to contact, for the given geometry-size rela- tion and optical excitation wavelength. Spectral analysis of the SHG response and polarization analysis support a coupled plasmon resonance origin of the nonlinear emis- sion. Since SHG is emitted from a nanometric volume at the dimer gap, it might be used as a nanosource for the development of high-resolution near-field imaging. Using the AFM-tip pushing technique, more complex nanos- tructures can be assembled, such as a trimer nanoparticle which can lead to a tunable nano-half wave plate [32], or a succession of metallic beads with decreasing sizes which forms an efficient plasmonic nano-lens with high field con- centration [33]. Such structures are likely to exhibit sur- prising nonlinear properties in the optical domain. FIG. 5: (a) SHG efficiency plotted on a log-scale as a function of the size difference h between the two GNs of an asymmetric dimer (see inset). The excitation wavelength is at λ = 990 nm and laser input mean power is (∼ 500 µW). (b) Simula- tion of the scattering cross-section for different values of h. Inset: Log-scale plot of the fourth power of the electric field amplitude at the dimer gap Egap4 as a function of h, for a constant 990 nm excitation wavelength. Acknowledgements We dedicate this work to our esteemed colleague Do- minique Chauvat who passed away during the prepara- tion of the manuscript. We are grateful to S. Perruchas and T. Gacoin for providing us with the GNs sample. We h=10nmh=20nmh=30nm1300120011001000900wavelength (nm)Scattering cross-section (a.u.)h(nm)h100090011001200Wavelength (nm)1300(a)(b)h=0nmEgap4101424101524101623020100456710023456710002403020100103210h(nm)201030400SHG emission (x10 cts.s )-1 3thank J.-J. Greffet and F. Marquier for priceless discus- sions. This work is supported by C'Nano Ile-de-France. ∗ Electronic address: [email protected] [1] J. C. Johnson, H. Yan, R. D. Schaller, P. B. Petersen, P. Yang and R. J. Saykally, "Near-Field Imaging of Non- linear Optical Mixing in Single Zinc Oxide Nanowires," Nano Lett. 2, 279-283 (2002). [2] A. B. Djurisi´c and Y. H. Leung, "Optical properties of ZnO nanostructures," Small 2, 944-961 (2006). [3] M. Zielinski, D. Oron, D. Chauvat, J. Zyss, " Second- harmonic generation from a single core/shell quantum dot," Small 5, 2835-2840 (2009). [4] M. Zielinski, S. Winter, R. Kolkowski, C. Nogues, D. Oron, J. Zyss, D. Chauvat, "Nanoengineering the sec- ond order susceptibility in semiconductor quantum dot heterostructures," Opt. Express 19, 6657-6670 (2011). [5] L. Bonacina, Y. Mugnier, F. Courvoisier, R. Le Dantec, J. Extermann, Y. Lambert, V. Boutou, C. Galez, J.-P. Wolf, " Polar Fe(IO3)3 nanocrystals as local probes for nonlinear microscopy," Applied Physics B 87, 399-403, (2007). [6] L. Le Xuan, C. Zhou, A. Slablab, D. Chauvat, C. Tard, S. Perruchas, T. Gacoin, P. Villeval, J.-F. Roch, "Photostable second-harmonic generation from a single KTiOPO4 nanocrystal for nonlinear microscopy," Small 4, 1332-1336 (2008). [7] C.-L. Hsieh, R. Grange, Y. Pu, D. Psaltis, " Three- dimensional harmonic holographic microcopy using nanoparticles as probes for cell imaging," Opt. Express 17, 2880-2891(2009). [8] A. V. Kachynski, A. N. Kuzmin, M. Nyk, I. Roy, P. N. Prasad, "Zinc Oxide Nanocrystals for Non-resonant Non- linear Optical Microscopy in Biology and Medicine," J. Phys. Chem. C 112, 10721-10724 (2008). [9] Y. Nakayama, P. J. Pauzauskie, A. Radenovic, R. M. Onorato, R. J. Saykally, J. Liphardt, P. Yang, "Tunable nanowire nonlinear optical probe," Nature 447, 1098- 1101 (2007). [10] V. L. Brudny, B. S. Mendoza, W. L. Mochan, "Second- harmonic generation from spherical particles," Phys. Rev. B 62, 11152-11162 (2000). [11] J. Dadap, J. Shan, T. F. Heinz, "Theory of optical second-harmonic generation from a sphere of centrosym- metric material: small-particle limit," J. Opt. Soc. Am. B 21, 1328-1347 (2004). [12] J. Butet, J. Duboisset, G. Bachelier, I. Russier-Antoine, E. Benichou, C. Jonin, P.-F. Brevet, "Optical Second Harmonic Generation of Single Metallic Nanoparticles Embedded in a Homogeneous Medium," Nano Lett. 10, 1717-1721 (2010). [13] J. Nappa, G. Revillod, I. Russier-Antoine, E. Benichou, C. Jonin, P.-F. Brevet, "Electric dipole origin of the second harmonic generation of small metallic particles," Phys. Rev. B 71, 165407-165410 (2005). [14] J. Shan, J. I. Dadap, I. Stiopkin, G. A. Reider, T. F. Heinz, "Experimental study of optical second-harmonic scattering from spherical nanoparticles," Phys. Rev. A 73, 023819-023822 (2006). [15] A. Bouhelier, M. R. Beversluis, A. Hartschuh, L. 5 Novotny, "Near-field second-harmonic generation in- duced by local field enhancement," Phys. Rev. Lett. 90, 013903-013906 (2003). [16] H. Husu, B. K. Canfield, J. Laukkanen, B. Bai, M. Kuit- tinen, J. Turunen, M. Kauranen, "Local-field effects in the nonlinear optical response of metamaterials," Meta- materials 2, 155-168 (2008). [17] T. Hanke, G. Krauss, D. Trautlein, B. Wild, R. Brats- chitsch, A. Leitenstorfer, "Efficient Nonlinear Light Emission of Single Gold Optical Antennas Driven by Few-Cycle Near-Infrared Pulses," Phys. Rev. Lett. 103, 257404-257407 (2009). [18] A. Benedetti, M. Centini, C. Sibilia, M. Bertolotti, "En- gineering the second harmonic generation pattern from coupled gold nanowires," J. Opt. Soc. Am. B 27, 408- 416 (2010). [19] W. Rechberger, A. Hohenau, A. Leitner, J. R. Krenn, B. Lamprecht, F. R. Aussenegg, "Optical properties of two interacting gold nanoparticles," Opt. Commun. 220, 137-141 (2003). [20] T. Atay, J.-H. Song, A. V. Nurmikko, "Strongly inter- acting plasmon nanoparticle pairs: From dipole-dipole interaction to conductively coupled regime," Nano Lett. 4, 1627-1631 (2004). [21] I. Romero, J. Aizpurua, G. W. Bryant, F. D. Garc´ıa de Abajo, "Plasmons in nearly touching metallic nanoparti- cles: singular response in the limit of touching dimers," Opt. Express 14, 9988-9999 (2006). [22] A. L. Lereu, G. Sanchez-Mosteiro, P. Ghenuche, R. Quidant, N. F. Van Hulst, "Individual gold dimers inves- tigated by far- and near-field imaging," Journal of Mi- croscopy 229, 254-258 (2008). [23] S. Schietinger, M. Barth, T. Aichele, O. Benson, "Plasmon-Enhanced Single Photon Emission from a Nanoassembled Metal Diamond Hybrid Structure at Room Temperature," Nano Lett. 9, 1694-1698 (2009). [24] P. K. Jain and M. A. El-Sayed, "Plasmonic coupling in noble metal nanostructures," Chem. Phys. Lett. 487, 153-164, (2010). [25] M. Danckwerts and L. Novotny, "Optical frequency mix- ing at coupled gold nanoparticles," Phys. Rev. Lett. 98, 026104-026107 (2007). [26] S. Palomba and L. Novotny, "Near-Field Imaging with a Localized Nonlinear Light Source," Nano Lett. 9, 3801- 3804 (2009). [27] R. W. Boyd, Nonlinear Optics (Academic Press, New York, 1992). [28] We use the gold dielectric constants reported in Ref. [29] for a 500-1400 nm wavelength range and we do not take into account the influence of the substrate. The two GNs are supposed to be in air and the polarization of the excitation field is linear oriented along the dimer axis. [29] P. B. Johnson and R. W. Christy, "Optical constants of the noble metals," Phys. Rev. B 6, 4370-4379 (1972). [30] S.-C. Yang, H. Kobori, C.-L. He, M.-H. Lin, H.-Y. Chen, C. Li, M. Kanehara, T. Teranishi, S. Gwo, "Plasmon Hy- bridization in Individual Gold Nanocrystal Dimers: Di- rect Observation of Bright and Dark Modes," Nano Lett. 10, 632-637 (2010). [31] P. Rooney, A. Rezaee, S. Xu, T. Manifar, A. Hassan- zadeh, G. Podoprygorina, V. Bohmer, C. Rangan, S. Mittler, "Control of surface plasmon resonances in dielec- trically coated proximate gold nanoparticles immobilized on a substrate," Phys. Rev. B 77, 235446-235452 (2008). [32] Z. Li, T. Shegai, G. Haran, H. Xu, "Multiple-particle nanoantennas for enormous enhancement and polariza- tion control of light emission," ACS Nano 3, 637-642 (2009). [33] K. Li, M. I. Stockman, D. J. Bergman, "Self-Similar Chain of Metal Nanospheres as an Efficient Nanolens," Phys. Rev. Lett. 91, 227402-227405 (2003). 6 

1303.3529

2

1303

Hall Drag and Magnetodrag in Graphene

Massless Dirac fermions in graphene at charge neutrality form a strongly interacting system in which both charged and neutral (energy) modes play an important role. These modes are essentially decoupled in the absence of a magnetic field, but become strongly coupled when a field is applied. We show that these ideas explain the recently observed giant magnetodrag, arising in classically weak fields when electron density is tuned near charge neutrality. We predict strong Hall drag in this regime, which is in stark departure from the weak coupling regime, where theory predicts the absence of Hall drag. Energy-driven magnetodrag and Hall drag arise in a wide temperature range and at weak magnetic fields, and feature an unusually strong dependence on field and carrier density.

cond-mat.mes-hall

cond-mat

Hall Drag and Magnetodrag in Graphene 1 Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA and 2 School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, USA Justin C. W. Song1,2 and Leonid S. Levitov1 Massless Dirac fermions in graphene at charge neutrality form a strongly interacting system in which both charged and neutral (energy) modes play an important role. These modes are essentially decoupled in the absence of a magnetic field, but become strongly coupled when a field is applied. We show that these ideas explain the recently observed giant magnetodrag, arising in classically weak fields when electron density is tuned near charge neutrality. We predict strong Hall drag in this regime, which is in stark departure from the weak coupling regime, where theory predicts the absence of Hall drag. Energy-driven magnetodrag and Hall drag arise in a wide temperature range and at weak magnetic fields, and feature an unusually strong dependence on field and carrier density. Graphene near charge neutrality (CN) hosts an intrigu- ing electron-hole system with unique properties[1 -- 10]. Our understanding of the behavior at CN would greatly benefit from introducing ways to couple the novel neu- tral modes predicted at CN to charge modes which can be easily probed in transport measurements. There is a long history of employing magnetic field for such a pur- pose, since transport in charge-neutral plasmas is ultra- sensitive to the presence of external magnetic fields[11]. A new interesting system in which magnetotransport at CN can be probed are atomically thin graphene double layer G/hBN/G structures[12, 13]. Strong Coulomb cou- pling between adjacent layers in these systems results in strong Coulomb drag, arising when current applied in one (active) layer induces a voltage in the adjacent (passive) layer[13 -- 22]. Recent measurements[13] revealed drag re- sistance that peaks near CN and has dramatic magnetic field dependence, with the peak value increasing by more than an order of magnitude (and changing sign) upon application of a relatively weak B field. Strong mag- netic field dependence of drag has been observed pre- viously in other double layer two-dimensional electron gas (2DEG) heterostructures[23 -- 25], however these ex- periments were carried out in the quantum Hall regime, whereas the anomalous magnetodrag found in Ref.[13] occurs at classically weak fields B <∼ 1 T. Here we explain this puzzling behavior in terms of an energy-driven drag mechanism which involves coupled energy and charge transport[20, 22] (see Fig.1). En- ergy transport plays a key role because of fast verti- cal energy transfer due to interlayer Coulomb coupling in G/hBN/G systems[20] and relatively slow electron- lattice cooling[28, 29]. As a result, current applied in one layer can create a spatial temperature gradient for elec- trons in both layers, giving rise to thermoelectric drag voltage. The effect peaks at CN, since thermoelectric re- sponse is large close to CN[8 -- 10] and diminishes as 1/EF upon doping away from CN[30, 31]. Drag arising from this mechanism depends on thermoelectric response and, unlike the conventional momentum drag mechanism, it is insensitive to the electon-electron interaction strength. Another interesting effect that can be probed in these FIG. 1: Energy-driven magnetodrag in a double layer graphene heterostructure close to CN. (a) Schematic of charge current, temperature gradients, and electric field in the two layers that give rise to a negative ρdrag(cid:107) . (b,c) Magnetodrag resistivity, ρdrag(cid:107) , obtained from Eqs.(11),(13). Parameter val- ues: B = 0.6 T, n0 = 1011 cm−2, T = 150 K, and ρ0 = h 3e2 . The B = 0 dependence taken from the model of drag at zero B field [20, 21].(d) Experimentally measured magnetodrag re- sistivity in G/hBN/G heterostructures, reproduced from Ref. [13] for the same B values as in (c). Application of magnetic field leads to a giant negative drag at CN. Note the simi- larity between data and theoretically predicted drag density dependence, B dependence, and sign. systems is that of Hall drag. It has long been argued that, at weak coupling, no Hall voltage can arise in the passive layer in the presence of current in the active layer [26, 27]. This is because transferred momentum is parallel to ve- locity, allowing only a longitudinal "back-current" to de- velop in the passive layer. As we shall see, a very differ- ent behavior arises at strong coupling, owing to the long- range energy currents leading to electron-lattice tempera- ture imbalance. Close to CN, the magnitude of the cross- couplings between charge and energy currents becomes large, producing a finite Hall drag, VH = Rdrag H I(cid:107). H and Rdrag(cid:107) As we will see, energy currents result in Hall and mag- netodrag resistances, Rdrag , that are large and peak near CN, see Fig.1 and Fig.2. These large values arise even for classically weak fields B ∼ 0.1 T, exceed- ing by two orders of magnitude the values found previ- ously in GaAs/AlGaAs 2DEG heterostructures [23 -- 25] at similar fields. The mechanism based on coupled energy and charge transport predicts large and negative drag at CN that matches recent experiments (see Fig.1c,d). Our mechanism naturally leads to Hall drag because verti- cal energy transfer between layers does not discriminate between longitudinal and transverse heat currents since temperature profile is a scalar field. This stands in con- trast to conventional momentum driven drag, where mo- mentum transfer is parallel to the applied current [26, 27]. Heat current and an electric field, induced by charge current and temperature gradients, are coupled via the thermoelectric effect altered by the B field, ∇T T . jq = Qj, E = Q (1) Here Q is a 2 × 2 matrix, of which diagonal compo- nents describe the Peltier and Thompson effects, and off- diagonal components describe the Nernst-Ettingshausen effect. Onsager reciprocity requires that Q in both the expressions for jq and E are the same [see analysis fol- lowing Eq.(9)]. As an example of how our mechanism produces drag consider the Hall bar geometry, see Fig.1. When a longitudinal charge current is applied in the ac- tive layer (for B (cid:54)= 0) a transverse (Ettingshausen) heat current develops in both layers through efficient verti- cal energy transfer. Nernst voltage in the passive layer results in a longitudinal magnetodrag of a negative sign. To obtain the electric field in layer 2 induced by cur- rent applied in layer 1, we first need to understand the coupling of temperature profiles T1,2(r) in the two layers. Energy transport in the system can be described by −∇κ1∇δT1 + a(δT1 − δT2) + λδT1 = −∇ ·(cid:0)Q(1)j(cid:1) −∇κ2∇δT2 + a(δT2 − δT1) + λδT2 = 0 with a the energy transfer rate between the two layers [20], λ the electron-lattice cooling rate, and δTi = Ti− T0 (here T0 is the lattice temperature, equal for both layers). Here we focus on a Hall-bar geometry of two parallel rectangular layers of dimensions L×W , L (cid:29) W , pictured in Fig.1a. For the sake of simplicity, we treat the electric and heat currents as independent of the x coordinate along the bar. In layer 1, current is injected at x = −L/2 and drained at x = L/2. In layer 2, the Hall drag voltage arising across the device, VH and the longitudinal drag voltage, V(cid:107), are evaluated as VH = E(2) y dy, V(cid:107) = L W E(2) x dy (3) (cid:90) W/2 −W/2 (cid:90) W/2 −W/2 2 The electric and thermal variables may depend on the transverse coordinate y, see below. Boundary conditions for a Hall bar require electric cur- rent being tangential to the side boundaries, y = ±W/2, and zero temperature imbalance at the ends, x = ±L/2, reflecting that the current and voltage contacts act as ideal heat sinks. The electric current parallel to the boundaries y = ±W/2 gives rise to the Ettingshausen heat current that may have a component transverse to the Hall bar. The divergence of this heat current, ap- pearing on the right hand side of Eq.(2), acts as an effec- tive boundary delta function source in the heat transport equations. Boundary conditions can profoundly influence the symmetry of the resultant drag resistivity, see below. We consider the case of a spatially uniform Q in both layers. The ideal heat sinks at x = ±L/2 mean that no temperature imbalance develops in the x-direction (ex- cept for some "fringing" heat currents near the contacts which give a contribution small in W/L (cid:28) 1, which we will ignore in the following discussion). Since no temper- ature gradients are sustained in the x-direction for from the ends, we can reduce our problem Eq.(2) to a quasi-1D problem with temperature profiles that only depend on the y-coordinate. As a result, the only heat source arises from the Ettingshausen effect Q(1)j = (Q(1) yx j)(cid:98)y. To describe transport in the presence of such a source, we will expand temperature variables in both layers in a suitable orthonormal set of functions. Here it will be convenient to use eigenstates of the operator ∂2 y with zero Neumann boundary conditions at y = ±W/2, given by (cid:19) (cid:18) 2πn W (cid:18) 2π(n + 1 (cid:19) 2 ) y , W un(y) = A cos y , vn(y) = A sin A = (2/W )1/2, n = 0, 1, 2... From the symmetry of the source in Eq.(2) we expect δT1,2(y) to be odd in y. Thus only the functions vn(y) are relevant, giving δT1,2(y) = δ T1,2(qn)A sin qny, qn = 2π(n + 1 2 ) W . (cid:88) qn nκ1δ T1 + a(δ T1 − δ T2) + λδ T1 = Fn q2 nκ2δ T2 + a(δ T2 − δ T1) + λδ T2 = 0 q2 and Fn = 2A(−1)nQ(1) where κ1,2 = κ(1,2) Eq.(4), we find the temperature profile in layer 2: xx yx j. Solving (4) δT2(y) = aFn L1(qn)L2(qn) − a2 vn(y), (5) (cid:88) n≥0 where Li(qn) = κiq2 n + a + λ (i = 1, 2). Since electron- lattice cooling is very slow [28, 29], with the correspond- ing cooling length values in excess of few microns, we will suppress λ in what follows. Because the boundaries (2) For each n we obtain a pair of algebraic equations 3 e2L11/T , and thermal conductivity is κ = L22/T 2. Com- paring to the heat current due to an applied charge cur- rent, Eq.(1), we identify L21 = −eQL11. Onsager reciprocity demands that the cross-couplings 21(−B) where B is the applied magnetic obey L12(B) = LT field (note the transposed matrix). In an isotropic sys- tem the off-diagonal components of L obey L(xy)(B) = L(yx)(−B). As a result, Onsager reciprocity reduces to L12(B) = L21(B) (9) in an isotropic system. Applying Eq. 9 to Eq. 8 in an 11 QL11∇T. open circuit, we find E = −e−1∇µ = T −1L−1 For an isotropic system Q = Qxx1 + iQxyσ2, L = Lxx1 + iLxyσ2, so that [Q, L] = 0, which gives Eq.(1). Several different regimes arise depending on the rela- tion between the interlayer cooling length ξ and the bar width W . Using Eq.(3) and Eq.(1) we obtain (cid:32) (cid:33) (cid:32) V(cid:107) VH = Rdrag(cid:107) Rdrag H −Rdrag Rdrag(cid:107) H , (10) (cid:33)(cid:32) (cid:33) I(cid:107) 0 −LG(ξ) W T κ giving the magnetodrag and Hall drag resistance values −G(ξ) T κ Q(1) Q(1) Rdrag H = xy Q(2) xx , Rdrag(cid:107) = xy Q(2) xy , (11) where we used Qxx = Qyy and Qxy = −Qyx for an isotropic system. For a narrow bar (or, slow cooling), we have ξ/W (cid:29) 1 and G → 0, giving vanishingly small H,(cid:107) . For a wide bar (or, fast cooling) we have G → 1 Rdrag so that Rdrag H,(cid:107) saturates to a universal value independent of the interlayer cooling rate. For typical device param- eters, we estimate ξ ≈ 40 nm at T = 300 K [20]. Since L, W are a few mircons for typical graphene devices, we expect them to be firmly in the G = 1 regime, with the Hall drag and magnetodrag attaining universal values in- dependent of the electron-electron interaction strength. To describe the density and B field dependence, we use a simple model for Q. Measurements indicate[8, 9] that thermopower and the Nernst effect in graphene are well described by the Mott formula [33], giving (cid:32) (cid:33) Q = π2 3e k2 BT 2ρ ∂[ρ−1] ∂µ , ρ = ρ(cid:107) ρH −ρH ρ(cid:107) , (12) with ρ the resistivity, e < 0 the electron charge, and µ the chemical potential. We use a simple phenomenological model [34] relevant for classically weak B fields: ρ0(cid:112)1 + n2/n2 0 ρ(cid:107) = , ρH = −Bn e(n2 + n2 0) , (13) where ρ0 is the resistivity peak value at the Dirac point, n is the carrier density, and parameter n0 accounts for broadening of the Dirac point due to disorder. We ac- count for disorder broadening of the density of states, dn/dµ = (n2 + n2 0)1/4(cid:0)2/(π¯h2v2 F )(cid:1)1/2 . FIG. 2: (a) Schematic of charge current, temperature gra- dients, and electric field in the two layers of a Hall bar that produces Hall drag. (b,c) Density dependence of Hall drag resistance, predicted from Eqs.(11),(13) for the same param- eter values as in Fig.1. (d) Density dependence of Qxx, Qxy, see text. in the transverse (y-direction) are free, a finite temper- ature imbalance between the edges can arise, given by ∆T = δT2(y = W/2) − δT2(y = −W/2). We find ∆T = 4A2(cid:88) ξ = (cid:112)κ1κ2/aκ. We evaluate the sum using the iden- tity(cid:80)∞ where we defined κ = κ1 + κ2 and a length scale aQ(1) yx j L1L2 − a2 = 1 − tanhπc n(1 + ξ2q2 q2 n) (cid:88) to obtain Q(1) yx j (cid:16) , (6) n≥0 n≥0 W κ 8 1 n=0 (n+ 1 2 )4+c2(n+ 1 2 )2 = π2 2c2 πc ∆T = W Q(1) yx j κ G(ξ), G(ξ) = 1 − 2ξ W tanh . (7) (cid:17) (cid:18) W 2ξ (cid:19) Connecting ∆T with the drag voltage, and in particu- lar determining its sign, requires taking full account of Onsager reciprocity. This analysis is presented below. In the same way that the applied charge current, j, in layer 1 causes a heat current (Peltier/Ettingshausen), a temperature imbalance in layer 2, ∆T , can sustain voltage drops across the sample (Thermopower/Nernst). These two effects are related by Onsager reciprocity con- straints. The cross couplings in the coupled energy and charge transport equations [32] arise from (cid:18) −j (cid:19) jq = (cid:18) eL11/T eL12 L21/T L22 (cid:19)(cid:32) ∇µ (cid:33) ∇ 1 T (8) where L are 2 × 2 matrices and e is the carrier charge. In this notation, the electrical conductivity equals σ = H = (W/L)Rdrag(cid:107) From Eqs.(11),(12),(13) and the Wiedemann-Franz re- lation for κ, we obtain ρdrag(cid:107) and Rdrag (see Fig.1b,c and Fig.2b,c, respectively). In that, we used the parameter values n0 = 1011 cm−2, ρ0 = h 3e2 , and a representative temperature, T = 150 K. These values match device characteristics (disorder broadening, n0, and peak resistivity, ρ0) described in Ref.[13]. As a sanity check, we plot the components of Q (in Fig.2d) which show the behavior near CN matching thermopower and Nernst effects measured in graphene[8, 9]. Analyzing magnetodrag, we find that ρdrag(cid:107) peaks at dual CN, taking on large and negative values (Fig.1b,c). Magnetodrag peak exhibits a steep B dependence, ρdrag(cid:107),peak ∝ −B2, bearing a striking resemblance to mea- surements reproduced in Fig.1d. In particular, our model explains the negative sign of the measured magnetodrag. Hall drag is large and sign-changing (see Fig.2b,c), tak- ing on values consistent with measurements[35]. Interest- ingly, the map in Fig.2b indicates that the sign of Rdrag is controlled solely by carrier density in layer 2, breaking the n1 ↔ n2 symmetry between layers. This behavior does not contradict Onsager reciprocity, it arises as a consequence of the asymmetric boundary conditions for the Hall bar: free boundary at y = ±W/2 and ideal heat sinks at the ends, δT (x = ±L/2) = 0. This allows for finite temperature gradients to be sustained across the bar but not along the bar, see Fig.2a. H For other geometries, the temperature gradient can be obtained by balancing the heat flux due to thermal conductivity against the net heat flux in the two lay- ers, (κ1 + κ2)∇δT = D Q(1)j1 (see Eq.(24) of Ref.[22]). The quantity D can in principle be obtained by solving heat transport equations. Adopting the same approach as above, we find a magneto and Hall-drag resistivity ρdrag = 1 T κ Q(2)D Q(1), E2 = ρdragj1. (14) clarify, We wish to in connection to For isotropic heat flow, D = 1. In this case, since Q(1) and Q(2) commute, the resulting drag is layer-symmetric, n1 ↔ n2[22]. In particular, Hall drag for D = 1 vanishes on the diagonal n1 = −n2. In contrast, for anisotropic heat flow, such as that discussed above, we expect a generic tensor D (cid:54)= 1 and thus no layer symmetry. recent measurements,[35] that layer symmetry n1 ↔ n2 implies a swap of current and voltage contacts. Layer symme- try, which implies D = 1 in Eq.(14), will therefore only hold for Hall bars equipped with wide voltage contacts, for which the contact and the bar widths are compa- rable. This is indeed the case for the cross-shaped de- vices used in Ref.[13]. However it is not the case for a Hall bar with noninvasive voltage probes which are much narrower than the bar width, as assumed in our analysis above. Noninvasive probes, which have little effect on 4 temperature distribution in the electron system, trans- late into D (cid:54)= 1 and no layer symmetry. In summary, magnetic field has dramatic effect on drag at CN because it induces strong coupling between neu- tral and charge modes, which are completely decoupled in the absence of magnetic field in a uniform system. Field- induced mode coupling leads to giant drag that dwarfs the conventional momentum drag contribution as well as a remnant drag due to spatial inhomogeneity[20]. Our es- timates indicate that these two contributions are orders of magnitude smaller than the predicted magnetodrag, which also has an opposite sign. The giant magnetodrag and Hall drag values attained at classically weak mag- netic fields, along with the unique density dependence and sign, make these effects easy to identify in experi- ment. The predicted magnetodrag is in good agreement with findings in Ref.[13]. Magnetic field, coupled with drag measurements at CN, provides a unique tool for probing the neutral modes in graphene. We acknowledge useful discussions with A. K. Geim, P. Jarillo-Herrero, L. A. Ponomarenko, and financial sup- port from the NSS program, Singapore (JS). [1] J. Gonzales, F. Guinea, and M.A.H. Vozmediano, Nucl. Phys. B424, 595 (1994); Phys. Rev. B 59, R2474 (1999). [2] D. E. Sheehy and J. Schmalian, Phys. Rev. Lett. 99, 226803 (2007). [3] D. T. Son, Phys. Rev. B 75, 235423 (2007). [4] O. Vafek, Phys. Rev. Lett. 98, 216401 (2007). [5] A. B. Kashuba, Phys. Rev. B 78, 085415 (2008). [6] L. Fritz, J. Schmalian, M. Mueller, S. Sachdev, Phys. Rev. B 78, 085416 (2008). [7] M. Mueller, L. Fritz, S. Sachdev, Phys. Rev. B 78, 115406 (2008). [8] Y. M. Zuev, W. Chang, and P. Kim, Phys. Rev. Lett. 102, 096807 (2009). [9] P. Wei, W. Bao, Y. Pu, C. N. Lau, and J. Shi, Phys. Rev. Lett. 102, 166808 (2009). [10] J. G. Checkelsky and N. P. Ong, Phys. Rev. B 80, 081413(R) (2009). [11] L. P. Pitaevskii, E.M. Lifshitz, Physical Kinetics (Perga- mon, Oxford, 1981) [12] L. Britnell, et al. Science 335, 947 (2012). [13] R. V. Gorbachev, et al. Nature Physics 8, 896 (2012). [14] S. Kim, I. Jo, J. Nah, Z. Yao, S. K. Banerjee, and E. Tutuc, Phys. Rev. B 83, 161401 (2011). [15] W.-K. Tse and S. Das Sarma, Phys. Rev. B 75, 045333 (2007). [16] B. N. Narozhny, Phys. Rev. B 76, 153409 (2007). [17] E. H. Hwang, R. Sensarma, and S. Das Sarma , Phys. Rev. B 84, 245441 (2011). [18] N. M. R. Peres, J. M. B. Lopes dos Santos, and A. H. Castro Neto, Europhys. Lett. 95, 18001 (2011). [19] M. I. Katsnelson, Phys. Rev. B 84, 041407 (2011). [20] J. C. W. Song and L. S. Levitov, Phys. Rev. Lett. 109, 236602 (2012). [21] B. N. Narozhny, M. Titov, I. V. Gornyi, and P. M. Os- trovsky, Phys. Rev. B 85, 195421 (2012). [22] J. C. W. Song, D. A. Abanin, L. S. Levitov, Nano Lett., 5 10.1021/nl401475u (2013). McEuen, Nature Physics, 9 103 (2013) [23] N. K. Patel, E. H. Linfield, K. M. Brown, M. Pepper, D. A. Ritchie, and G. A. C. Jones, Semicond. Sci. Technol., 12, 309 (1997). [29] A. C. Betz, S. H. Jhang, E. Pallecchi, R. Ferreira, G. F´eve, J-M. Berroir, and B. Placais, Nature Physics, 9 109 (2013). [24] H. Rubel, A. Fischer, W. Dietsche, K. von Klitzing, and [30] J. M. Ziman, Principles of the Theory of Solids, Cam- K. Eberl, Phys. Rev. Lett. 78 1763 (1997). bridge University Press (1979). [25] M. P. Lilly, J. P. Eisenstein, L. N. Pfeiffer, and K. W. [31] E. H. Hwang, E. Rossi, and S. Das Sarma Phys. Rev. B West, Phys. Rev. Lett., 80 1714 (1998). 80, 235415 (2009). [26] A. Kamenev and Y. Oreg, Phys. Rev. B, 52 7516 (1995). [27] M. C. Bonsager, K. Flensberg, B. Y.-K. Hu, and A.-P. Jauho, Phys. Rev. Lett., 77 1366 (1996). The authors note that Hall drag vanished for systems with inversion symmetry and when the carriers in the active layer can be described by a drifted Fermi-Dirac distribution. [28] M. W. Graham, S-F. Shi, D. C. Ralph, J. W. Park, P. L. [32] H. B. Callen, Phys. Rev., 73 1349 (1948). [33] M. Jonson and S. M. Girvin, Phys. Rev. B, 29 1939 (1984). [34] D. A. Abanin, et. al., Science, 332 328 (2011) and ac- companying online supplement. [35] A. K. Geim, private communication 

1005.4780

1

1005

Power laws in surface state LDOS oscillations near a step edge

In this paper we indicate a general method to calculate the power law that governs how electronic LDOS oscillations decay far away from a surface step edge (or any local linear barrier), in the energy range when only 2D surface states are relevant. We identify the critical aspects of the 2D surface state band structure that contribute to these decaying oscillations and illustrate our derived formula with actual examples.

cond-mat.mes-hall

cond-mat

Power laws in surface state LDOS oscillations near a step edge Rudro R. Biswas1,3∗ and Alexander V. Balatsky2,3 1Department of Physics, Harvard University, Cambridge, MA 02138 2Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM 87545 3Center for Integrated Nanotechnologies, Los Alamos National Laboratory, Los Alamos, NM 87545 (Dated: May 25, 2018) In this paper we indicate a general method to calculate the power law that governs how electronic LDOS oscillations decay far away from a surface step edge (or any local linear barrier), in the energy range when only 2D surface states are relevant. We identify the critical aspects of the 2D surface state band structure that contribute to these decaying oscillations and illustrate our derived formula with actual examples. PACS numbers: 03.65.Nk, 07.79.Cz, 73.20.-r, 73.20.At I. INTRODUCTION For over a couple of decades now Scanning-Tunneling Microscopy (STM) experiments have been used to ob- serve the effects of perturbations to electronic surface states, in the form of atomic defects, corrals and step edges1–4. The quantum electronic response to defects can give us basic information about the band structure of the scattered quasiparticles5; it also encodes informa- tion about their nature and this is useful when probing correlated phases2,6–8. In this paper, we shall consider the case of a step edge on a 2D surface and calculate the spatial decay of standing waves created in the surface LDOS, far away from the step edge. We shall show, by a simple process of power counting, that the geometry of the constant energy cut of the quasiparticle band struc- ture and a qualitative knowledge of the character of the quasiparticle wavefunctions provide enough information to pin down the power with which the LDOS oscillations decay far away from the step edge. II. THEORY We begin by considering a general band structure for the 2D surface states, whose cross-section at the energy of observation Eobs is shown in Figure 1. Also shown in the figure is the orientation of the surface step edge - parallel to the y-direction. Because of the conservation of momentum parallel to the edge during a scattering pro- cess, the incoming and outgoing states must be connected by straight lines perpendicular to the step edge. Some such processes are also marked in Figure 1. The arrows joining the initial and final states denote the wave-vector of LDOS oscillations that particular scattering process would give rise to. Obtaining the total LDOS involves summing up these oscillations. The most coherent con- tributions to this sum come from regions where the scat- tering wave-vectors change the slowest as we move par- allel to the step edge, i.e, changing only the ky of the scattering states. We denote the 'identifying' scattering wave vector in each such region as the 'characteristic' wave vector of that region. As a very common example, for a circular constant energy cut as in a 2DEG (Figure 2), the most coherent contributions come from the scatterings around the di- ameter - the characteristic wave vector in this case is thus the diameter, ∆0. The new electronic LDOS far away from the step edge is now provided by (below, 'new' refers to the new energy eigenstates while 'init' and 'fin' refer to the initial and FIG. 1: (Color online) A constant energy cut of a generic 2D electronic surface state (SS) band structure, taken at the energy Eobs of an STM probe; on the right is shown the ori- entation of the surface step edge (or any linear barrier) in question. Quasiparticles are scattered 'horizontally', preserv- ing ky. The regions where the scattering wave-vectors vary the slowest (locally) with ky are shaded – blue for incident and pink for reflected states. The 'characteristic' scattering wave vectors are also indicated by arrows and numbers. final scattering states, respectively): ψnew2 E (cid:88) (cid:88) (cid:122) (cid:88) E E ρ(x, E) = = = ψinit + rψfin2 + transmitted from other side x−independent part (cid:16)ψinit2 + rψfin2 + transmitted (cid:105) (cid:125)(cid:124) (cid:88) (cid:104) + 2 Re r ψ † init · ψfin (cid:123) (cid:17) E (1) Writing the energy-momentum eigenstates as ψk(x) = χeik·x, where χ denotes an 'internal' part involving the spin and other internal components, the x-dependent part of the LDOS can be summarized as δρ(x, E) ∝(cid:88)(cid:90) (cid:104) (cid:16) † f · χi χ (cid:17) ei∆kxx(cid:105) (2) dky ρ0(ky)Re r(ky) 0 The sum is over the various regions of coherent scatter- ing, each one corresponding to a characteristic scattering vector. The outer limits of these integrals are not impor- tant as the oscillations there de-cohere rapidly. ρ0(ky) is a DOS factor (it multiplicatively converts the measure dky to a product of the length of the band curve enclosed between ky and ky + dky and the DOS in that region). The x-dependence of a characteristic oscillation far away from the step edge may be found from the above expression by writing down the lowest order ky- dependencies of the relevant quantities near each charac- teristic wave-vector (δky is the ky-displacement from the associated characteristic wave vector): ρ0(ky) ∼ ρ0δkα r(ky) ∼ r0δkβ † f · χi ∼
δkγ χ ∆kx ∼ ∆0 + ∆1δkη (3) y y y y (cid:104) r0
ei(∆0x+∆1µ)(cid:105) α+β+γ+1 η Re Changing the integration variable δky to the variable µ = δkη y x in (2) and using (3), we obtain our central result δρ(x, E) ∝(cid:88) ρ0 (cid:90) dµ ∼(cid:88)ρ0r0
sin(∆0x + φ) α+β+γ+1 µ µ x η x(α+β+γ+1)/η (4) This asymptotic behavior is correct for x (cid:29) (∆k)−1, where ∆k is the characteristic size of the region in mo- mentum space where the scaling laws (3) hold. The power of decay of the oscillations coming from each char- acteristic region is thus given by (α + β + γ + 1)/η, which may be evaluated using our knowledge of the band struc- ture in that region. Note, however, that we haven't been able to evaluate the total strength of the scattering pro- cess which requires a much more detailed calculation in- cluding evaluating the reflection amplitudes themselves. This exercise provides us with the combination of possi- ble power laws we can try to fit actual experimental data to, if an idea of the band structure exists. 2 III. EXAMPLES A. 2DEG1 The scattering wave-vector varies slowest around the equator (see Figure 2). Thus, ∆0 = diameter of circle = 2kE, α = β = γ = 0 (assuming, quite reasonably, that the reflection amplitude is nonzero and smooth across normal scattering). Also, from the geometry of the band, we get η = 2. Using (4), this tells us that: δρ(x, E) ∼ sin(∆0x + φ) x1/2 (∆0 = 2kE) (5) FIG. 2: (Color online) The characteristic wave-vector ∆0 in the case of a circular band (for 2DEGs with rotational invari- ance) B. Strong Topological Insulator (circular band cut) 1. Generic barrier This case is illustrated in Figure 3 and is realized for the gapless surface states in Strong Topological Insula- tors like Bi2Se3 and Bi2Te3 (at energies near the Dirac point). The scattering wave-vector varies slowest around the equator (as in the 2DEG case), where ∆0=diameter of circle, α = 0, β = 1 since the reflection coefficient changes sign9 as one crosses the diameter/case of normal reflection (can be any odd power; should be linear gener- ically), γ = 1 because the spins are exactly antiparallel for scattering states at the diameter and thus the lowest order overlap is linear in δky, and η = 2 as in the 2DEG case. This gives rise to: δρ(x, E) ∼ sin(∆0x + φ) x3/2 (∆0 = 2kE) (6) This result agrees with numerical calculations for partic- ular cases of the model describing the step edge10. FIG. 3: (Color online) Constant energy cut of a circular sur- face band on a STI surface. The spins, indicated as block arrows, are antiparallel for normal scattering (the character- istic scattering process for the STI surface state band that is circular) – the spin overlap magnitude is thus generically a linear function of the angle of incidence. The same may be said for the reflection amplitude magnitude which is a certain gauge is antisymmetric in the angle of incidence. 2. 'Perfect' reflection The scattering wave-vector varies slowest around the equator (as in the 2DEG case), where ∆0=diameter of circle, α = β = 09 (since the reflection amplitude is con- stant in magnitude near normal incidence), γ = 1 as argued for the previous case, and η = 2. This gives rise to: δρ(x, E) ∼ sin(∆0x + φ) x (∆0 = 2kE) (7) In the actual case, there will always be a region near normal incidence where the reflection amplitude will be- come linear (because it is antisymmetric). This means that 'very' far away ∼ the inverse of the ky-span of the region where r is linear, the previous result (6) for the generic barrier should hold. C. Bi2Te3 (with hexagonal warping)3 If we are far away from the Dirac point, the surface band of Bi2Te3 exhibits hexagonal warping11. The fol- lowing results hold when the STM bias maintains our observation energy in that regime. 1. Step edge ⊥ ΓM direction The scattering wave-vector varies slowest around the equator (as in the 2DEG case), where ∆0=diameter of 3 circle, α = 0, β = 1, γ = 1 and η = 2 exactly as argued before for the circular STI band. However, the extent of this region is very small and the scattering is found to be dominated by processes connecting the hexagonal 'cor- ners' (marked by bold arrow in Figure 4), with a charac- 3. For the latter case, since teristic scattering vector knest the spin states have a finite overlap with each other at the hexagon corners11, we have γ = 0. Also, assuming that the reflection coefficient is smooth for the relevant scattering processes, β = 0. Finally, α = 0 (DOS is finite and smooth) and the overwhelmingly 'linear' nature of the bands yield η = 1. Putting these together, we obtain the observed variation3 δρ(x, E) ∼ sin(knestx + φ) x (8) FIG. 4: (Color online) Scattering processes from a step edge on Bi2Te3 oriented perpendicular to the ΓM direction, in the energy range where the band exhibits hexagonal warping. The scattering vector varies linearly near knest, as indicated by the angle made by the dotted lines, leading to η = 1. The weaker characteristic scattering process is denoted by the dotted ar- row. Spins are indicated as block arrows. 2. Step edge ⊥ ΓK direction The wave-vectors vary slowest around the equator (as in the 2DEG case) and from the considerations of the cicular STI surface band above, we can conclude that there should be characteristic oscillations at 2kΓK decay- ing as x−3/2 (or 1/x for a 'perfect' reflector). In this case, because of the larger extent of the characteristic scattering region around the diameter, these oscillations may be strong and observable. Because of the presence of the linear band shape with larger spectral presence and reflection strengths near the corners, we can also observe LDOS oscillations from the corner→corner scat- tering processes, decaying as 1/x. Of course, from our simple calculation we cannot reliably predict which of the two processes discussed above have the stronger sig- nature. 4 over the long-distance behavior) for features at the 'char- acteristic' scattering vectors. The FT near those points can then be fitted to the abovementioned power laws (or a logarithm, for the case of a 1/x decay) to recover the spatial decay power laws. V. CONCLUSION We have outlined a method to calculate the possible oscillatory power laws governing the decay of LDOS per- turbations next to a step edge or some such linear barrier on a surface, in the energy range when only electronic surface states are relevant and the surface band struc- ture is qualitatively known. To find these laws we need to identify the characteristic scattering regions (Figure 1), compute the scaling powers of the relevant quantities (3) and from that obtain the possible oscillatory powers (4). Acknowledgments This work was supported by the US DOE thorough BES and LDRD and by the University of California UCOP program T027-09. We would also like to acknowl- edge illuminating discussions with M. Crommie, D. Hal- dane, H. Manoharan and members of M. Z. Hasan and A. Yazdani's research groups. FIG. 5: (Color online) Scattering processes from a step edge on Bi2Te3 oriented perpendicular to the ΓK direction, in the energy range where the band exhibits hexagonal warping. Spins are indicated as block arrows. IV. ISOLATING CONTRIBUTIONS USING THE 1-D FOURIER TRANSFORM The Fourier Transform of the LDOS data may be used to observe signatures from more than one set of scattering processes. For oscillations decaying as sin(Kx + φ)/xn, scaling analysis tells us that the Fourier transform looks like F (k) ∼ k ∓ Kn−1, when k ∼ ±K. Thus, one way to look for contributions to these oscillations would be to scan the 1-D Fourier transform of the LDOS (taken ∗ Electronic address: [email protected] 1 M. F. Crommie, C. P. Lutz, and D. M. Eigler, Nature 363, 524 (1993). 2 A. Yazdani, B. A. Jones, C. P. Lutz, M. F. Crommie, and D. M. Eigler, Science 275, 1767 (1997). 3 Z. Alpichshev, J. G. Analytis, J.-H. Chu, I. R. Fisher, Y. L. Chen, Z. X. Shen, A. Fang, and A. Kapitulnik, Phys. Rev. Lett. 104, 016401 (2010). 4 A. Richardella, P. Roushan, S. Mack, B. Zhou, D. A. Huse, D. D. Awschalom, and A. Yazdani, Science 327, 665 (2010). 5 L. Petersen, P. T. Sprunger, P. Hofmann, E. Laegsgaard, B. G. Briner, M. Doering, H.-P. Rust, A. M. Bradshaw, F. Besenbacher, and E. W. Plummer, Phys. Rev. B 57, R6858 (1998). 6 J. E. Hoffman, E. W. Hudson, K. M. Lang, V. Madhavan, H. Eisaki, S. Uchida, and J. C. Davis, Science 295, 466 (2002). 7 J. E. Hoffman, K. McElroy, D. H. Lee, K. M. Lang, H. Eisaki, S. Uchida, and J. C. Davis, Science 297, 1148 (2002). 8 S. H. Pan, E. W. Hudson, K. M. Lang, H. Eisaki, S. Uchida, and J. C. Davis, Nature 403, 746 (2000). 9 R. R. Biswas and A. V. Balatsky, arXiv:0912.4477 (2009). 10 X. Zhou, C. Fang, W.-F. Tsai, and J. Hu, Phys. Rev. B 80, 245317 (2009). 11 L. Fu, Phys. Rev. Lett. 103, 266801 (2009). 

1104.3104

1

1104

Observation of Intra- and Inter-band Transitions in the Optical Response of Graphene

The optical conductivity of freely suspended graphene was examined under non-equilibrium conditions using femtosecond pump-probe spectroscopy. We observed a conductivity transient that varied strongly with the electronic temperature, exhibiting a crossover from enhanced to decreased absorbance with increasing pump fluence. The response arises from a combination of bleaching of the inter-band transitions by Pauli blocking and induced absorption from the intra-band transitions of the carriers. The latter dominates at low electronic temperature, but, despite an increase in Drude scattering rate, is overwhelmed by the former at high electronic temperature. The time-evolution of the optical conductivity in all regimes can described in terms of a time-varying electronic temperature.

cond-mat.mes-hall

cond-mat

Observation of Intra- and Inter-band Transitions in the Optical Response of Graphene Leandro M. Malard1, Kin Fai Mak1, A. H. Castro Neto2,3, N. M. R. Peres4, and Tony F. Heinz1 1Departments of Physics and Electrical Engineering, Columbia University, 538 West 120th Street, New York, New York 10027, USA 2Department of Physics, Boston University, 590 Commonwealth Avenue, Boston, Massachusetts 02215, USA 3Graphene Research Centre, National University of Singapore, 2 Science Drive 3, 117542, Singapore 4Departamento de Física e Centro de Física, Universidade do Minho, P-4710-057 Braga, Portugal (April 15, 2011) The optical conductivity of freely suspended graphene was examined under non-equilibrium conditions using femtosecond pump-probe spectroscopy. We observed a conductivity transient that varied strongly with the electronic temperature, exhibiting a crossover from enhanced to decreased absorbance with increasing pump fluence. The response arises from a combination of bleaching of the inter-band transitions by Pauli blocking and induced absorption from the intra-band transitions of the carriers. The latter dominates at low electronic temperature, but, despite an increase in Drude scattering rate, is overwhelmed by the former at high electronic temperature. The time-evolution of the optical conductivity in all regimes can described in terms of a time-varying electronic temperature. PACS: 78.67.Wj, 78.47.D-, 78.47.jb The optical properties of graphene have been the subject of much attention [1-3]. Interest in this topic is generated both by the insight that the optical response of graphene provides into the nature of its excited states and their interactions and by the importance of understanding graphene’s optical response for emerging photonic applications [2, 4]. The response of graphene to excitation by femtosecond laser pulses has attracted particular interest [5-11]. Probing dynamics on the femtosecond time scale provides information about electron-electron, electron- phonon, and phonon-phonon interactions and also has important implications for the behavior of recently developed ultrafast photonic devices [2, 4]. To date all ultrafast pump-probe measurements of graphene have observed an induced bleaching of the absorption under optical excitation [7-9]. This behavior arises from Pauli blocking of the strong inter-band optical transitions [7-9] and is expected for electronic temperatures sufficiently high to induced filling of the states. The optical response of graphene arises, however, from two fundamental processes: inter-band and intra-band optical transitions [3, 12-14]. Here we report observation of a dominant intra-band contribution to the transient optical response, with a corresponding enhanced absorption. We this effect identify through probing freely suspended graphene samples at relatively low pump fluence. With increasing pump fluence, the transient optical response exhibits a crossover from enhanced absorption to bleaching as Pauli blocking of inter-band transitions becomes more prominent. We explain the observed temporal evolution of the optical absorption for all pump fluences within the framework of a thermalized electronic energy distribution that is in equilibrium with a set of strongly coupled optical phonons. Cooling of this subsystem, and the decrease in electronic temperature, is controlled by energy loss of the strongly coupled optical phonons to lower energy phonons through anharmonic decay. A fluence-independent time constant of 1.4 ps is deduced, which is consistent with recent time-resolved Raman scattering measurements of the cooling of zone-center optical phonons [15]. In addition to the importance of these measurements for understanding ultrafast carrier dynamics and applications in ultrafast photonics, response at high carrier intra-band the temperature provides insight into carrier dynamics in a regime relevant to high-field charge transport in graphene [16-19]. In particular, our experiments reveal a sharp increase in the Drude scattering rate with increasing temperature of the electrons and optical phonons, as also suggested by high-field electrical transport measurements in carbon nanotubes [20, 21] and graphene [16-19]. Our results shown to be compatible with the behavior expected if electron-optical phonon scattering plays a dominant role in determining the carrier scattering rate in this high temperature regime. In our experimental study we made use of freely suspended graphene samples. Such samples allow us to eliminate both accidental doping effects and energy transfer associated with the substrate. We prepared exfoliated graphene samples that were freely suspended over trenches patterned in transparent SiO2 substrates [22]. The substrates, with trenches of width 4 – 5 µm and depth of ~3 µm (as characterized by atomic force microscopy), were carefully cleaned by chemical etching (nanostrip) before we deposited graphene by mechanical exfoliation from kish graphite. Areas of graphene of single-layer thickness were identified by optical microscopy and further characterized by Raman spectroscopy [22]. The low doping levels of the samples were reflected in the large (~15 cm-1) width of the G-mode and characteristic asymmetry of the 2D-mode, while the low level of defects was indicated by the absence of the disorder-induced D-mode. The optical pump-probe measurements were performed using a modelocked Ti-sapphire laser producing pulses of ~300-fs duration at an 80-MHz repetition rate. The 800-nm wavelength output of the laser provided the probe pulses, while radiation at 400 nm, obtained by frequency doubling in a β-barium borate (BBO) crystal, served as the pump excitation. Both the pump and probe beams were focused onto the samples with a single 40× objective to yield Gaussian spots of comparable widths of ~1.5 µm (FWHM). The absorbed pump fluence was determined using the absorbance of graphene at 400 nm (1.8πα = 4.14 % [23]). To account for the spatial variation of the pump and probe beams, the effective fluence was determined by weighting the absorbed probe fluence using the spatial 1 profile of the probe beam. Effective absorbed fluences F 4.0 µJ/cm2 were between 0.5 and investigated experimentally. We note that over this range of pump fluences, saturation of absorbance in graphene at the pump wavelength pump-probe [4]. The negligible is measurements were performed by modulating the pump laser at 20 kHz and detecting the synchronous change in probe transmission. The induced modulation of the probe beam for the lowest pump fluence was ~10-6. For a suspended thin film of material like our graphene sample, the fractional change in transmittance, , is given by the change in the sample absorbance . The later is proportional to the change in the real part of the optical , according sheet conductivity of graphene, to [14]. We can therefore directly convert our experimentally observed change in transmission into a fundamental material property of the graphene, namely, . Figure 1(a) shows the measured transient response of graphene for a comparatively low absorbed pump fluence of F = 0.5 µJ/cm2. An increase in the optical conductivity (enhanced absorption) is seen. The rise time of this response is comparable to our experimental time resolution, while the relaxation can be fit to a single exponential decay with a time constant of τexp = 3.1 ps. FIG. 1 (color online): Measured dynamics (closed circles) of the transient optical conductivity of graphene for excitation with different absorbed pump fluences F: (a) 0.50 µJ/cm2, (b) 2.03 µJ/cm2, (c) 3.04 µJ/cm2, and (d) 4.06 µJ/cm2. The red curves are fits based on the model described in the text, which include both intra- and inter-band contributions to the optical response. To analyze our data, we consider models for the optical conductivity of graphene with the electronic system described by a Fermi-Dirac distribution at (electronic) temperature T. The assumption of a thermalized energy distribution for the electronic excitations is justified by the time resolution of >100 fs in our measurements. On this time scale, the excited electrons and holes are equilibrated with one another, as well with the strongly coupled optical phonons (near the Γ- and K-points) [5, 24]. The predicted change in the optical conductivity of graphene under excitation arises from both intra- and inter- band contributions, . For the case of thermalized electronic energy distributions relevant in our case, the optical response of graphene has been examined in several theoretical investigations [3, 12, 13]. Within a picture of non-interacting electrons, the induced change in the real part of the optical sheet conductivity follows directly from the corresponding changes in the intra- and inter-band terms: (1) Here denotes the photon energy of the probe beam, and Γ is the carrier scattering rate. For our suspended graphene samples, we have negligible doping and take the chemical potential to be at the Dirac point. We also neglect any transient separation of the chemical potentials for the electrons and holes. While a dynamical separation of the chemical potentials is common in semiconductors under ultrafast excitation, on the relevant time scale the effect is expected to be insignificant in graphene because of the presence of rapid Auger processes for this zero-gap material [4, 11]. The photo-induced the optical in change conductivity in Eqn. (1) from intra-band optical transitions, , is positive in sign. The enhanced conductivity arises from the presence of additional free carriers after optical excitation and scales linearly with the electronic temperature T. The inter-band term yields a negative contribution to the conductivity, a bleaching of the absorption, from Pauli blocking of the strong inter-band transitions at the photon energy . In our regime of >> kBT , the fractional change in the inter- band conductivity is slight and varies nearly exponentially in T. Accordingly, the intra-band contribution dominates at comparatively low temperatures (low fluence), but, as shown in Fig. 2, is overwhelmed by the inter-band terms as the temperature increases. Thus the transient increase in absorption seen in Fig. 1(a) for low pump fluence reflects the change in intra-band optical response. In order to apply the analysis of graphene optical response presented in Eqn. (1) to our experiment, we introduce a simple model to describe the temporal evolution of the electron temperature T(t) induced by the pump laser. As we shall see, this treatment predicts transient optical conductivities compatible both with the low-fluence data already presented and the high-fluence response discussed below. To describe T(t), we first note that on the time scale of our measurement, the deposited energy from the pump pulses rapidly equilibrates among the electronic excitations and the strongly coupled optical phonons [5, 24]. Energy then leaves this coupled subsystem through anharmonic decay of the phonons on the picosecond time scale, which 2 we describe phenomenologically by an effective phonon lifetime τph [25]. To determine the resulting electronic temperature T(t) we need to know the heat capacity of the sub-system consisting of the electrons and the strongly coupled phonons. For this purpose, we use an Einstein model consisting of two phonon modes, the Γ- and K-point phonons. Since only a small part of the phonon branch gets populated, we considered only a fraction of the graphene Brillouin zone to obtain the phonon heat capacity. The fraction is determined by comparing results from [24] and the initial condition is fixed by the absorbed fluence F. (For details of this analysis, please refer to the Auxiliary Materials [26].) The electronic contribution to the heat capacity is minor and is neglected. Using this heat capacity, we the find T(t) for any given excitation condition and, using Eqn. (1), then obtain the temporal evolution of the optical conductivity. In fitting our results, we treating the phonon lifetime τph and the Drude scattering rate Γ as adjustable parameters. As we discuss below, we use an effective scattering rate Γ that depends on the pump fluence, but is independent of time, to account for the variation of the scattering rate with the temperature of the electrons and optical phonons. Fig. 2 (color online): Change in the optical conductivity of graphene as a function of electronic temperature calculated from Eq. (1). Black and red curves show the expected contributions from intra- and inter-band terms, respectively. The blue curve shows the total response. The measured the optical time evolution of conductivity in Fig. 1(a) is reproduced well with this model. We use an effective scattering rate of Γ = 27 meV and a phonon lifetime of τph =1.4 ps. This lifetime differs significantly from time constant (τexp = 3.1 ps) of the conductivity transient, a difference that reflects the strong temperature dependence of the phonon heat capacity: The electronic temperature, which tracks that of the strongly coupled optical phonons and is described by τexp, falls more slowly than the energy content of the optical phonons, which is characterized by τph. The inferred phonon lifetime of τph =1.4 ps, we note, lies between that obtained by time- resolved Raman scattering measurements for graphite (2.2 ps) [24] and for graphene (1.2 ps) [15]. The slightly longer phonon lifetime compared to that measured for graphene [15] is attributed to the fact that our graphene samples are suspended, thus eliminating possible decay channels involving the generation with surface polar phonons in the substrate [16, 17, 19]. At higher pump fluences (and, correspondingly, higher electronic temperatures), the optical conductivity is expected to decrease as state filling of the inter-band transitions begins to dominate the transient response (Fig. 2). The measured conductivity transients at higher pump fluence are shown in Fig. 1 (b-d). With increasing pump fluence, the initial response is indeed negative. At later times when the electronic temperature drops, the intra-band contribution becomes dominant and a positive conductivity change is observed. The experimental data of Fig. 1(b-d) can be fit (red curves) with the same model of the electronic temperature described above, but now with the inclusion of the inter- band optical response of graphene. Thus model can then accurately describe the dynamics of the conductivity transients at all fluences [Fig. 1(a-d)]. In this fitting process, we use a single, fluence-independent parameter of τph =1.4 ps for the optical phonon lifetime. This parameter reflects the anharmonic decay rate of these optical phonons. It should not increase with pump fluence unless the excitation of the resulting phonon decay modes also increased significantly [27], which is not expected this regime. On the other hand, the fitting procedure implies that the effective carrier scattering rate Γ increases with pump fluence. The inferred variation of Γ is shown in Fig. 3 as a function of absorbed fluence (bottom scale) and of the peak electronic temperature (top scale) in our model. At high fluence, corresponding to a peak temperature of 1,700 K, we find Γ = 52 meV. Both this rate and that inferred for lower fluences are significantly higher than the scattering rates implied by conventional transport measurements [28] obtained for a more modest temperature range. To understand the origin of the enhanced scattering rate, recall that the electron-phonon coupling in graphene is especially strong for the optical phonon modes located at the center and edge of the Brillouin zone [5, 25]. This coupling is not, however, important for conventional low-bias charge transport measurements at room temperature. In this conventional regime, the optical phonons are not thermally excited and cannot be absorbed in a scattering event, while the electron energy is too low to scatter through the emission of optical phonons. This situation changes when the temperature of the electrons and optical phonons becomes comparable to that characterizing the optical phonon, [20, 21]. Electron scattering with optical phonons is then allowed and the contribution of this scattering process to the temperature dependence of the Drude is significant [29]. To obtain specific predictions for the increased Drude scattering rate Γ from electron – optical phonon interactions, we consider the zone-center and zone-edge phonons to be dispersionless branches with energies of 200 meV and 150 meV, respectively. We then can then calculate (see Auxiliary Materials [26]), the expected temperature dependence of Γ from electron- optical phonon scattering. As shown in Fig. 3, this contribution increases nearly linearly with T for the relevant temperatures. By using published values for the electron-phonon coupling for both the zone-center [30] and zone-edge [31] optical phonons, we obtain semi-quantitative agreement with the experimental results for Γ as a function of pump fluence. Better agreement is not anticipated, since the treatment involves several significant simplifications. We first note that the deduced effective Drude scattering rate Γ actually 3 corresponds to fitting the response over a range of electronic temperatures. Here we assume that the result is dominated by the behavior near the peak electronic temperature. More elaborate modeling could take this effect into account. In terms of the optical the underlying description of conductivity at high temperature, our treatment could be improved by a more detailed analysis of the different phonon modes, considered here as two dispersionless the scattering process. More in branches, involved examine possible should fundamentally, one also contributions to Γ from electron-electron scattering processes [32], which are known to influence the dc conductivity [33]. Finally, at high electronic temperatures (T > 1600K), the experimental values for Γ level off with increasing T. This effect is absent in the model and suggests that screening of the electron interactions [34] may play a significant role in this regime of high carrier densities. Fig. 3 (color online): Drude scattering rate (closed circles) as a function of absorbed pump fluence (bottom scale) and the inferred peak electronic temperature (top scale). The points are inferred from modeling of the experimental transient optical response of the graphene sample. The black line corresponds to the predicted temperature dependence of arising from electron- optical phonon interactions, as described in the text and in the Auxiliary Materials [26]. An interesting point of comparison for our results is with those obtained in high-field dc transport studies. We find that the values for Γ deduced here for high electronic temperatures are also broadly consistent with those obtained in high-field transport measurements in metallic carbon nanotubes at current saturation [20, 21]. In this regime, electron- optical phonon scattering is also considered to be the dominant contribution to Γ. For high-bias transport measurements in graphene [16, 17, 19], the carrier scattering rate is also significantly enhanced compared to the low-field behavior. This effect has, however, been attributed to interaction with of the carriers with polar phonons in the substrate on which the graphene is deposited. Our results imply that even for suspended graphene, which lacks these substrate-mediated interactions, the scattering rate will be strongly enhanced at elevated electronic temperatures. This effect will places a fundamental limit on current flow in graphene at electrical high bias. In conclusion, we have examined ultrafast carrier dynamics in freely suspended graphene samples by optical pump-probe measurements. A crossover of the graphene optical conductivity transients from enhanced to decreased absorption is observed as the pump fluence is increased. This behavior can be understood as the result of the existence of both intra- and inter-band contributions to the optical response of graphene. Analysis of the data implies an optical phonon lifetime of 1.4 ps and enhanced carrier scattering rates at high electronic temperatures. Our experiment demonstrates the importance of free carrier absorption in the visible spectral range for graphene under non-equilibrium conditions and opens up new opportunities for probing fundamental charge transport properties of this 2-dimensional system by means of optical measurements. We thank Drs. Y. Wu, M. Koshino, D. Song and H. Yan for fruitful discussions and C. H. Lui for sample preparation. We acknowledge support from the Nanoscale Science and Engineering Initiative of the National Science Foundation and from the MURI program of the Air Force Office of Scientific Research. AHCN acknowledges the partial support of the U.S. DOE under grant DE-FG02- 08ER46512, and ONR grant MURI N00014-09-1-1063. LM acknowledges support from the CNPq program in Brazil. [1] A. H. Castro Neto et al., Rev. Mod. Phys. 81, 109 (2009). [2] F. Bonaccorso et al., Nat. Photonics 4, 611 (2010). [3] N. M. R. Peres, Rev. Mod. Phys. 82, 2673 (2010). [4] Z. P. Sun et al., ACS Nano 4, 803 (2010). [5] T. Kampfrath et al., Phys. Rev. Lett. 95, 187403 (2005). [6] M. Breusing, C. Ropers, and T. Elsaesser, Phys. Rev. Lett. 102, 086809 (2009). [7] P. A. George et al., Nano Lett. 8, 4248 (2008). [8] D. Sun et al., Phys. Rev. Lett. 101, 157402 (2008). [9] R. W. Newson et al., Opt. Express 17, 2326 (2009). [10] C. H. Lui et al., Phys. Rev. Lett. 105, 127404 (2010). [11] T. Winzer, A. Knorr, and E. Malic, Nano Lett. 10, 4839 (2010). [12] N. M. R. Peres, F. Guinea, and A. H. Castro Neto, Phys. Rev. B 73, 125411 (2006). [13] L. A. Falkovsky, and A. A. Varlamov, Eur. Phys. J. B 56, 281 (2007). [14] K. F. Mak et al., Phys. Rev. Lett. 101, 196405 (2008). [15] K. Kang et al., Phys. Rev. B 81, 165405 (2010). [16] I. Meric et al., Nat. Nano. 3, 654 (2008). [17] M. Freitag et al., Nano Lett. 9, 1883 (2009). [18] A. Barreiro et al., Phys. Rev. Lett. 103, 076601 (2009). [19] A. M. DaSilva et al., Phys. Rev. Lett. 104, 236601 (2010). [20] J. Y. Park et al., Nano Lett. 4, 517 (2004). [21] E. Pop et al., Phys. Rev. Lett. 95, 155505 (2005). [22] S. Berciaud et al., Nano Lett. 9, 346 (2009). [23] K. F. Mak, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 106, 046401 (2011). [24] H. Yan et al., Phys. Rev. B 80, 121403 (2009). [25] N. Bonini et al., Phys. Rev. Lett. 99, 176802 (2007). [26] See EPAPS Document for more information. [27] I. Chatzakis et al., submitted (2011). [28] K. I. Bolotin et al., Phys. Rev. Lett. 101, 096802 (2008). [29] T. Stauber, N. M. R. Peres, and A. H. Castro Neto, Phys. Rev. B 78, 085418 (2008). [30] A. H. C. Neto, and F. Guinea, Phys. Rev. B 75, 045404 (2007). [31] M. Lazzeri et al., Phys. Rev. B 78, 081406 (2008). [32] E. H. Hwang, B. Y. K. Hu, and S. Das Sarma, Phys. Rev. B 76, 115434 (2007). [33] A. A. Kozikov et al., Phys. Rev. B 82, 075424 (2010). [34] E. H. Hwang, and S. Das Sarma, Phys. Rev. B 79, 165404 (2009). 4  Auxiliary Material Observation of Intra- and Inter-band Transitions in the Optical Response of Graphene Leandro M. Malard1, Kin Fai Mak1, A. H. Castro Neto2,3, N. M. R. Peres4, and Tony F. Heinz1 1Departments of Physics and Electrical Engineering, Columbia University, 538 West 120th Street, New York, New York 10027, USA 2Department of Physics, Boston University, 590 Commonwealth Avenue, Boston, Massachusetts 02215, USA 3Graphene Research Centre, National University of Singapore, 2 Science Drive 3, 117542, Singapore 4Departamento de Física e Centro de Física, Universidade do Minho, P-4710-057 Braga, Portugal 1) Determination of the electronic temperature Here we describe the model employed for the heat capacity of the strongly coupled optical phonons. These excitations are assumed to be in thermal equilibrium with the electronic excitations, but, because of the very low heat capacity of electronic excitations in graphene, the heat capacity is dominated by the phonons. Knowledge of this heat capacity then allows us to convert the experimental value of the absorbed pump fluence into the electronic temperature of this subsystem. With the temperature in hand, we predict the optical response from Eq. (1) of the main text to calculate for a comparison with the experimental data. Strong electronic coupling is present for the zone-center phonons (G band, energy of 200 meV) and the zone-edge phonons (D band, energy of 150 med) [A1, A2, A3]. Using the Einstein model for heat capacity including these two phonon modes, we obtain for the energy density (per unit area) as a function of temperature [A4]: 5  . (E1) Here A is the Brillouin zone area, f is the fraction of Brillouin zone filled by the G and D optical phonons and denotes the G or D phonon energy. We determined f in Eq. (E1) by comparison with a previous experiment in which transient phonon temperatures for femtosecond laser excitation of graphite were determined by time- resolved Raman scattering [A5, A6]. Fitting these earlier measurements to a polynomial yields the following relation between the absorbed fluence (or energy density) per graphene layer and the temperature: . (E2) We find the fraction f by comparing Eq. (E1) and (E2). Eq. (E1) then yields the energy density as a function of temperature, even for temperatures below the range for which the experimental data from the earlier experiments are available. To model the temporal evolution of the sub-system of the electronic excitations and the strongly coupled phonons, we assume that the energy relaxes as a rate of 1/τph through the anharmonic decay, with Eq. (E1) yielding the variation of the temperature. This temperature dynamics is then insert in Eq. (1) from the main text of the paper to fit our pump-probe dynamics experiment. Because of the non-linear relationship between the energy density and temperature, the decay time of these two quantities are not the same, as can see in Fig. S1. 6  Figure S1: (color online) Temporal evolution of the energy density (black) and temperature (red) calculated for the case of an absorbed fluence of 0.5 . The decay times for the energy density and temperature are, respectively, 1.4 and 3.1 ps. 2) Calculation of the electronic scattering rate in graphene at high temperatures The intra-band contribution to the optical conductivity scales with the electron scattering rate Γ. For the regime of temperature of the electrons and strongly coupled optical phonons, we expect that scattering of the electrons by these phonons will contribute significantly to the overall scattering rate Γ. Here we calculate this contributions and find that it yields a rate similar to that inferred from the experiment. As above, we model the strongly coupled phonons by two branches of dispersionless optical phonons, the 200 meV zone-center and 150 meV zone-edge phonons. 7 The self-energy of the electrons in graphene due to optical phonons is given by [A7]: 1 1 9 2 N c β  Q , iωn − iνm ) , [E3] 0 (  k − D0 (iνm ) Gbb ∑ m Σ(  k , iωn ) /  = − ∂t ⎛ ⎜ ∂a ⎝ 2 ⎞ ⎟ ⎠  M cω0 ∑ Q→ where is the electron-phonon coupling as defined in Ref. [A8], is the € carbon-carbon distance, is the hopping in the tight-binding model for graphene, MC is the carbon mass, is the phonon frequency. All the other parameters are equally defined by Ref. [A7]. After performing the Matsubara summation, the imaginary part of the self-energy is given by 1 9π 2 ∂t ℑΣ(  k , iωn ) /  = − ⎛ ⎞  ∑ ⎜ ⎟ 4 N c M cω0 ∂a ⎝ ⎠  Q ] δ(ε k −ε k − [  Q + ω0 ) + n B (ω0 ) + n F (ε k −  Q ) ] δ(ε k −ε k − [ + n B (ω0 ) + 1 − n F (ε k −  Q )  Q − ω0 ) [ ] δ(ε k +ε k −  Q ) + n B (ω0 ) + n F (−ε k −  Q + ω0 ) [ ] δ(ε k +ε k −  Q − ω0 ),  Q ) + n B (ω0 ) + 1 − n F (−ε k − where nB and nF are the Bose and Fermi distribution functions, [E3] and € . The two first terms are intra-band scattering processes and the last two are inter-band ones. The scattering rate is obtained from . [E4] To compute , we need to solve the following integrals appearing in Eq. [E3]: 8 [E5] where . The -functions in Eq. [E3] guarantee that we can replace in the Fermi functions with for the first and second integral, and with in the fourth one. The third integral does not contribute to the scattering rate, since its argument is always positive. Computing the three integrals, we find that the scattering rate of an electron with momentum has, as a consequence of phase- space restrictions, different forms for different energy ranges: 1. For : 2. For : where C is given by [E6] [E7] [E8] 9 and . The electron phonon couplings used for our calculations are given by [A8] and [A9] for the zone-center and zone-edge phonons, respectively. Now we need to average the energy dependent scattering rate given by Eq. [E6-E7] to find the average scattering rate [A10]: [E9] where is the density of states of graphene. Finally to compute the results for the temperature dependence of the scattering rate shown in Fig. 3 of the main manuscript, we summed the contributions from the two optical phonon branches: [E10] References: [A1] – S. Piscanec et al., Phys. Rev. Lett. 93, (2004). [A2] – M. Lazzeri and F. Mauri, Phys. Rev. B 73, 165419 (2006). [A3] – N. Bonini et al., Phys. Rev. Lett. 99, 176802 (2007). [A4] – N. W. Ashcroft and N. D. Mermin, Solid State Physics (Thomson, 1976). [A5] – H. Yan et al., Phys. Rev. B 80, 121403R (2009). [A6] – C. H. Lui et al., Phys. Rev. Lett. 105, 127404 (2010). [A7] – T. Stauber, N. M. R. Peres and A. H. Castro Neto, Phys. Rev. B 78, 085418 (2008). [A8] – A. H. Castro Neto and F. Guinea, Phys. Rev. B 75, 045404 (2007). [A9] – M. Lazzeri et al., Phys Rev. B. 78, 081406R (2008). [A10] – Paul Harrison, Quantum Wells, Wires and Dots: theoretical and computational physics of semiconductor nanostructures (Wiley, 2005). 10 

1508.04433

2

1508

Helical Quantum Edge Gears in 2D Topological Insulators

We show that two-terminal transport can measure the Luttinger liquid (LL) parameter $K$, in helical LLs at the edges of two dimensional topological insulators (TIs) with Rashba spin-orbit coupling. We consider a Coulomb drag geometry with two coplanar TIs and short-ranged spin-flip inter-edge scattering. Current injected into one edge loop induces circulation in the second, which floats without leads. In the low-temperature ($T \rightarrow 0$) perfect drag regime, the conductance is $(e^2/h)(2 K + 1)/(K + 1)$. At higher $T$ we predict a conductivity $\sim T^{-4K+3}$. The conductivity for a single edge is also computed.

cond-mat.mes-hall

cond-mat

Helical Quantum Edge Gears in 2D Topological Insulators Yang-Zhi Chou,1, ∗ Alex Levchenko,2 and Matthew S. Foster1, 3 1Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA 2Department of Physics, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA 3Rice Center for Quantum Materials, Rice University, Houston, Texas 77005, USA (Dated: September 11, 2018) We show that two-terminal transport can measure the Luttinger liquid (LL) parameter K, in helical LLs at the edges of two-dimensional topological insulators (TIs) with Rashba spin-orbit coupling. We consider a Coulomb drag geometry with two coplanar TIs and short-ranged spin-flip interedge scattering. Current injected into one edge loop induces circulation in the second, which floats without leads. In the low-temperature (T → 0) perfect drag regime, the conductance is (e2/h)(2K + 1)/(K + 1). At higher T we predict a conductivity ∼ T −4K+3. The conductivity for a single edge is also computed. PACS numbers: 71.10.Pm, 72.15.Nj, 74.25.F- The edge states that encircle two-dimensional (2D) topological insulators (TIs) realize a novel electronic heli- cal Luttinger liquid (HLL) phase [1 -- 3]. Distinct from an ordinary one-dimensional (1D) quantum wire and from a quantum Hall edge, a helical edge consists of two coun- terpropagating modes forming a Kramers pair. The left- and right-moving channels interact through Coulomb re- pulsion, but time reversal symmetry protects the edge from the opening of a gap and from Anderson localization due to impurities. The combination of topological pro- tection and electron correlations implies that a TI edge is an ideal Luttinger liquid at low temperatures [4, 5]. Experimental evidence for helical edge states in HgTe [6] and InAs/GaSb [7] includes a quantized conductance G ≃ 2e2/h [7, 8]. In the absence of electrical contacts and magnetic fields, a HLL forms a closed, unbreakable loop. This topology of the edge has so far received little atten- tion. In this Letter, we propose a TI device geome- try in which edge loops rotate as interlocking "gears" through Coulomb drag [9 -- 13]. Our main result is that the strength of electron correlations encoded in the Lut- tinger parameter can be directly obtained in such a device using a two-terminal dc conductance measurement. Correlations are generically strong in 1D electron fluids because two particles cannot exchange positions without scattering or tunneling. These correlations are encoded in the Luttinger parameter K [14]. Measuring K in a nontopological 1D electronic system (or "wire") is pos- sible but delicate. For instance, the zero-temperature (T = 0) dc transport through a perfectly clean wire gives a quantized conductance independent of K [15 -- 17]. In a long wire, disorder tends to induce Anderson insulating behavior. At temperatures T & vkF /kB, inelastic scat- tering due to irrelevant umklapp interactions gives a con- ductivity that depends on T through a power law [18]; here v and kF , respectively, denote the charge velocity and Fermi wave vector. The disorder-induced scattering may lead to a qualitatively similar effect [19]. The tem- perature exponent in conductance can reveal the Lut- tinger parameter K, but a large temperature range is needed to fit the data. The tunneling zero bias anomaly is also predicted to encode K, but measurements often contain contributions from other mechanisms [20]. In the simplest version of HLL physics that realizes the quantum spin Hall effect [4, 5, 21, 22], the z component of spin is assumed to be conserved in a TI. As a result, the edge electrons carry well-defined Sz currents. When 0 I' 1 I 1 I 2 V FIG. 1. Using helical quantum edge gears to measure the Luttinger parameter. We consider Z2 TI edge states in two adjacent topological regions. The blue and red arrows indi- cate the propagation directions of edge electrons with oppo- site helicities. The left TI is connected to external leads; I1 and I ′ 1 denote the currents of the edges connected to these. The right TI edge floats as an electrically isolated closed loop. Rashba spin-orbit coupling [1] enables Coulomb drag due to short-ranged spin-flip scattering [23] between the adjacent edges. This induces a current I2 that circulates in the right edge. In the case of identical TIs with an interacting edge region of size L → ∞, at zero temperature strong backscat- tering "locks" the currents I1 = I2, associated to perfect drag [10, 12]. We then predict that the zero-temperature conduc- 1)/V = (e2/h)(2K + 1)/(K + 1), where K tance is G = (I1 + I ′ is the Luttinger parameter. In a real system of finite length L ≫ ξ and at temperatures T satisfying v/L . kBT ≪ ∆ [11] with ξ = v/∆ and ∆ the Mott gap of the antisymmetric mode, the result for G holds up to terms exponentially small in L/ξ and ∆/kBT [10 -- 12]. Here v is the charge velocity. Rashba spin-orbit coupling (SOC) is present [1] (generi- cally expected in the absence of inversion symmetry), Sz symmetry is sabotaged. New spin-flip interactions [23 -- 25] are then allowed on TI edges. We show that the Luttinger parameter enters the con- ductance in a Coulomb drag geometry consisting of two coplanar TI regions with Rashba SOC. Over a segment of length L, proximate HLL edge states are separated by a gap narrow enough to allow short-ranged Coulomb scat- tering but wide enough to prevent tunneling. In Fig. 1, we consider two identical helical edges. Current I1 is in- jected by external leads. Short-ranged spin-flip scatter- ing [23] between edges induces a current I2 in the right TI edge loop, which floats without leads. At zero tempera- ture, the two proximate edge segments develop a locking state of perfect drag (I1 = I2) [10, 12] for an infinitely long interacting region L → ∞. An additional current I′1 flows in parallel between the contacts. The zero- temperature two-terminal conductance G = (I1 + I′1)/V is G = e2 h h1 + (1 + 1/K)−1i = e2 h (cid:18) 2K + 1 K + 1 (cid:19) , (1) where (1 + 1/K) is the dimensionless resistance of the locked edges, as explained below. For a finite locking length L ≫ ξ and at temperatures T satisfying v/L . kBT ≪ ∆ [11], Eq. (1) holds up to exponentially small corrections in L/ξ and ∆/kBT [10 -- 12]. Here ξ ≡ v/∆ is the length scale associated to the gapped "antilocking" mode with I1 = −I2; ∆ is the energy gap. We also discuss dissipative finite-temperature trans- port in this geometry. In contrast to the usual setup for Coulomb drag [9, 13], the system is naturally character- ized in terms of conductances or conductivities: I2 (cid:21)=(cid:20) G11 G12 (cid:20) I1 G21 G22 (cid:21)(cid:20) V1 V2 (cid:21) , Gij = σij /L, where the labels 1 and 2 indicate the active and passive systems respectively. For our TI edges, the passive sys- tem is a closed HLL loop with V2 = 0, I1 = σ11V1/L, and I2 = σ21V1/L. We compute the intraedge and transcon- ductivities using the Kubo formula and bosonization, em- ploying the effective potential formalism [26, 27]. Both σ11 and σ21 give T −4K+3 (T −4K+2) behavior in the ab- sence (presence) of disorder, above the locking transition. Finally, we compute the conductivity of a single edge due to the least irrelevant symmetry-allowed (one- particle umklapp) interaction term. We find asymptotic T −2K−1 (T −2K−2) behavior in the high- (low-) T limits, in the presence of disorder, consistent with [28], and we also obtain the full result for the clean limit. Power-law scaling of conductance as a function of temperature and bias voltage that may be attributable to Luttinger liquid physics was recently observed in InAs/GaSb quantum spin Hall devices [29]. Model. -- The edge states of a 2D TI can be expressed in terms of right (R) and left (L) mover fermion fields. The kinetic term is 2 (2) H0 = −ivF Z dx(cid:2)R†(x)∂xR(x) − L†(x)∂xL(x)(cid:3) , where vF is the Fermi velocity of the edge band. Time- reversal symmetry is encoded by R(x) → L(x), L(x) → −R(x), and i → −i. Left and right movers interact via intraedge Coulomb repulsion. We focus on the coplanar geometry in Fig. 1 and consider the backscattering components of the interedge Coulomb interaction. An additional interedge Luttinger interaction does not modify our results for the locking regime if the distal portion of edge loop 2 is much longer than the interacting segment of length L; otherwise the parameter K in Eq. (1) encodes a combination of inter- and intraedge correlations. In the presence of Rashba SOC, the following interedge backscattering terms are allowed by symmetry [23]: H− = U−Z dxhei2(kF 1−kF 2)xL†1R1R†2L2 + H.c.i , H+ = U+Z dxhei2(kF 1+kF 2)xL†1R1L†2R2 + H.c.i , (3) (4) where kF 1 (kF 2) indicates the Fermi momentum in the first (second) edge. These are defined relative to an edge Dirac point, which is a commensurate (time-reversal in- variant) momentum [2]. The U− interaction describes normal backscattering, while U+ is a two-particle umk- lapp interaction. Additional one-particle umklapp inter- action terms are also allowed, HU = Xa=1,2 UaZ dxhe−i2kF a xR†aLaR†¯aR¯a − ei2kF a xL†aRaL†¯aL¯a + H.c.i, (5) where a is the index of the edge, ¯1 = 2, and ¯2 = 1. It is worth mentioning that all of these interactions are disallowed in the presence of Sz conservation (in each edge) [23, 30]. For simplicity, we assume the two HLLs are identi- cal, so that kF 1 = kF 2 ≡ kF and U1 = U2 ≡ U . The dominant interedge interaction at T = 0 is the nonumk- lapp backscattering H−; the others are irrelevant at long In order to include Lut- wavelengths for kF 6= 0 [14]. tinger liquid effects, we use bosonization [14, 31]. The individual edge loop HLLs are described by Hb,0 = v 2 Xa=1,2Z dx(cid:20)K (∂xφa)2 + 1 K (∂xθa)2(cid:21) , (6) where K is the Luttinger parameter and v is the charge velocity. K = 1 and v = vF corresponds to the free fermion limit. The density (n) and current (I) can be expressed in terms of the axial fields as na = ∂xθa/√π and Ia = −∂tθa/√π, respectively. The interedge interac- tion H− is bosonized to Hb,− = U− 2π2α2 Z dx cosh√4π (θ1 − θ2)i , (7) where α is an ultraviolet length scale. Perfect current drag and dc conductance. -- At zero temperature, two infinite HLLs form an interedge lock- ing state [10, 12] due to the two-particle backscatter- ing term in Eq. (7). The locking state is characterized by θ1(t, x) = θ2(t, x) + cm, where cm = (m + 1/2)√π is a constant and m ∈ Z. This state exhibits per- fect current drag [10], I1 = I2 in Fig. 1. The conduc- tance of the locked edges (both carrying current I1) is I1/V = (e2/h)[K/(K + 1)]. This can be understood as the series resistor combination of a spinless LL connected to leads with resistance h/e2 [15 -- 17] and one with peri- odic boundary conditions and resistance h/Ke2 [32, 33]. An explicit Green's function calculation confirms this re- sult [34], which is also independent of disorder. Equation (1) is obtained by adding the parallel I′1 edge channel. For a finite interacting region of length L and nonzero temperature T , we require that L ≫ ξ and kBT ≪ ∆. Occasional phase slips between the drive and slave cir- cuits give rise to corrections that are exponentially small in L/ξ and ∆/kBT [10 -- 12]. For L = 1 µm in InAs/GaSb with v ∼ vF = 3 × 104 m/s, this gives a lower bound for ∆ of the order of v/L = 0.02 meV. We assume that kBT is larger than the latter to avoid coherent in- stanton effects [11]. By comparison, the bulk minigap is of the order of 4 meV [7]. The Mott gap takes the form [14] ∆ ∼ pK U−v /α. Using α = 1 nm gives ∆ ∼ 20 meVpK (U−/v). The interaction strength U− is obtained from the inter-edge Coulomb potential, medi- ated by matrix elements determined by the Rashba SOC in each TI (since it vanishes in its absence). The result will depend on microscopic details that we do not analyze here. Finite temperature corrections. -- Above a crossover temperature T ∗ ∼ ∆/kB[12], inelastic electron-electron collisions due to the interedge interactions in Eqs. (3) -- (5) can be treated perturbatively. In addition, we consider intraedge collisions due to electron-electron interactions [Eq. (14), below] and forward-scattering potential disor- der. Ordinary backscattering (random mass) disorder is forbidden by time-reversal symmetry. We ignore irrel- evant backscattering disorder terms with extra deriva- tives that are not expected to impact the conductivity in isolation [28] and which give subleading corrections in combination with interactions. Forward-scattering disor- der is encoded in Himp =Pa=1,2R dx ηa(x) na(x), where ηa(x) is a random potential obeying ηa(x) = 0 and ηa(x)ηb(x′) = gη δa,bδ(x − x′). The ··· denotes disorder averaging, while gη characterizes the disorder strength. 3 To compute the conductivity, we evaluate interaction corrections to the inverse boson propagator via the effec- tive potential method [26, 27]. We use replicas to average over disorder. The retarded boson correlation function is ab , (8) =h G(R)(ω, k)i−1 ab −h Π(R)(ω, k)iab h G(R)(ω, k)i−1 where a, b ∈ {1, 2} indicate the edges. The noninter- acting propagator is G(R)(ω, k), while Π(R) denotes the self-energy describing the interaction corrections. Equa- tion (8) is a matrix Dyson equation. At second order in the coupling constants, Π(R) contains an imaginary part that determines the scattering rates; the real part does not contribute to dc conductivity. In the limit ω → 0 with k = 0, ImhΠ(R) ab i = −2ωΞab + O(cid:0)ω2(cid:1) , (9) 2 Ξ− + ΞU + ΞW and Ξ12 = Ξ21 = 1 where Ξab is the "rate" (inverse mean free path) as- sociated to Πab. The components are Ξ11 = Ξ22 = 2 Ξ+ + 1 1 2 Ξ−. ΞU is due to the one-particle backscattering term HU . Ξ+ and Ξ− correspond to the two-particle backscatter- ing interactions H+ and H−, respectively. ΞW is due to intraedge inelastic electron-electron collisions, HW in Eq. (14). ΞU and ΞW affect only the diagonal elements of the self-energy, while Ξ+ and Ξ− contribute to all of the components. 2 Ξ+ − 1 Temperature dependences of all the scattering rates are obtained analytically. For kBT ≫ max(gη/v, vkF ), ΞU ∝ T 2K−1, ΞW ∝ T 2K+1, and Ξ± ∝ T 4K−3. In the presence of disorder and for kBT ≪ gη/v, ΞU ∝ T 2K, ΞW ∝ T 2K+2, and Ξ± ∝ T 4K−2. The additional power of T comes from disorder smearing of kF [35]. Full crossovers with and without disorder are determined by the explicit forms of ΞU,W,± provided in Ref. [34]. The dc conductivity can be obtained through Kubo formula [36]. The intraedge dc conductivity is e2  1 σ11 = − π e2 2h(cid:20) = lim ω→0 Imhω G(R) 11 (ω, k)i 1 Ξ+ + ΞU + ΞW + 1 Ξ− + ΞU + ΞW(cid:21) . (10) This expression is very different from the conductivity of an isolated edge, discussed below. Both intraedge and interedge interactions contribute to Eq. (10), but the intraedge contribution ΞW is subleading comparing to the interedge rates. The finite-temperature behavior for σ11 is summarized as follows. For clean edges and kBT ≫ vkF , σ11 ∼(T −4K+3, T −2K+1, for K ≤ 1, for K > 1. (11) With smooth disorder and kBT ≪ gη/v, for K ≤ 1, for K > 1. σ11 ∼(T −4K+2, T −2K, K ≤ 1 (repulsive interactions) is the physical situation. The transconductivity is σ21 = e2 2h(cid:20) 1 Ξ+ + ΞU + ΞW − 1 Ξ− + ΞU + ΞW(cid:21) . (13) This can be measured by shorting the distal part of pas- sive edge loop with an ideal (zero input impedance) cur- rent meter. The leading temperature dependence of the drag conductivity is the same as the intraedge conduc- tivity. In the usual case, one measures instead the drag resistivity [9, 13]. Here this evaluates to ρD = −ρ12 = (h/2e2) [Ξ− − Ξ+] , independent of the interedge U and intraedge W interactions. In the case of clean identi- cal edges, a positive drag resistivity with leading T 4K−3 behavior is obtained. This is the same result found previ- ously for spinless Luttinger liquids [11, 12] and TI edges with small magnetic fields [30]. Single edge. -- Finally, we consider dc conductivity of a single edge in isolation, in the presence of Rashba SOC. The least irrelevant intra-edge electron-electron interac- tion term allowed by time-reversal symmetry that can give a finite transport lifetime is HW = W Z dx :(cid:8)ei2kF xL†(x)R(x)R†(x)[−i∂xR(x)] +e−i2kF xR†(x)L(x)L†(x)[−i∂xL(x)] + H.c.(cid:9) :, (14) where :O: denotes the normal ordering of O. This is a one-particle spin-flip umklapp term. Similar one-particle backscattering interactions appear in Refs. [24, 25, 28], but the full temperature dependence of the dc conductiv- ity was not determined. The interaction correction due to Eq. (14) can be described by a self-energy with imag- inary part ImhΠ(R) k = 0 and ω → 0. We find that W (ω, k)i = −2ωΞW + O(cid:0)ω2(cid:1) , when 22Kπ2K+3KΓ [−K − 3] W 2 α2K ΞW = (v)2 l2K+1 T Γ [K + 2] + γ2 γ/π −∞ dy" ×Z ∞ 2π (cid:1)2 (cid:0)y − kF lT cosh (2πy) − cos (πK)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) × sin (πK) + (kF → −kF )# (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Γ(cid:2) 4+K 2 + iy(cid:3) Γ(cid:2) 2−K 2 + iy(cid:3) 2 , (15) where W = W/(π3/2α) and lT ≡ v/kBT denotes the thermal de Broglie wavelength. The disorder is en- coded in γ ≡ lT (K/v)2gη/2π. The dc conductivity is σdc = (e2/h)(1/ΞW ). At zero temperature where the HLL exhibits ballistic transport, σdc diverges. For a clean noninteracting edge (K = 1 and v = vF ), the conductivity reduces to 4 (12) σdc = e2 h (v)2 l3 T W 2π3 6 [cosh (kF lT ) + 1] 2π )4 + 5 2 ( kF lT 2π )2 + 9 . (16) (cid:2)( kF lT 16(cid:3) At high temperatures kF lT ≫ 1, this is proportional to T −3; in the opposite limit the umklapp scattering is thermally activated, giving σdc ∼ T exp (kF lT ). Lut- tinger interactions modify the high-temperature behav- ior to T −2K−1, while disorder leads to σdc ∼ T −2K−2 for lT ≫ (v)2g−1 η , again due to smearing of the Fermi momentum [35]. The disordered result is consistent with earlier predictions [24, 25, 28]. The T −2K−1 behavior is the most important temperature dependence in the clean edge due to the intraedge interactions, in the pres- ence of Rashba SOC. The responsible interaction term in Eq. (14) will be generated by renormalization irrespective of whether it arises in a particular microscopic model. Summary and discussion - In this work, we have shown that low-temperature edge state transport measurements for two proximate HLLs can quantify the value of the Luttinger parameter in the presence of spin-flip interedge electron-electron scattering. The latter is enabled by Rashba SOC within each TI, as can arise in InAs/GaSb. In contrast to the usual setup for Coulomb drag, the passive circuit floats without leads and provides a much stronger source of scattering for the active circuit edge than intraedge interactions, which are negligible at low temperature. Because of the topological protection, this result is immune to disorder but receives exponentially small corrections for a long, but finite interacting region. In the same device geometry, both the intraedge con- ductivity and the transconductivity show the same lead- ing temperature dependence for T above the crossover scale to the low-temperature locking regime. Thus, two- terminal conductivity gives an alternative route to detect Coulomb drag physics. We have also computed the con- ductivity correction due to the least irrelevant symmetry- allowed interaction in a given edge. This gives T −2K−1 temperature dependence for a clean edge. We close with some observations and avenues for fu- ture work. In general, negative drag is possible when kF 1 − kF 2 ≫ kF 1 + kF 2. Eq. (4) instead of Equation (3) dominates the interedge interactions at low temper- ature in this case. For two almost identical edges but kF 1 = −kF 2, a perfect antiparallel current locking can oc- cur; the two-terminal conductance is still given by Eq. (1) in the T → 0 limit. The finite-temperature behavior will be qualitative the same as the parallel drag situation. The generic kinetic theory of Coulomb drag between he- lical edge states, that also includes the forward-scattering long-ranged component of the Coulomb interaction, is an important topic for future work [37]. Understanding how a HLL edge state thermalizes via the various scattering mechanisms has crucial implications for nonequilibrium spectroscopy [38, 39]. It will also be interesting to study the noise [40] for the two-helical-edge setup described here. Y.-Z.C. and M.S.F. thank R.-R. Du, L. Du, D. Na- telson, and A. Nevidomskyy for useful discussions. A.L. thanks N. Kainaris, I. Gornyi and D. Polyakov for multi- ple important discussions and ongoing collaboration on a related problem. A.L. and M.S.F. acknowledge the hos- pitality of the Spin Phenomena Interdisciplinary Center (SPICE), where this work was completed. Y.-Z.C. and M.S.F. acknowledge funding from the Welch Foundation under Grant No. C-1809 and from an Alfred P. Sloan Research Fellowship (No. BR2014-035). Y.-Z.C. also ac- knowledges hospitality of the Michigan State University. A.L. acknowledges funding from NSF Grants No. DMR- 1401908 and No. ECCS-1407875. ∗ [email protected] 5 [23] Y. Tanaka and N. Nagaosa, Phys. Rev. Lett. 103, 166403 (2009). [24] T. L. Schmidt, S. Rachel, F. von Oppen, and L. I. Glaz- man, Phy. Rev. Lett. 108, 156402 (2012). [25] N. Lezmy, Y. Oreg, and M. Berkooz, Phys. Rev. B 85, 235304 (2012). [26] M. E. Peskin and D. V. Schroeder, An Introduction to Quantum Field Theory (Westview, Boulder, 1995). [27] Z. Ristivojevic, P. Le Doussal, and K. J. Wiese, Phys. Rev. B 86, 054201 (2012). [28] N. Kainaris, I. V. Gornyi, S. T. Carr, and A. D. Mirlin, Phys. Rev. B 90, 075118 (2014). [29] T. Li, P. Wang, H. Fu, L. Du, K. A. Schreiber, X. Mu, X. Liu, G. Sullivan, G. A. Csathy, X. Lin, R.-R. Du, Phys. Rev. Lett. 115, 136804 (2015). [30] V. A. Zyuzin and G. A. Fiete, Phys. Rev. B 82, 113305 (2010). [31] R. Shankar, Acta Phys. Pol. B 26, 1835 (1995). [32] W. Apel and T. M. Rice, Phys. Rev. B 26, 7063(R) (1982). [33] C. L. Kane and M. P. A. Fisher, Phys. Rev. B 46, 15233 (1992). [34] See supplemental material. [35] G. A. Fiete, K. Le Hur, and L. Balents, Phys. Rev. B 73, [1] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 165104 (2006). (2005). [36] J. Sirker, R. G. Pereira, and I. Affleck, Phys. Rev. B 83, [2] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 035115 (2011). (2010). [37] N. Kainaris, A. Levchenko, I. Gornyi, and D. Polyakov, [3] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 in preparation. [38] C. Altimiras, H. Le Sueur, U. Gennser, A. Cavanna, D. Mailly, and F. Pierre, Nature Physics 6, 34 (2010). [39] S. S. Apostolov and A. Levchenko, Phys. Rev. B 89, 201303(R) (2014). [40] Y. M. Blanter and M. Buttiker, Physics Reports 336, 1 (2000). (2011). [4] C. Xu and J. E. Moore, Phys. Rev. B 73, 045322 (2006). [5] C. Wu, B. A. Bernevig, and S.-C. Zhang, Phys. Rev. Lett. 96, 106401 (2006). [6] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buh- mann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). [7] I. Knez, R.-R. Du, and G. Sullivan, Phys. Rev. Lett. 107, 136603 (2011). [8] A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X.-L. Qi, and S.-C. Zhang, Science 325, 294 (2009). [9] A. G. Rojo, Journal of Physics: Condensed Matter 11, R31 (1999). [10] Y. V. Nazarov and D. V. Averin, Phys. Rev. Lett. 81, 653 (1998). [11] V. V. Ponomarenko and D. V. Averin, Phys. Rev. Lett. 85, 4928 (2000). [12] R. Klesse and A. Stern, Phys. Rev. B 62, 16912 (2000). [13] B. preprint Levchenko, Narozhny and A. arXiv:1505.07468 (2015). [14] T. Giamarchi, Quantum Physics in One Dimension (Ox- ford University Press, Oxford, 2004). [15] D. L. Maslov and M. Stone, Phys. Rev. B 52, R5539 (1995). [16] V. V. Ponomarenko, Phys. Rev. B 52, R8666 (1995). [17] I. Safi and H. J. Schulz, Phys. Rev. B 52, R17040 (1995). [18] T. Giamarchi, Phys. Rev. B 44, 2905 (1991). [19] D. L. Maslov, Phys. Rev. B 52, R14368 (1995). [20] V. V. Deshpande, M. Bockrath, L. I. Glazman, and A. Yacoby, Nature 464, 209 (2010). [21] C. L. Kane and E. J. Mele, Phy. Rev. Lett. 95, 226801 (2005). [22] B. A. Bernevig and S.-C. Zhang, Phys. Rev. Lett. 96, 106802 (2006). Helical Quantum Edge Gears in 2D Topological Insulators: SUPPLEMENTAL MATERIAL We set  = kB = 1 except as noted. BOSONIZATION CONVENTIONS We adopt the standard field theoretic bosonization method. The fermionic fields can be described by chiral bosons via 6 R(x) = 1 √2πα ei√π[φ(x)+θ(x)], L(x) = 1 √2πα ei√π[φ(x)−θ(x)], where α is some ultraviolet length scale. The time reversal operations in the bosonic language read: φ → −φ + √π 2 , θ → θ − √π 2 , and i → −i. In the imaginary time formalism, the Luttinger action for the ath edge reads BOSONIC ACTIONS i (∂xθa) (∂τ φa) + v 2 Zτ,x (cid:20)K (∂xφa)2 + xφa(cid:1) cos(cid:16)√4π θa + 2kF,ax(cid:17) , S0,a =Zτ,x SW,a = WaZτ,x (cid:0)∂2 Simp,a =Zτ,x ηa(x) 1 √π ∂xθa, 1 K (∂xθa)2(cid:21) , where Rτ,x is the short hand notation for R dτ dx. The inter-edge interactions correspond to SU = Xa=1,2 S+ = −U+ 2π2α2 Zτ,x U− 2π2α2 Zτ,x S− = Ua π3/2α Zτ,x (∂xφ¯a) sinh√4πθa + 2kF axi , cosh√4π (θ1 + θ2) + 2(kF 1 + kF 2)xi , cosh√4π (θ1 − θ2) + 2(kF 1 − kF 2)xi , where a labels the edge, with ¯1 = 2 and ¯2 = 1. We can integrate over ∂xφ1 and ∂xφ2 exactly. The θ-only action is a (∂xθa)2i 1 Sθ = Xa=1,2 − Xa=1,2 − Xa=1,2 i Wa 2vaKa Zτ,x h(∂τ θa)2 + v2 vaKa Zτ,x vaKaπ3/2α Zτ,x iU¯a (∂τ ∂xθa) cos(cid:16)√4π θa + 2kF ax(cid:17) (∂τ θa) sin(cid:16)√4πθ¯a + 2kF ¯ax(cid:17) + S+ + S−, (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) where we have dropped terms not contributing to the boson self-energy at the second homogeneous order of the coupling constants. 7 Forward-scattering potential disorder is averaged using the replica trick. The quadratic part of the action becomes Sθ,0 = 1 2vK Xa=1,2 R Xn=1Z dτ dxh(∂τ θa,n)2 + v2 (∂xθa,n)2i − gη 2π Xa=1,2 R Xn,m=1Z dτ dτ′dx [∂xθa,n(τ )] [∂xθa,m(τ′)] , (27) where n, m are the replica indexes, R is the number of replica, and gη is the variance of the disorder potential [η(x), introduced in the main text]. All interaction terms are diagonal in the replica space. DC CONDUCTANCE We derive the zero temperature dc conductance result [Eq. (1)] in this section. We consider the device geometry in Fig. 1. The edge carrying current I1 has Luttinger parameter K and charge velocity v. It is connected to external leads, which we treat as non-interacting free fermion reservoirs with K = 1 and v = vF . The electric field is applied between the leads. The passive circuit edge carrying current I2 forms a closed loop with uniform Luttinger parameter K and velocity v. For simplicity, we assume that this loop is infinitely long (but see below). The two edges interact via Eq. (7) in the region −L/2 ≤ x ≤ L/2. At zero temperature, the locking condition holds across this span of length L. We can replace the sine-Gordon term in Eq. (7) by a mass term, Hb,M = M 2 2 Z L/2 −L/2 dx (cid:20) θ1(x) − θ2(x) − c0 √2 (cid:21)2 . (28) The constant c0 can be absorbed by shifting θ2 → θ2 − c0; the locking condition becomes θ1 = θ2. The current I1 can be expressed in terms of retarded Green's functions, e2 πZ L/2 hI1(x)i = i 12 (ω; x, x′)i , (29) where E is the external electric field in the region −L/2 ≤ x ≤ L/2, and η1,2(x) are the random forward scattering potentials. The retarded Green's functions in the above formula are determined by [∂x′η2(x′)]hω G(R) 11 (ω; x, x′)i − i dx′ [E − ∂x′η1(x′)] lim ω→0hω G(R) dx′ lim ω→0 −L/2 −∞ e2 πZ ∞ ω2 v1(x)K1(x) + ∂xh v1(x) K1(x) ∂xi − M 2(x) M 2(x) 2 2   where M 2(x) 2 ω2 vK + v K ∂2 x − M 2(x) 2     G(R) 11 (ω; x, x′) G(R) 21 (ω; x, x′) G(R) 12 (ω; x, x′) 22 (ω; x, x′) G(R)  = δ(x − x′)1, 1, K1(x) =(K, −L/2 ≤ x ≤ L/2 x > L/2 v1(x) =(v, −L/2 ≤ x ≤ L/2 x > L/2 M (x) =(M, −L/2 ≤ x ≤ L/2 vF , 0, x > L/2 (30) (31) (32) (33) The retarded Green function can be solved by imposing the following boundary conditions [15, 19]. (i) G(R) ab continuous everywhere in x. (ii) v1(x) x ∼ x′, K1(x) ∂x G(R) 11 , v1(x) K1(x) ∂x G(R) 12 , ∂x G(R) 21 , and ∂x G(R) is 22 are continuous for x 6= x′. (iii) For K1(x) K1(x) (cid:20) v1(x) (cid:20) v1(x) h v h v K K ∂x G(R) ∂x G(R) 11 (ω; x, x′)(cid:21)x=x′+0+ −(cid:20) v1(x) 12 (ω; x, x′)(cid:21)x=x′+0+ −(cid:20) v1(x) K1(x) ∂x G(R) ∂x G(R) 11 (ω; x, x′)(cid:21)x=x′−0+ 12 (ω; x, x′)(cid:21)x=x′−0+ ∂x G(R) ∂x G(R) 21 (ω; x, x′)ix=x′+0+ −h v 22 (ω; x, x′)ix=x′+0+ −h v K K K1(x) ∂x G(R) ∂x G(R) 21 (ω; x, x′)ix=x′−0+ 22 (ω; x, x′)ix=x′−0+ = 0 = 1. = 1, = 0, (34) (35) (36) (37) (iv) The retarded Green functions obey the outgoing wave conditions. 8 Expanding G(R) ab ditions, we find that in each region in terms of propagating and/or evanescent waves and imposing all boundary con- lim ω→0hω G(R) 11 (ω; x, x′)i = lim ω→0hω G(R) 12 (ω; x, x′)i = −i K 2(1 + K) . The result is independent of x. The current expression in Eq. (29) becomes hI1(x)i = e2 π K 2(1 + K)Z L/2 −L/2 dx′ [E − ∂x′ η1(x′)] − e2 π K = (1 + K) e2 h [EL − η1(L/2) + η1(−L/2)] − (1 + K) K 2(1 + K)Z ∞ K dx′ [∂x′ η2(x′)] −∞ [η2(∞) − η2(−∞)] , e2 h (38) (39) where η1(L/2) = η1(−L/2) = 0 because the first edge is connected to free fermion leads, and η2(∞) = η2(−∞) due to the periodic boundary condition. The conductance of the locked edges is then G = hI1i EL = K 1 + K e2 h . (40) Adding the contribution of the parallel edge carrying I′1 gives Eq. (1). In the presence of the inter-edge Luttinger interactions, the Luttinger parameter in the region −L/2 ≤ x ≤ L/2 is modified. The two-terminal conductance is unchanged if we assume that the distal part of the passive edge is much longer than the interacting part, as above. Then this passive edge is effectively connected to the Luttinger liquid leads with Luttinger parameter K, and this gives resistance (1/K)h/e2. For a finite total length l > L of the passive edge, the value of the measured Luttinger parameter is between the intra-edge value K and the Luttinger parameter for the symmetric mode, and it also depends on the ratio of the interacting region length L to the total length l. SCATTERING RATES At second order in the interaction coupling strengths, there are four distinct self-energies, ΠW , ΠU , Π+, and Π−. Coupling constant mixing will occur at higher orders, but is prevented here by vertex operator charge neutrality conditions. We are interested in the imaginary part of the retarded self-energies. In the long-wavelength and low- energy limits, Im(cid:2)Π(R)(ω, k)(cid:3) ≈ −2ωΞ. In this section we provide the explicit results for the scattering rates that enter the intra-edge and transconductivities in Eqs. (10) and (13). The scattering rate due to Eq. (14) already appears in Eq. (15). Here γ = K 2βgη/2πv is a parameter indicating the ratio of the effective disorder strength to the temperature. In the clean limit (γ → 0), the Lorentzian distributions in Eq. (15) becomes delta functions. The scattering rate reduces to sin (πK) Γ [−K − 3] Γ [K + 2] ΞW =(cid:16) W αK(cid:17)2 22K+1π2K+3 v2K+3β2K+1 K where β is the inverse temperature. When T ≫ vkF , the clean scattering rate is proportional to T 2K+1. 2π i 2π i In the presence of disorder and at low temperatures such that γ ≫ 1, the scattering rate becomes v2K+3β2K+2  ΞW =(cid:16) W αK(cid:17)2 22Kπ2K+1  cosh (vβkF ) − cos (πK)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) √π Γ(cid:2)2 + K 2(cid:3) Γ(cid:2) K Γ(cid:2) 5 2 + K(cid:3) 2 + i vβkF 2 + i vβkF Γh 4+K Γh 2−K 2π (cid:17)2 (cid:16) vβkF 2(cid:3) Γ(cid:2) 5 2(cid:3) Γ(cid:2) 1 K sin (πK) 2 + K   βγ/π + γ2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 2 + K 2(cid:3) Γ [−K − 3] . (42) 2 , (41) The term in square brackets is independent of temperature, so that ΞW ∼ T 2K+2. The inter-edge single particle backscattering Hamiltonian in Eq. (5) contains two terms, U1 and U2. U1 and U2 correct the dc conductivity in edge 1 and 2, respectively. The scattering rates are 9 ΞU,a =(cid:16) UaαK(cid:17)2 22K−3π2K+1 v2K+1β2K−1 1 K Γ [1 − K] Γ [2 + K]   2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)  γ/π + γ/π   dy  ×Z ∞ −∞ 2π (cid:17)2 (cid:16)y − vβkF,a In the high-temperature limit γ → 0 and vβkF,a → 0, ΞU is proportional to T 2K−1. where Ua = Ua/π3/2α. In the low-temperature limit γ ≫ 1, ΞU is proportional to T 2K. In the main text, we assume two identical edges and U1 = U2, so that ΞU,1 = ΞU,2 = ΞU . 2π (cid:17)2 (cid:16)y + vβkF,a + γ2 + γ2 sin (πK) cosh (2πy) − cos (πK)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Γ(cid:2) K 2 + iy(cid:3) Γ(cid:2)− K 2 + iy(cid:3) , (43) The scattering rates due to Eqs. (3) and (4) are given by Γ [1 − 2K] Γ[2K] v4K−1β4K−3 Ξ± =(cid:16) U±α2K(cid:17)2 24K−3π4K 4π (cid:17)2 (cid:16)y − vβk± ×Z ∞   −∞ F ¯γ/π 2  where ¯γ = K 2βgη/πv, U± = ∓U±/2π2α2, and k±F = 2(kF 1 ± kF 2). vβk±F → 0, Ξ± is proportional to T 4K−3. In the low-temperature limit ¯γ ≫ 1, it is proportional to T 4K−2. cosh (2πy) − cos(2πK)(cid:12)(cid:12)(cid:12)(cid:12) Γ [1 − K + iy](cid:12)(cid:12)(cid:12)(cid:12) 4π (cid:17)2 (cid:16)y + vβk± dy  Γ [K + iy] sin(2πK)   + ¯γ2 + ¯γ2 ¯γ/π + F In the high-temperature limit ¯γ → 0 and , (44) 

1905.04034

1

1905

Stochastic dynamics and pattern formation of geometrically confined skyrmions

Ensembles of magnetic skyrmions in confined geometries are shown to exhibit thermally driven motion on two different time scales. The intrinsic fluctuating dynamics ($t\sim 1~$ps) is governed by short-range symmetric and antisymmetric exchange interactions, whereas the long-time limit ($t\gtrsim10\,$ns) is determined by the coaction of skyrmion-skyrmion-repulsion and the system's geometry. Micromagnetic simulations for realistic island shapes and sizes are performed and analyzed, indicating the special importance of skyrmion dynamics at finite temperatures. We demonstrate how the competition between skyrmion mobility and observation time directly affects the addressability of skyrmionic bits, which is a key challenge on the path of developing skyrmion-based room-temperature applications. The presented quasiparticle Monte Carlo approach offers a computationally efficient description of the diffusive motion of skyrmion ensembles in confined geometries, like racetrack memory setups.

cond-mat.mes-hall

cond-mat

a Stochastic dynamics and pattern formation of geometrically confined skyrmions Alexander F. Schaffer,1, ∗ Levente R´ozsa,2 Jamal Berakdar,1 Elena Y. Vedmedenko,2 and Roland Wiesendanger2 1Institut fur Physik, Martin-Luther-Universitat Halle-Wittenberg, D-06099 Halle (Saale), Germany 2Institut fur Nanostruktur- und Festkorperphysik, Universitat Hamburg, D-20355 Hamburg, Germany (Dated: May 13, 2019) Ensembles of magnetic skyrmions in confined geometries are shown to exhibit thermally driven motion on two different time scales. The intrinsic fluctuating dynamics (t ∼ 1 ps) is governed by short-range symmetric and antisymmetric exchange interactions, whereas the long-time limit (t (cid:38) 10 ns) is determined by the coac- tion of skyrmion-skyrmion-repulsion and the system's geometry. Micromagnetic simulations for realistic island shapes and sizes are performed and analyzed, indicating the special importance of skyrmion dynamics at finite temperatures. We demonstrate how the competition between skyrmion mobility and observation time directly affects the addressability of skyrmionic bits, which is a key challenge on the path of developing skyrmion-based room-temperature applications. The presented quasiparticle Monte Carlo approach offers a computationally ef- ficient description of the diffusive motion of skyrmion ensembles in confined geometries, like racetrack memory setups. Magnetic skyrmions1 -- 3 are quasiparticles which are con- sidered as possible carriers of information for future storage devices. Their specific chirality is determined by the antisym- metric Dzyaloshinskii-Moriya exchange interaction (DMI)4,5. The DMI can be induced by a broken inversion symmetry in a crystal itself (e.g. MnSi)2 or by interfacing a heavy metal layer (e.g. Pt, W, Ir) with a ferromagnetic material (e.g. Fe, Co)6,7. Conceptually the utilization of these topologically non- trivial8 quasiparticles in so-called racetrack setups is of great interest 9 -- 11: Skyrmions can be manipulated (written and deleted)12,13 and addressed individually14 -- 16 on a magnetic stripe, allowing a memory device extension from a pure surface density into the third dimension by pushing the quasiparticle-hole-train back and forth, e.g. by applying elec- trical currents17,18. The low threshold driving current19 along with the small size and high stability of the skyrmions are key features of this concept. To connect experimental and theoretical model systems with technological applications, investigating the influence of finite temperatures is of crucial importance. The bits on a racetrack-based memory device need to fulfill two main fea- tures, stability against external perturbations and addressabil- ity. Both are affected by thermal fluctuations, as shown below. The stability of skyrmions was examined in several publi- cations over the last years20 -- 24. R´ozsa et al.21 investigated the- oretically periodic two-dimensional Pd/Fe double-layers on Ir(111) and determined the phase diagram as a function of ex- ternal field and temperature, which includes field-polarized, skyrmion lattice, spin spiral, fluctuation-disordered and para- magnetic regions. Skyrmion lifetimes in the fluctuation- disordered regime were calculated. The lifetimes of iso- lated skyrmions in racetrack geometries for Pd/Fe/Ir(111) and Co/Pt(111) systems were investigated in Refs.22,23, and differ- ent mechanisms were revealed for the collapse of a skyrmion inside the track and at the boundary. The diffusive motion of skyrmions at finite temperature has also attracted significant research attention lately25 -- 27. Previous studies concentrated on diffusion in infinite or ex- tended geometries, but a clarification of the role of the sam- ple shape still seems to be missing in the case where the sys- tem size becomes comparable to that of the skyrmions. Ef- fects of this kind directly impede the addressability, which is indispensable when using skyrmions for storage technol- ogy, e.g. in racetrack memory devices. Since the number of skyrmions during the diffusive motion remains constant, quasiparticle models have been developed for their description in this limit25,26,28,29, which are primarily based on the Thiele equation30. The advantage of such a collective-coordinate de- scription over micromagnetic or atomistic spin dynamics sim- ulations is its significantly lower computation cost. Due to the finite temporal resolution, imaging techniques on the atomic length scale like spin-polarized scanning tun- neling microscopy (SP-STM)31 or magnetic force microscopy (MFM)32 cannot access temporally the diffusive motion of the skyrmions. Instead, the time-averaged skyrmion proba- bility distribution is imaged which may well be different from the zero-temperature configuration or a snapshot of a simula- tion. Here we will discuss the formation, stability and address- ability of diffusive skyrmion configurations. Using micro- magnetic simulations, it is shown that the complex interplay of the repulsive interaction between the skyrmions33,34 along with the confinement effect of the nanoisland and the ther- mally induced skyrmion diffusion leads to a pattern formation of the skyrmion probability distribution on the nanosecond time scale. A computationally efficient quasiparticle Monte Carlo method is introduced, which is found to yield compa- rable results to time-averaged micromagnetic simulations re- garding the skyrmion probability distribution. The simple im- plementation and high speed of such a method may make it advantageous over micromagnetic simulations when the ac- tual number of skyrmions has to be determined based on a time-averaged experimental image. RESULTS Skyrmion stability. We study metastable skyrmions and their motion and mobility at finite temperatures. At first the phase space has to be explored with respect to the exter- nal magnetic field and temperature. A detailed description of the phase diagram for extended systems of ultrathin bi- 2 fields lead to shrunken skyrmions, which in turn enables the magnetic island to host a larger number of quasiparticles. In this regime the topological charge is always close to an in- teger, with the small deviation caused by the tilting of the spins at the edge of the sample. Reaching field values above B ≈ 3.5 T, the system becomes completely field-polarized and skyrmions are not stabilized anymore starting from a ran- dom configuration, even at zero temperature. Previous calcu- lations37 -- 39 indicated that skyrmions on the lattice collapse at around B = 4.5 T in the system. In order to prevent the appearance of spin spiral states and to be able to consider the skyrmions as well-defined quasi- particles, we choose B = 1.5 T for the following calcula- tions, unless mentioned otherwise. Figure 1 shows that sev- eral metastable configurations associated with different topo- logical charges are accessible starting from randomly gener- ated initial states. However, in this approach not all stable structures are covered, as the included ones are obtained from relaxing random initial configurations. Different configura- tions may also be generated in a controlled way by adding skyrmions to the field-polarized state one-by-one, and relax- ing the state at zero temperature. This is shown in Fig. 2 for the disk-shaped and a triangular geometry for B = 1.5 T. On top of the transitions due to a variation of the magnetic field, the impact of finite temperature is also of crucial im- portance. In Fig. 3a temperature effects on the topological charge are displayed. Starting from a relaxed configuration at T = 0 K, the temperature is increased every 500 ps by ∆T = 2 K and the topological charge is averaged over time at each temperature. The results show a first discontinuity at T1 ≈ 40 K. Up to this temperature the thermal fluctuations are relatively weak, and the number of well-defined skyrmions in- side the sample remains constant. The small standard devia- tion of the topological number in this regime can be attributed to the thermal motion of the spins at the edge. However, the combination of thermal fluctuations and the repulsion be- tween the skyrmions triggers the escape of a single skyrmion out of the sample at T1 ≈ 40 K. Hence, the total topological number is changed from NSk ≈ −5 to NSk ≈ −4, see Sup- plementary Video 1. This change is also shown in the plot of the topological number susceptibility in Fig. 3b, defined as (cid:0)(cid:104)N 2 Sk(cid:105) − (cid:104)NSk(cid:105)2(cid:1) , (1) χN = 1 T (cid:82) t1 t0 with (cid:104)NSk(cid:105) = 1 NSk(t)dt. Only a minor deviation t1−t0 from the otherwise smooth function is visible. This means that the thermal fluctuations are still not strong enough to perturb the quasiparticles drastically. A first characteristic change in the slope of the topological susceptibility can be seen at T2 ≈ 50K. Comparing Figs. 3a and b it is obvious that the fluctuations of topological num- ber are increasing with temperature, and around this point the lifetime of skyrmions is reduced below the averaging window. The average topological charge continuously approaches zero above this temperature. In this disordered regime the lifetime of skyrmions is shorter than the observation time, as the fluc- tuations allow a collapse inside the sample. Hence, this tem- perature range is not favorable for information storage appli- FIG. 1. Field dependence of the total topological charge. Red re- gion: external fields below the strip-out instability leading to spin spiral segments; dark green: transition regime; bright green: en- semble of skyrmions; blue: completely field-polarized island, where skyrmions collapse due to the strong external field. Solid red line indicates the magnetic field B = 1.5 T used for subsequent calcula- tions. For each step in the vertically applied field 20 randomly mag- netized configurations of a 21 nm diameter nanodisk were relaxed towards the nearest metastable state on the energy hypersurface at T = 0 K. atomic layers of Pd/Fe on an Ir(111) substrate hosting mag- netic skyrmions can be found, e.g., in Refs.21,35. Here, simula- tions are performed for the same system in the micromagnetic framework, by solving the Landau -- Lifshitz -- Gilbert equation (LLG)36, as described in the Methods section. In Fig. 1 possible equilibrium spin configurations are in- vestigated in a 21-nm-diameter nanodisk of 0.4 nm thickness at zero temperature. For each fixed external magnetic field (B = Bz), 20 different randomly generated initial states are considered and relaxed towards the closest local mini- mum of the total energy hypersurface. Subsequently, the total skyrmion number NSk of the relaxed states is calculated and plotted against the external magnetic field in Fig. 1 to gain an overview of the possible metastable states. Below the strip- out instability field37, skyrmions as localized cylindrical spin structures are not stable and the configuration consists of spin spiral segments. For these the topological number is not a well-defined quantity, as spin spiral structures can extend over the whole island, and therefore the boundaries have a signif- icant effect. This is reflected by the continuous distribution of topological charge values found below 0.75 T in the red regime in Fig. 1, whereas discrete skyrmion numbers corre- spond to discrete steps in the total topological charge. The upper boundary of this region is reasonably close to the strip- out field of B = 0.65 T37,38 determined for extended systems. With increasing the external field, individual skyrmions may be stabilized in the system, where they will coexist with spin spiral segments. This corresponds to the darker green area in Fig. 1 around B ≈ 1 T, where besides the continuous distribution also discrete steps can be observed in the topolog- ical charge. In the following bright green region stronger magnetic 01234-8-6-4-2024B(T)NSkB03 FIG. 3. Temperature dependence of the topological number. Disk-shaped island (21 nm diameter) for B = 1.5 T. a Total topo- logical number (cid:104)NSk(cid:105) averaged over 0.5 ns with error bars indicat- ing the standard deviation. b Topological number susceptibility χN (Eq. (1)), with the dashed black lines indicating three characteristic temperatures discussed in the main text. Solid red line marks the temperature used for successive simulations. Skyrmion diffusion. With these first results, giving in- formation on the segment of the parameter space we deal with, we will focus on the influence of thermal fluctuations on the dynamics of the skyrmions. In Fig. 4a and 5 results for different initial numbers of skyrmions presented in Fig. 2 are shown for the circular and triangular geometries, respec- tively. The first row shows the magnetic configuration after 20 ps of simulation time and the second row the final state, which means the magnetization state after our total simula- tion time of 20 ns. Only the out-of-plane z component is shown in grayscale. The magnetic configuration after only 20 ps is qualitatively very similar to the final one, displaying slightly deformed skyrmion configurations due to the thermal fluctuations compared to the zero-temperature initial states in Fig. 2. This short time scale on which the short-range Heisen- berg and Dzyaloshinskii-Moriya interaction dominate the dy- namics and cause shape deformations, is examined more ex- tensively below. The third row corresponds to the time-averaged z compo- nent, calculated over the simulation time divided into 1000 snapshots. In our understanding this result comes as close as possible to real-space scanning-probe measurements, due to the limited time resolution and finite measurement dura- tion in such techniques. According to Ref.37, typical limits FIG. 2. Relaxed states for different initial skyrmion numbers. Brightness corresponds to the out-of-plane component of the mag- netization characterized by the polar angle θ, whereas the color de- notes the in-plane orientation described by the azimuthal angle ϕ, as shown in the color map. B = 1.5 T, T = 0 K. System size: a 21 × 21 × 0.4 nm3; b 30 × 30 × 0.4 nm3. The saturation magneti- zation is locally set to zero in the dark gray areas. cations. Finally, at about T3 ≈ 100K the paramagnetic regime is en- tered, which is characterized by a completely disordered time- averaged magnetic configuration. Here the average topologi- cal charge is very close to zero. This behavior is also indicated by a change of sign in the first derivative dχN (T )/dTT =T 3. In the following, the external parameters are fixed to T = T0 = 15 K and B = B0 = 1.5 T, shown by the solid red lines in Figs. 1 and 3. This ensures that the thermal fluctu- ations do not lead to a collapse or escape for the skyrmions, only to a diffusive motion, and that the number of particle-like skyrmions will be a constant during the simulation time. of the time resolution in SP-STM are tres > 5 ms. With the present simulation parameters we found that increasing the simulation time further does not lead to a significant change in the time-averaged images, indicating that similar observations may be expected on the much longer experimental time scales as well. The observed two characteristic time scales can be attributed to different interaction types of clearly distinguish- able energy scales: local deformations on the picosecond time scale of the magnetic texture are dominated by thermal excita- tions competing with the short-range symmetric and antisym- metric exchange interactions, while the translational motion leading to the complex pattern formation over tens of nanosec- onds is caused by the repulsive interactions between pairs of skyrmions and between skyrmions and the boundaries. In case of the disk, it is extremely challenging to make a clear statement about the number of skyrmions in the system based on the time-averaged images only. As the symmetry of the system is a cylindrical one, also the resulting picture is cylindrically symmetric, consisting of concentric bright and dark circles and rings. The small contrast differences along the angular direction within a single diffusive ring-like area (cf. third row in Fig. 4a) arise because of the finite simula- tion time, but also as an artifact of the finite grid, leading to a deviation from the ideal radial symmetry. For all initial con- figurations the total skyrmion number is conserved. Since the skyrmions repulse each other, the radius of the resulting gray ring increases for higher skyrmion densities, before one of them becomes localized quite strongly in the center. This be- havior is analyzed in Fig. 4b, where the out-of-plane magneti- zation component depending on the distance from the center of the disk is calculated by integration over the angular coor- dinate. For comparison, the static profile of a single skyrmion in the center of the disk is plotted as well (black curve). The positions of the local minima of the integrated functions are shown in Fig. 4c. Without performing the averaging over time, the profile of the skyrmion rotates from mz = −1 in the cen- ter towards mz = 1 in the field-polarized background. Due to the diffusive motion of the skyrmions, the time-averaged im- ages show a smaller difference compared to the background magnetization at the approximate positions of the skyrmions, providing a measure for the localized nature of the quasiparti- cles in the sample. For an increasing number of skyrmions not only the radius of the resulting ring-shaped pattern increases, but so does the localization of the skyrmions. For the triangular system in Fig. 5, the lower symmetry of the geometry is reflected in the resulting time-averaged pictures. The symmetry of the spin configuration coincides with that of the sample for one, three and six skyrmions without thermal fluctuations, as shown in Fig. 2b. In the time-averaged images in the third row of Fig. 5, this re- sults in well-localized skyrmions with a strong dark contrast for these numbers of quasiparticles in the system. Interest- ingly, also the adjacent configurations show almost the same time-averaged pattern as the highly symmetric configuration, e.g. two or four skyrmions compared to the case of three skyrmions. For the case where one skyrmion is missing from the highly ordered configuration, the remaining skyrmions im- itate the absent skyrmion and effectively show up as an addi- 4 FIG. 4. Diffusive skyrmion motion on a circular island. a z com- ponent of the magnetization (white +z, black −z) for an ascending total skyrmion number. First row shows the configuration after 20 ps, the second row after 20 ns of elapsed time, and the third row repre- sents the time average for the total simulation time over 1000 snap- shots. b Time-averaged signal as a function of the distance from the center of the disk for different skyrmion numbers, averaged over the angular coordinate. Black line shows the static profile of a central single skyrmion. c Radii rmin corresponding to the local minima of the time-averaged signal in panel b, in dependence on the skyrmion number. Disk system (21 × 21 × 0.4 nm3), T = 15 K, B = 1.5 T. The saturation magnetization is locally set to zero in the dark gray areas. tional "phantom skyrmion". In contrast, a surplus skyrmion smears out in the time-averaged picture, so that mostly the high-symmetry points remain observable, even though they can also be slightly broadened -- cf. the 6th and 7th columns in Fig. 5. For an even larger numbers of skyrmions, the repul- sive interaction between them in combination with tempera- ture fluctuations is strong enough for the skyrmions to escape at the boundary during the simulation time, as shown in the decreased number of quasiparticles in the second row com- pared to the first row in the 8th and 9th columns in Fig. 5. 1Skprofilemin20ps20nsav.02468-1.0-0.50.00.51.05 FIG. 5. Diffusive skyrmion motion on a triangular island. z component of the magnetization (white +z, black −z). First row shows the configuration after 20 ps, the second row after 20 ns of elapsed time, and the third row represents the time average for the total simulation time over 1000 snapshots. Triangular system (30 × 30 × 0.4 nm3), T = 15 K, B = 1.5 T. The saturation magne- tization is locally set to zero in the dark gray areas. To identify the different time regimes more clearly, the convergence behavior of the time-averaged pictures is inves- tigated. The time-integrated pictures are calculated up to each snapshot time and compared to the averaged image after 20 nsvia a matching parameter, where a complete agreement is denoted by a value of 1. The mathematical definition of the matching parameter is given in the Methods. The results are plotted in Fig. 6. Panels a and b show the convergence behavior for different skyrmion occupations for the disk and for the triangular geometry. For the circle there are no ma- jor differences in terms of the convergence speed for different skyrmion numbers. Only the ensembles of three and seven skyrmions exhibit a slightly faster convergence, which is ex- plicable by the resemblance of those states to the close-packed arrangement. As expected, the close-packed array possesses a lower mobility. In the case of the triangle, for three and six skyrmions the value 1 is reached much faster than for the other configurations. This effect can again be explained by the strong localization of the skyrmions in these cases, preventing an exchange of their positions. A different trend is visible for the eight-skyrmion case (purple line in Fig. 6b), which shows a much slower convergence. This delayed behavior arises due to the non-conserved skyrmion number, which can be seen in Fig. 5. Consequently, after the escape of the skyrmions it takes more time until the averaged picture has adapted to this change. For exploring the stochastic dynamics on the short time scale, which we suppose to be governed by the skyrmion deformations in contrast to the slow skyrmion displacement, multiple simulations are performed for the same initial con- figuration, two skyrmions inside the disk relaxed at zero tem- perature, but with different pseudorandom number seeds. The simulations cover a time of 100 ps each with 50 different ini- tializations. After every 0.5 ps a snapshot is taken and the positions of the two skyrmions extracted. On the scale of the simulation time no noteworthy displacement of each of the skyrmions takes place, however a deformation of the mag- netic textures is visible. The mean skyrmion radius relative to the zero temperature radius calculated as described in the Methods, is averaged over the different simulations, shown in Fig. 7. Here, two features are remarkable. Firstly, the radius FIG. 6. Convergence behavior of the time-integrated magneti- zation for different skyrmion numbers. The match between the final time-averaged pattern after 20 ns and the time average taken until time t is defined such that a value of 1 corresponds to perfect agreement, see the Methods for details. a and b show the conver- gence of the matching parameter for the disk and triangular setup, respectively, for different skyrmion numbers. increases, approaching an enhancement of about 10 % after 100 ps. Secondly, the radius is oscillating with a high fre- quency of approximately 60 GHz, estimated from the average time between the peaks. This oscillation can be identified with internal skyrmion modes, like the so-called breathing modes (cf. Ref.38: at T = 0 K, B = 1.5 T and the triangular atomic lattice the breathing mode frequency is f = 70.88 GHz, which is expected to decrease at higher temperature). These findings support the result that in confined geometries, differ- ent time scales are important for the skyrmion dynamics. On the short time scale the internal modes and deformations of the skyrmions are dominant, whereas on the longer time scale the interaction leads to a slowly converging pattern formation. All of the discussed examples showcase the strong influ- ence of the interaction between skyrmions as well as of the ge- ometrical aspects on their mobility, that can vary between de- localization and a strong localization, e.g., in the center of the disk in Fig. 4a. This feature of variable mobility, depending on the geometry and packing density of skyrmions, is of gen- eral nature. In real experimental systems further contributions may affect the mobility, and ultimately the measured position of the skyrmion as well. As an example of such a feature, defects in the grown thin film nanoislands will be discussed. In Fig. 8, a nanoisland geometry inspired by the experimental results in Refs.7,12 is considered. Additionally to the previous simulation parameters, a magnetic defect is inserted in the top- left corner of the island, treated as a simulation cell with the 12345678920ps20nsav.6 FIG. 8. Diffusive skyrmion motion on an asymmetric nanoisland including a defect. z component of the magnetization (white +z, black −z) is imaged. In the top-left corner a defect is modeled by freezing the magnetization direction in a single simulation cell along the +x direction. a Snapshots of the magnetization at different times, as well as the resulting time-averaged picture (av.). b Time-averaged signal along the dashed arrow shown in the lower right part of panel a. Simulation box: 30 × 30 × 0.4 nm3, T = 25 K, B = 1.5 T. The saturation magnetization is locally set to zero in the dark gray areas. skyrmion and the boundary needs to be quantified. Due to the Neumann boundary condition40 at the edge of nanois- lands, the DMI leads to a noncollinear spin texture, similar to a skyrmion41. Therefore, the interaction mechanism is the same between the skyrmions and between a skyrmion and the boundary, as has been studied in previous publications33. For details on the calculation and the explicit functions of the po- tentials see the Methods section. potentials quasiparticle Starting from the modeled skyrmion-skyrmion and skyrmion-boundary simulations are executed, in order to compare the calculated probability densities with the time-averaged configurations obtained from full-fledged micromagnetic simulations. The main assump- tion for this endeavor is that neither the thermal fluctuations lead to the collapse, creation or escape of skyrmions, nor does a high skyrmion density lead to a merging of the skyrmions on the island as discussed in Ref.39. A triangular setup containing two quasiparticles serves as the model system. The side of the triangle is chosen to be 30 nm, the same as in FIG. 7. Fast deformation dynamics of two skyrmions on a disk- shaped island. Two skyrmions on a disk (21 × 21 × 0.4) nm3 are initialized from a relaxed zero-temperature configuration. Afterward, 50 different seeds for the pseudorandom number generator are used to simulate the skyrmion dynamics over 100 ps, saved every 0.5 ps. The results are analyzed by calculating the centers of the skyrmions, followed by the computation of the average skyrmion radius (see Methods) for each of them as shown in a and b. Averaging over the different simulations smears out most of the pure stochastic dynam- ics, yielding an oscillating pattern superposed with a general trend of an increased skyrmion radius compared to the zero-temperature configuration. Bright blue points show data averaged over different simulations, the solid blue line corresponds to data smoothened by a Gaussian filter in order to calculate the peak positions (red dots). magnetization frozen along the in-plane +x direction. The initial configuration including 5 skyrmions is evolved in time over a span of 20 ns, as shown in the snapshots in panel a. During the time evolution one skyrmion is pinned at the de- fect, whereas the others are moving rather freely. The result- ing time average of the images reflects this behavior, as only a single spot is visible with a well-defined position in the up- per left corner, and the rest of the quasiparticle features form a blurred trace mostly following the geometry of the island. Additionally, the profile of the out-of-plane spin component is shown in Fig. 8b along the dashed arrow in the lower right panel of Fig. 8a. In this representation the discrepancy be- tween the measured profiles of the different skyrmions is even more pronounced. Where the localized skyrmion appears as a sharp dip almost reaching mz = −1, the superposition of the other four skyrmions leads to broader and shallower features, and therefore no clear indication of their positions. Quasiparticle model. The localization of the quasiparti- cles discussed above may also be interpreted as the probabil- ity density of finding a skyrmion at a selected position over the complete simulation time. Such a probabilistic interpreta- tion is capable of describing the different contrast levels ob- served for nominally equivalent skyrmions, and it proves to be valuable in the interpretation of scanning-probe experimental results where the characteristic measurement times are signif- icantly longer than the time scale of the diffusive motion. In this section, a simplified quasiparticle model is introduced, motivated by the time-averaged data of the micromagnetic simulations, offering a time-efficient opportunity in predict- ing the complex pattern formation of the skyrmion probability distribution discussed above. For the purpose of using a quasiparticle approach, the re- pulsive potential between pairs of skyrmions and between the 5ns10ns20nsav.10nmab051015202530-1.0-0.50.00.51.0y(nm)〈mz〉the case of the micromagnetic calculations shown in Figs. 2b and 5. The motion of the skyrmions inside the potential is calculated by modeling them as interacting random walkers, using an implementation of a Markov-chain Monte Carlo algorithm with Metropolis transition probabilities to study their diffusion, described in detail in the Methods section. This method offers a fast way of obtaining the cumulative probability density function of the positions of the two particles, which in turn can be compared to the time-averaged 4 and 5 obtained from micromagnetic images in Figs. simulations. In the following, a quantitative comparison between full micromagnetic simulations and Monte Carlo calculations will be presented to review the validity of the simplified quasiparticle approach. For this endeavor, we tracked the positions of two skyrmions in the triangular system from the micromagnetic simulation snapshots after every 20 ps and calculated the probability density distribution of the center coordinates. The resulting distribution along with the Monte Carlo result after 105 Monte Carlo steps is shown in Fig. 9. The largest difference between the two results arises because of the substantially less data for the micromagnetic calculation. As the histogram is calculated with only 103 snapshots compared to 105 Monte Carlo steps, this behavior is expected. Nevertheless, the characteristic features are similar. In both results the three peak densities are not radially symmetric, but are resembling the triangular boundary shape. Hence this deformation is not an artifact of the Monte Carlo simulation. Additionally, the average distance between the density maxima may serve as a good quantity for comparing both methods. As Fig. 9b lacks the resolution for calculating the peak positions reliably, we start instead from the time-integrated magnetization pictures in Fig. 5. Based on the averaged result for two skyrmions, we extract again the peak positions and calculate subsequently the mean distance. The positions for the Monte Carlo calcu- lations are at (11.0, 6.5) nm, (15.0, 13.5) nm, (19.0, 6.5) nm, the micromagnetic (11.1, 6.6) nm, (15.0, 13.5) nm, (19.0, 6.5) nm. This leads to an average distance of 8.0 nm for both the Monte Carlo calculations and the micromagnetic simulations, supporting the validity of the simplified quasiparticle approach in the investigated parameter space. simulations deliver The model is capable of predicting the skyrmion distribu- tion in the long time limit and thereby serves as a method for computationally efficient calculations for larger or more com- plex systems. As an example, the probability density func- tions of the Monte Carlo simulations are shown in Fig. 10 af- ter 105 Monte Carlo steps for different temperatures and sam- ple sizes. Due to the repulsion between the two skyrmions, equilibrium positions are found close to the vertices of the triangle, with energy barriers between these local energy min- ima. At low temperature (T = 1 K) and a small island sizes (30 nm edge length) in the upper left panel in Fig. 10, no tran- sitions between the minima occur during the duration of the simulation, and two skyrmions may be observed in the ob- tained probability densities. As the temperature is increased to T = 15 K or T = 30 K in the first row of Fig. 10, the transition rate between the preferential positions is enhanced, 7 FIG. 9. Comparison of the Monte Carlo and micromagnetic sim- ulations. The probability density function for two skyrmions inside a triangular island as obtained from a: micromagnetic simulations, and b: from calculations following the Monte Carlo algorithm. For the micromagnetic distribution, the coordinates of the skyrmions are extracted from the snapshots taken every 20 ps over 20 ns of simu- lation time at T = 15 K. Panel b is generated over 105 Monte Carlo steps. Probability density function for two skyrmions in- FIG. 10. side a triangular model potential calculated following a Monte Carlo algorithm. The temperatures are T = 1 K, T = 15 K and T = 30 K, the edge lengths of the triangle are 30 nm, 45 nm and 60 nm. Probability distributions for the centers of the two skyrmions are calculated over 105 Monte Carlo steps. and an additional "phantom skyrmion" appears in the proba- bility density function, in remarkable agreement with the mi- cromagnetic results as analyzed before. By enlarging the size of the island (second and third rows in Fig. 10 for 45 nm and 60 nm edge length), the energy surface becomes flatter, which in turn leads to higher transition probabilities between the en- ergy minima at the same temperature. Therefore, raising the temperature or increasing the size of the system have similar effects on the resulting probability density functions for the skyrmions. Consequently, large island sizes or high tempera- tures result in completely delocalized skyrmions as indicated in the bottom right panel in Fig. 10. solve the LLG equation, mi(t) = − γ 1 + α2 mi × Beff (cid:104) i + αmi ×(cid:16) (cid:17)(cid:105) 8 , (2) mi × Beff i for every simulation cell mi of the discretized magnetization vector field. Here, γ0 = 1.76 × 1011(T−1s−1) is the gyro- magnetic ratio of an electron and α is the Gilbert damping parameter. The time- and space-dependent effective magnetic field, Beff i (t) = Bext i + Bexch i + Bd i + Ba i + Bdmi i + Bth i (t) , (3) i i ; the external field Bext exchange in- is composed of teraction field Bexch = 2Aexch/M sat∆mi, with Aexch the saturation magne- the exchange stiffness and M sat i = M sat Kij ∗ mj, tization; demagnetizing field Bd where details on the calculation of the demagnetiz- ing kernel K can be found in Ref.40; uniaxial mag- netocrystalline anisotropy field Ba i = 2K u/M satmzez, with K u the anisotropy constant; and the field gener- ated by the Dzyaloshinskii-Moriya interaction Bdmi = 2D/M sat(∂mz/∂x, ∂mz/∂y,−∂mx/∂x− ∂my/∂y)T , with D the strength of the interfacial Dzyaloshinskii-Moriya in- teraction. The simulation parameters were determined ex- perimentally in Ref.7 based on the field dependence of the skyrmion profile: M sat = 1.1 MA/m, D = 3.9 mJ/m2, Aexch = 2 pJ/m, K u = 2.5 MJ/m3 and α = 0.05. Additionally, an effective thermal field is included in Eq. (3) i as (cid:115) Bth i = η 2αkBT M satγ0∆V ∆t , (4) where η is a random vector generated according to a standard normal distribution independently for each simulation cell and time step. kB is Boltzmann's constant, T is the temperature, ∆V is the size of the simulation cell and ∆t is the time step of simulation. During the simulations, the cell size was set to ∆V = 0.3 × 0.3 × 0.4 nm3. The linear size of the cell is comparable to the lattice constant a = 0.271 nm of the Pd/Fe bilayer on Ir(111). This means that the micromagnetic simulations per- formed here should closely resemble the results of atomistic simulations where the skyrmions collapse when their size be- comes comparable to the lattice constant37,39, and where the use of temperature-independent model parameters is justified. The total skyrmion number N Sk in the simulations was cal- culated via (cid:90) (cid:18) ∂m ∂x m · × ∂m ∂y (cid:19) dxdy . (5) DISCUSSION In this paper we studied the temperature-driven diffusive motion of ensembles of magnetic skyrmions in finite mag- netic islands. Based on experimentally determined system parameters for ultrahin Pd/Fe bilayers on Ir(111), we per- formed full-fledged micromagnetic simulations for various nanoisland geometries. For moderate temperatures and exter- nal magnetic fields, magnetic skyrmions with a long lifetime are present in the system, so that the topological charge is con- served. Our study of this model system showed two different time scales on which the stochastic dynamics of skyrmions takes place. On the short time scale (t ∼ 1 ps), the fluctu- ating dynamics governed by the symmetric and antisymmet- ric exchange interaction is most prominent, which essentially leads to local deformations, exciting internal modes of the skyrmions. In the long-time limit (t (cid:38) 10 ns), interactions between pairs of skyrmions, as well as between the quasi- particles and the boundaries, are dominant. These thermally activated mechanisms lead to complicated time-averaged pic- tures of the skyrmion distribution, where the number of quasi- particles in the system is not immediately apparent. These results can be compared directly to images obtained from time-integrating measurement techniques such as SP-STM or MFM, the time-resolution of which (∼ 5 ms37) is typically much longer than the timescale of the inherent dynamics of skyrmions. Our findings open an alternative way compared to conventional interpretations of experimental results obtained with scanning probe methods on mobile arrays of skyrmions, by differentiating between the various time scales of magneti- zation dynamics. Furthermore, we proposed and developed a Monte Carlo quasiparticle model, which can be of great use to efficiently calculate the stochastic motion of larger systems of skyrmions in arbitrarily shaped geometries. According to the results, the time-averaged images of the micromagnetic simulations can be qualitatively well reproduced by the probability density distributions obtained from the simple quasiparticle model, as long as it is possible to treat the skyrmions as stable ob- jects with lifetimes significantly longer than the simulation time at the given temperature. This method can be used to describe larger systems containing more skyrmions at vari- ous temperatures in a computationally efficient way, including model applications in storage technology like racetrack mem- ory devices. Here, not only the stability of the information bit is important, but also its addressability, which is immediately affected by the diffusive motion of the skyrmions. METHODS NSk = 1 4π Micromagnetic simulations. Finite-temperature mi- cromagnetic calculations, using the open-source, GPU- accelerated software package mumax340, were performed to Processing of the micromagnetic results. For the results shown in Fig. 6, the z component of the magnetization aver- aged up to time t is stored in grayscale pictures as a matrix, Aij(t) = (mz(xi, yj, t) + 1)/2, where xi and yj denote the position of the micromagnetic simulation cell. The distance induced by the Frobenius norm between the image averaged up to time t, Aij(t), and the image averaged over the whole simulation length τ = 20 ns, Aij(τ ) is divided by the square root of the number of matrix elements and subtracted from 1, yielding the matching parameter M (t) = 1 − (Aij(t) − Aij(τ ))2 NxNy . (6) (cid:118)(cid:117)(cid:117)(cid:116) Nx(cid:88) Ny(cid:88) i=1 j=1 This procedure gives a matching parameter of 1 for a perfect agreement of the compared pictures. In order to determine the skyrmion radius as a function of time in Fig. 7, firstly the contour lines where the z component of the magnetization is zero are calculated for the skyrmion at each time step. The contour lines are discretized on N points in space, Li(t) = (xi(t), yi(t)), i ∈ (1, ..., N ). Subsequently, the center of the skyrmion is calculated via Li(t) , (7) N(cid:88) i=1 c(t) = 1 N N(cid:88) i=1 which in turn is used to obtain the average skyrmion radius ¯r(t) = 1 N Li(t) − c(t) . (8) Finally, the radius is normalized with respect to the zero tem- perature radius r0 obtained from the relaxed initial configura- tion of choice. Calculation of the skyrmion-skyrmion and skyrmion- boundary interaction potentials. For the calculation of the potential, the total energy of a stripe-shaped model system (75 × 30 × 0.4 nm3) was investigated with micromagnetic simulations for two different cases. In the first case, Neu- mann boundary conditions were used along the long side, and a homogeneous magnetization was relaxed leading to a canted rim. Afterwards a skyrmion was added to the system at a cer- tain distance from the boundary. Because of the repulsive na- ture of the interaction, a standard relaxation of the skyrmion position was not suitable for determining the distance depen- dence. Instead, we fixed the central magnetic moments in the skyrmion (in a circular region of 0.9 nm diameter) at a specific position and relaxed the magnetic texture in the other cells. Analogously, in the second case the skyrmion-skyrmion inter- action was calculated by fixing one skyrmion and changing the position of the other skyrmion, using periodic boundary conditions along both directions in the plane. From the ob- tained total energies we subtracted a reference spectrum for a homogeneously magnetized periodic surrounding around the skyrmion fixed at different positions in the mesh. Small devi- ations arising due to the finite mesh are mostly canceled out by this procedure. The results for the interaction strengths are shown in Fig. 11. A smooth energy function rapidly decaying with 9 increasing distance between skyrmion and boundary is ob- tained. For small distances between the boundary and a skyrmion or between two skyrmions, a deformation of the quasiparticles is visible in the micromagnetic simulations, also leading to a bending of the potential in Fig. 11. Dur- ing the construction of the quasiparticle model it is assumed that the thermal energy kBT is lower than the energy value where the potential function starts to bend, which is around 5 meV (60 K) for the skyrmion-skyrmion interaction potential in Fig. 11. In this regime it can be assumed that for suffi- ciently short simulation times the strong deformation of the skyrmions due to the interactions becomes exceedingly rare, and the basic spin structure is maintained. This temperature regime is in reasonable qualitative agreement with the range where the escape, collapse or creation of skyrmions may also be excluded, discussed in detail in Fig. 3. It was shown in Ref.33 that the interaction potentials in this regime may be approximated by the exponential model function E(r) = a exp(−r/b). This is confirmed by the simulation results in Fig. 11, with the fitted model parameters a = 0.211 meV, b = 1.343 nm for skyrmion-boundary (Sk-Bnd) repulsion and a = 1.246 meV, b = 1.176 nm for skyrmion-skyrmion (Sk- Sk) interaction. The similar characteristic length scales b be- tween the two cases support that the same physical mecha- nism is responsible for the interactions, namely the formation of chiral noncollinear spin structures due to the DMI as dis- cussed in the main text. The different values of the parameter a can be reformulated as a shift of the exponential function along the r axis. While this distance is measured between two magnetization cells pointing oppositely to the homogeneous background in the case of two skyrmions, in the case of the skyrmion-boundary repulsion it is measured between the cen- ter of the skyrmion and the magnetization at the edge which is only slightly tilted from the homogeneous background, mean- ing that the same interaction strength is reached at a smaller distance in the latter scenario. Quasiparticle simulations. For the calculation of the stochastic motion of skyrmions following the quasiparticle ap- proach, the Metropolis algorithm42 was utilized. With this method, the probability density function converges to the Boltzmann distribution determined by the energy functional of the system. For the triangular geometry, the potential sur- face was computed by taking the superposition of the poten- tials shown in Fig. 11 from the three boundaries for every grid point of the chosen finite, rectangular mesh. The cell size was ∆V = 0.5 × 0.5 × 0.4 nm. The initial positions of the skyrmions were randomly chosen inside the confined struc- ture, excluding the case in which both quasiparticles start from the same simulation cell. At each time step, a possible adja- cent position is selected for each skyrmion simultaneously via pseudo-random numbers. If the energy of the attempted new state, including interaction between the skyrmions, is lower than the energy of the initial one, the skyrmion will move there. If not, the transition into this state happens with a prob- ability of p(∆E) = exp(−∆E/kBT ) following the Boltz- mann distribution. Here ∆T = E2 − E1 is the energy dif- ference between the final and the initial states, kB is Boltz- mann's constant and T is the temperature. Subsequently, new attempted positions are generated and accepted or refused as before until a fixed number of simulation steps is reached. 10 FIG. 11. Potential energy depending on the skyrmion-skyrmion and skyrmion-boundary distance. Micromagnetic interaction en- ergies between skyrmion and skyrmion (blue squares), and be- tween skyrmion and boundary (red circles) on a finite nanoisland. Solid lines display exponential fits based on the model function E(r) = ae−r/b; the fitting parameters are skyrmion-boundary (Sk- Bnd): a = 0.211 meV, b = 1.343 nm; skyrmion-skyrmion (Sk-Sk): a = 1.246 meV, b = 1.176 nm. ∗ Corresponding author. [email protected] 1 Bogdanov, A. & Yablonskii, D. Thermodynamically stable vor- tices in magnetically ordered crystals. The mixed state of magnets. Zh. Eksp. Teor. Fiz 95, 178 -- 182 (1989). 2 Muhlbauer, S. et al. Skyrmion lattice in a chiral magnet. Science 323, 915 -- 919 (2009). 3 Nagaosa, N. & Tokura, Y. Topological properties and dynamics of magnetic skyrmions. Nat. Nanotechnol. 8, 899 -- 911 (2013). 4 Dzyaloshinsky, I. A thermodynamic theory of weak ferromag- netism of antiferromagnetics. J. Phys. Chem. Sol. 4, 241 -- 255 (1958). 5 Moriya, T. Anisotropic superexchange interaction and weak fer- romagnetism. Phys. Rev. 120, 91 -- 98 (1960). 6 Heinze, S. et al. Spontaneous atomic-scale magnetic skyrmion lattice in two dimensions. Nat. Phys. 7, 713 -- 718 (2011). 7 Romming, N., Kubetzka, A., Hanneken, C., von Bergmann, K. & Wiesendanger, R. Field-dependent size and shape of single magnetic skyrmions. Phys. Rev. Lett. 114, 177203 (2015). 8 Bogdanov, A. & Hubert, A. The stability of vortex-like structures in uniaxial ferromagnets. J. Magn. Magn. Mater. 195, 182 -- 192 (1999). 9 Parkin, S. S., Hayashi, M. & Thomas, L. Magnetic domain-wall racetrack memory. Science 320, 190 -- 194 (2008). 10 Sampaio, J., Cros, V., Rohart, S., Thiaville, A. & Fert, A. Nu- cleation, stability and current-induced motion of isolated mag- netic skyrmions in nanostructures. Nat. Nanotechnol. 8, 839 -- 844 (2013). 11 Tomasello, R. et al. A strategy for the design of skyrmion race- track memories. Sci. Rep. 4, 6784 (2014). 12 Romming, N. et al. Writing and deleting single magnetic skyrmions. Science 341, 636 -- 639 (2013). 13 Schaffer, A. F., Durr, H. A. & Berakdar, J. Ultrafast imprinting of topologically protected magnetic textures via pulsed electrons. Appl. Phys. Lett. 111, 032403 (2017). 14 Hanneken, C. et al. Electrical detection of magnetic skyrmions by non-collinear magnetoresistance. Nat. Nanotechnol. 10, 1039 -- 1042 (2015). 15 Gobel, B., Mook, A., Henk, J. & Mertig, I. Magnetoelectric ef- fect and orbital magnetization in skyrmion crystals: Detection and characterization of skyrmions. Phys. Rev. B 99, 060406 (2019). 16 Maccariello, D. et al. Electrical detection of single magnetic skyrmions in metallic multilayers at room temperature. Nat. Nan- otechnol. 13, 233 -- 237 (2018). 17 Slonczewski, J. C. Current-driven excitation of magnetic multi- layers. J. Magn. Magn. Mater. 159, L1 -- L7 (1996). 18 Iwasaki, J., Mochizuki, M. & Nagaosa, N. Universal current- velocity relation of skyrmion motion in chiral magnets. Nat. Com- mun. 4, 1463 (2013). 19 Jonietz, F. et al. Spin transfer torques in MnSi at ultralow current densities. Science 330, 1648 -- 1651 (2010). 20 Hagemeister, J., Romming, N., von Bergmann, K., Vedmedenko, E. & Wiesendanger, R. Stability of single skyrmionic bits. Nat. Commun. 6, 8455 (2015). 21 R´ozsa, L., Simon, E., Palot´as, K., Udvardi, L. & Szunyogh, L. Complex magnetic phase diagram and skyrmion lifetime in an ul- trathin film from atomistic simulations. Phys. Rev. B 93, 024417 (2016). 22 Lobanov, I. S., J´onsson, H. & Uzdin, V. M. Mechanism and activa- tion energy of magnetic skyrmion annihilation obtained from min- imum energy path calculations. Phys. Rev. B 94, 174418 (2016). 23 Bessarab, P. F. et al. Lifetime of racetrack skyrmions. Sci. Rep. 8, 3433 (2018). 24 von Malottki, S., Dup´e, B., Bessarab, P., Delin, A. & Heinze, S. Enhanced skyrmion stability due to exchange frustration. Sci. Rep. 7, 12299 (2017). 25 Schutte, C., Iwasaki, J., Rosch, A. & Nagaosa, N. Inertia, diffu- sion, and dynamics of a driven skyrmion. Phys. Rev. B 90, 174434 (2014). 26 Miltat, J., Rohart, S. & Thiaville, A. Brownian motion of mag- netic domain walls and skyrmions, and their diffusion constants. Phys. Rev. B 97, 214426 (2018). 27 Z´azvorka, J. et al. Thermal skyrmion diffusion used in a reshuffler device. Nat. Nanotechnol. 1 (2019). 28 Lin, S.-Z., Reichhardt, C., Batista, C. D. & Saxena, A. Particle model for skyrmions in metallic chiral magnets: Dynamics, pin- ning, and creep. Phys. Rev. B 87, 214419 (2013). 29 Reichhardt, C., Ray, D. & Reichhardt, C. J. O. Collective transport properties of driven skyrmions with random disorder. Phys. Rev. Lett. 114, 217202 (2015). 30 Thiele, A. A. Steady-state motion of magnetic domains. Phys. Rev. Lett. 30, 230 -- 233 (1973). 31 Wiesendanger, R. Spin mapping at the nanoscale and atomic scale. Rev. Mod. Phys. 81, 1495 -- 1550 (2009). 32 Martin, Y. & Wickramasinghe, H. K. Magnetic imaging by force microscopywith 1000 A resolution. Appl. Phys. Lett. 50, 1455 -- 1457 (1987). 33 Lin, S.-Z., Reichhardt, C., Batista, C. D. & Saxena, A. Particle model for skyrmions in metallic chiral magnets: Dynamics, pin- ning, and creep. Phys. Rev. B 87, 214419 (2013). 34 R´ozsa, L. et al. Skyrmions with attractive interactions in an ultra- thin magnetic film. Phys. Rev. Lett. 117, 157205 (2016). 35 Bottcher, M., Heinze, S., Egorov, S., Sinova, J. & Dup´e, B. B -- T phase diagram of Pd/Fe/Ir (111) computed with parallel tempering Monte Carlo. New J. Phys. 20, 103014 (2018). 36 Gilbert, T. L. A phenomenological theory of damping in ferro- magnetic materials. IEEE Trans. Magn. 40, 3443 -- 3449 (2004). 37 Leonov, A. et al. The properties of isolated chiral skyrmions in thin magnetic films. New J. Phys. 18, 065003 (2016). 38 R´ozsa, L., Hagemeister, J., Vedmedenko, E. Y. & Wiesendanger, R. Localized spin waves in isolated k π skyrmions. Phys. Rev. B 98, 224426 (2018). 39 Siemens, A., Zhang, Y., Hagemeister, J., Vedmedenko, E. & Wiesendanger, R. Minimal radius of magnetic skyrmions: stat- ics and dynamics. New J. Phys. 18, 045021 (2016). 40 Vansteenkiste, A. et al. The design and verification of MuMax3. AIP Advances 4, 107133 (2014). 11 41 Rohart, S. & Thiaville, A. Skyrmion confinement in ultrathin film nanostructures in the presence of Dzyaloshinskii-Moriya interac- tion. Phys. Rev. B 88, 184422 (2013). 42 Metropolis, N. et al. Equation of state calculations by fast com- puting machines. J. Chem. Phys. 21, 1087 -- 1092 (1953). ACKNOWLEDGEMENTS Financial support was provided by the Deutsche Forschungsgemeinschaft (DFG) via CRC/TRR 227 and SFB 762, by the European Union via the Horizon 2020 research and innovation program under Grant Agreement No. 665095 (MAGicSky), by the Alexander von Humboldt Foundation, and by the National Research, Development and Innovation Office of Hungary under Project No. K115575. AUTHOR CONTRIBUTIONS A.F.S. conducted and analyzed the micromagnetic and quasiparticle Monte Carlo simulations. E.Y.V. proposed and developed a general concept of this investigation. All authors discussed the results and contributed to the manuscript. COMPETING INTERESTS STATEMENT The authors declare no competing financial interests. DATA AVAILABILITY The code and the data that support this work's findings are available from the corresponding author on request. 

1608.03879

1

1608

Magnetotransport signatures of the proximity exchange and spin-orbit couplings in graphene

Graphene on an insulating ferromagnetic substrate---ferromagnetic insulator or ferromagnetic metal with a tunnel barrier---is expected to exhibit giant proximity exchange and spin-orbit couplings. We use a realistic transport model of charge-disorder scattering and solve the linearized Boltzmann equation numerically exactly for the anisotropic Fermi contours of modified Dirac electrons to find magnetotransport signatures of these proximity effects: proximity anisotropic magnetoresistance, inverse spin-galvanic effect, and the planar Hall resistivity. We establish the corresponding anisotropies due to the exchange and spin-orbit couplings, with respect to the magnetization orientation. We also present parameter maps guiding towards optimal regimes for observing transport magnetoanisotropies in proximity graphene.

cond-mat.mes-hall

cond-mat

a Magnetotransport signatures of the proximity exchange and spin-orbit couplings in graphene Jeongsu Lee∗ and Jaroslav Fabian Institute for Theoretical Physics, University of Regensburg, 93040 Regensburg, Germany Graphene on an insulating ferromagnetic substrate -- ferromagnetic insulator or ferromagnetic metal with a tunnel barrier -- is expected to exhibit giant proximity exchange and spin-orbit cou- plings. We use a realistic transport model of charge-disorder scattering and solve the linearized Boltz- mann equation numerically exactly for the anisotropic Fermi contours of modified Dirac electrons to find magnetotransport signatures of these proximity effects: proximity anisotropic magnetoresis- tance, inverse spin-galvanic effect, and the planar Hall resistivity. We establish the corresponding anisotropies due to the exchange and spin-orbit coupling, with respect to the magnetization orien- tation. We also present parameter maps guiding towards optimal regimes for observing transport magnetoanisotropies in proximity graphene. PACS numbers: 72.20.Dp 72.80.Vp 73.22.Pr 73.63.-b Dirac electrons in pristine graphene have wesizableak spin-orbit coupling [1] and no magnetic moments, lim- iting prospects for spintronics [2]. This can be partly remedied by functionalizing graphene with adatoms and admolecules, which can induce sizable local magnetic mo- ments and spin-orbit coupling, leading to marked spin transport fingerprints [3 -- 8]. A more systematic and, im- portant, spatially uniform way to induce spin properties in graphene is by proximity effects. Being essentially a surface (or two), graphene can "borrow" properties from its substrates. Placing graphene on a slab of a ferromag- netic insulator, or a ferromagnetic metal with a tunnel barrier, is expected to induce giant proximity exchange as well as spin-orbit coupling in the Dirac electron band structure. This is supported by first principles calcula- tions [9 -- 13] as well as by recent experiments on graphene on yttrium iron garnet [14, 15] and on graphene on EuS [16]. In effect, proximity graphene on ferromagnetic sub- strates should be an ultimately thin ferromagnetic layer, with giant spin-orbit coupling, forming a perfect play- ground for both spintronics experiment and theory [17]. An important question is: what transport ramifica- tions can we expect in such a magnetic graphene with strong spin-orbit coupling? On one hand, in ferromag- netic metals the exchange coupling is typically much greater than spin-orbit coupling. On the other hand, in semiconductor heterostructures, which are the best case studies for structure-induced spin-orbit coupling in its transport signatures [18, 19], there is no ferromag- netic exchange and spin splitting can be due to the Zee- man interaction which is, for realistic values of magnetic field, much weaker than spin-orbit coupling. Proxim- ity graphene should be intermediate between those two extremes: the proximity exchange and spin-orbit cou- plings are expected to be similar, on the order of 1 - 10 meV [9, 10, 20]. Perhaps the main effect of the interplay of exchange and spin-orbit couplings -- magnetotransport anisotropies -- should be well pronounced and make for useful, experimentally testable signatures of the spin proximity effects. In this paper we solve numerically exactly a realistic Boltzmann transport model, with long-range charge scat- terers, for Dirac electrons in the presence of proximity exchange and spin-orbit couplings. We start with the anisotropic band structure, as in Fig. 1, and explore its ramifications in transport. Specifically, we introduce and calculate the proximity anisotropic magnetoresistance, as an analog of the anisotropic lateral magnetoresistance in ferromagnetic metal/insulator slabs [21], characteriz- ing interfacial spin-orbit fields. We also present magne- toanisotropies of the planar Hall effect and inverse spin- galvanic effect. Finally, we give parameter maps indicat- FIG. 1. (Color online). Scheme of magnetoanisotropic trans- port experiment in proximity graphene. Polar (θ) and az- imuthal (φ) angles define the magnetization orientation with respect to the applied electric field. (a) Linear energy dis- persion of pristine graphene can be modified by (b) (intrin- sic and Bychkov-Rashba) spin-orbit coupling or (c) exchange field, both leading to spin splitting. (d) The interplay of the two interactions makes the bands anisotropic with respect to the magnetization orientation, here shown as out-of-plane and in-plane. ESOC (a) intrinsic(b) proximity SOC(c) proximity exchange(d) SOC and exchangeEEESOC + exchange exchange Ein-planeout-of-planeMagnetic Insulator mI ✓EmzkEkcontours shifted relative to each other. 2 To investigate electrical transport, we solve the lin- earized Boltzmann equation for the above model, assum- ing spatial homogeneity. In the presence of a longitudinal electric field E, the non-equilibrium distribution function is f = f0 + δf , where f0 is the equilibrium Fermi-Dirac function. We use the ansatz δf = −e(−∂f0/∂E)u·E and consider long-range Coulomb scattering, which is the es- tablished model for resistivity in graphene [23]. The un- known vector u is found by solving the integral equation (obtained from the Boltzmann equation in linear order in E), v(k) = 2πni(vF rs)2 ×(cid:72) (cid:48) dk ∇k(cid:48) Ek(cid:48) F (k,k q2ε(q)2 [u(k) − u(k(cid:48))] , (cid:48) ) EF (2) where Ek is the energy corresponding to wave vector k, v(k) is the group velocity, ni is the concentration of scat- terers, the effective fine structure constant rs ≈ 0.8 [24], F (k, k(cid:48)) = Ψ(k)†Ψ(k(cid:48))2 is the overlap between the in- cident (k) and scattered (k(cid:48)) states Ψ. For example, for pristine graphene F (k, k(cid:48)) = (1 + cos θkk(cid:48))/2. For sim- plicity, spin and pseudospin indices are implicit in the momentum labels k. The integral is over the Fermi con- tour of Fermi energy EF , and the transferred momentum is q = k−k(cid:48). The dielectric function ε is calculated from the random phase approximation [24 -- 26]. ε(q) =(cid:26) 1+ qs q 1+ πrs 2 + qs q − qs √ q2−4k2 2q2 F −rs sin −1( 2kF if q≤2kF q ) if q>2kF ,(3) where qs = 4kF rs. The Fermi wave vector kF is taken from the pristine graphene case corresponding to a given electron density. The integral equation, Eq. (2), is solved numerically exactly [27], taking the energy spectrum and eigenstates of the effective hamiltonian, Eq. (1). Knowing vectors u for the Fermi contour momenta k, the conductivity tensor is obtained from, viujδ(Ek − EF ). (4) (cid:90) dk 2π σij = e2 h We plot the calculated longitudinal conductivity of graphene as a function of carrier density n, with and without proximity effects, in Fig. 3(a). We use ni = 80× 1010 cm−2 as a representative density of long-range scat- terers. The carrier density, unlike in pristine graphene, depends not only on the Fermi level but also on the strength of the proximity interactions, λI and λex, n(EF ) = 2 × 1 2π F + 2EF λI + λ2 ex (5) (cid:3) /(vF ). (cid:2)E2 The factor 2 takes into account the valley degeneracy. The carrier density is independent of λBR and the di- rection of the magnetization. In all the plots we fix the (Color online). Fermi contours and spin texture FIG. 2. (left), and the band structure along kx (middle) and ky (right) for different directions of the exchange field. The direc- tion of the exchange field is indicated by the large arrows; small arrows give the spin projections. (a) The out-of-plane (OOP) exchange field separates two spin subbands, while the Bychkov-Rashba field leads to a distinctive spin texture de- pending on the z-projection of the real spin, which interacts with exchange field. (b) The in-plane (IP) exchange field splits the bands, but also deforms the Fermi circles. We used EF = 100 meV, λI = 0 meV, λBR = 10 meV, and λex = 25 meV. ing regions of large transport magnetoanisotropies. Dirac electrons in graphene in the presence of proxim- ity exchange and spin-orbit couplings are described by the minimal Hamiltonian [17], H = H0 + HI + HBR + Hex. (1) pristine Here, graphene Hamiltonian is H0 = vF (τzσxkx + σyky) with pseudospin (sublattice) Pauli matrices σσσ and τz = ±1 for K and K(cid:48) points. The Fermi velocity is vF = (3/2)ta0 ≈ 8.6 × 107 cm/s for t = 2.7 eV and the inter-atomic distance of carbons in graphene a0 = 1.42 A[22]. The proximity intrinsic-like spin-orbit coupling is given by the Hamiltonian HI = λI τzσzsz, with parameter λI and (true) spin Pauli matrices s. The intrinsic coupling opens a gap of 2λI . The Bychkov- Rashba Hamiltonian, HBR = λBR(τzσxsy − σysx), with parameter λBR describes the proximity spin-splitting due to spin-orbit coupling and lack of space inversion sym- metry. Finally, the spin-dependent hybridization with the ferromagnet leads to a proximity exchange, Hex = λexm·s with parameter λex and magnetization orienta- tion m. There are two important magnetic configurations to consider: out-of-plane and in-plane magnetizations, de- picted in Fig. 2 (see also Fig. 1). In the out-of-plane case the spin up and spin down bands are spin split, but the band structure (and thus Fermi contour) remains isotropic. On the other hand, in the in-plane case the band structure is markedly anisotropic, with the Fermi kxkxkykykxkxkyky3 (Color online). FIG. 4. (a) Calculated proximity induced anisotropic magnetoresistance (PAMR) and (b) planar Hall resistivity are shown when the exchange field is in-plane (θ = π/2) for λBR = 20 meV. PAMR quantifies the longitudinal magnetoresistance as a function of magnetization orientation φ. The interplay of spin-orbit coupling and exchange field leads to a net anisotropic resistivity with C2v symmetry while the off diagonal elements of the resistivity tensor are non-zero. Other parameters are n = 1012 cm−2, λI = 0 meV, λex = 10 meV. The insets are polar plot representations. Rashba field, a spin density transverse to the current appears as a demonstration of the inverse spin-galvanic effect [28 -- 30] . The shift of the spin subbands due to the electric field, combined with the spin texture due to the Bychkov-Rashba spin-orbit interaction, leads to a spin polarization. This is also expected to happen in graphene [31, 32], along with the spin-galvanic effect [33, 34]. The non-equilibrium spin density caused by the electric field can be calculated as, δ(cid:104)S(cid:105) = (cid:90) dk (2π)2 δf (k)s(k), (6) where s(k) is the spin (represented by Pauli matrices s) expectation value of the state k. For our proximity model in the presence of long-range Coulomb scatterers, the cal- culated inverse spin-galvanic effect is shown in Fig. 3(b) as a function of exchange coupling. With increasing mag- netization the induced transverse spin is reduced, as the exchange coupling aligns the spins and deforms the ro- tational spin texture of the Bychkov-Rashba field. The spin densities can be giant. In fields of 1 V/µm, which FIG. 3. (a) (Color online). Calculated longitudinal conduc- tivity as a function of carrier density, for pristine and proxim- ity graphene. For proximity graphene we show the conductiv- ity in the presence of Bychkov-Rashba and exchange coupling only, and in the presence of intrinsic spin-orbit coupling only. (b) Inverse spin-galvanic effect (scheme in the left inset) in proximity graphene. Spin density induced (and normalized) by electric field (which is along x-axis) with respect to the exchange interaction, when the magnetization is out-of-plane (OOP) and in-plane (IP). In-plane magnetization can be ei- ther parallel (along x-ais) or perpendicular (along y-axis) to the electric field E. In the right inset, the angle dependance of Sy(φ) is shown in the polar plot with ∆Sy = Sy−Smin , for the electric field E = 1 V/cm, and for λex = 10 meV in percent- y = Sy(EM). The carrier density in age with respect to Smax this plot is n = 1012 cm−2 and λI = 0 meV, λBR = 20 meV. y carrier density, instead of the Fermi level. The conduc- tivity for three different combinations of parameters is shown in Fig. 3(a). The linear dependence on n is well reproduced. While λBR and λex bring about relatively in- significant changes ((cid:46) 2.5 %), the presence of λI =10 meV lowers the conductivity by about ∼10 % at a fixed car- rier density. Thus, in terms of modifying the magni- tude of the conductivity, the proximity effects (unless not inducing additional scattering, which would need to be investigated case-by-case) are rather weak, being more pronounced with the inclusion of the intrinsic spin-orbit coupling, than with the Bychkov-Rashba and exchange effects. However, as we will see shortly, the anisotropic effects are quite pronounced. When current flows in the presence of a Bychkov- ������������������n/niσxx(e2/h)λI=λBR=λex=0λI=0,λBR=λex=10λI=10,λBR=λex=0ni=80×1010cm-2(a) ������������������������������λex(meV)<Sy>(107cm-2)OOPIP(EM)IP(E⟂M)E=1V/cm(b) Sjinv. spin-galvanicSy/Smaxy(%)10y x �π�π-���-���-���-������ϕPAMR(%)θ=π/���(a) �π�π-�-����ϕρxy(Ω)θ=π/���(b) 4 FIG. 5. (Color online). Parameter maps of PAMR(θ = π/2, φ = π/2). (a) PAMR as a function of λBR and λI , for a fixed λex = 10 meV. (b) PAMR as a function of λBR and λex, for λI = 0. Note that PAMR changes sign around the λBR = λex line. In both maps the carrier density is 1012 cm−2. are still achievable in graphene, the spin density could reach 1011 cm−2, corresponding to about 10% of spin polarization. The largest induced spin accumulation is in the out-of-plane configuration for large exchange. The magnetoanisotropy of the inverse spin-galvanic effect can be very large, as seen in Fig. 3(b). The presence of the current-induced spin accumulation, as well as its mag- netoanisotropy, could be detected in the same proximity structure, by measuring the transverse voltage, as in non- local spin injection [2]. To quantify the transport magnetoanisotropy, we in- troduce proximity induced anisotropic magnetoresistance (PAMR), as a ratio of the resistivities R (or conductiv- ities σ) for a given magnetization orientation (θ, φ) (see Fig. 1), PAMR[E] = = R(θ, φ) − R(θ, 0) σxx(θ, 0) − σxx(θ, φ) R(θ, 0) σxx(θ, φ) , (7) analogously to the tunneling anisotropic magnetoresis- tance effect [35 -- 37]. PAMR refers to the changes in the longitudinal magnetoresistance as the magnetization di- rection varies with respect to the direction of the external electric field that drives the current. When the magne- tization is out-of-plane (θ = 0), the broken time reversal symmetry and strong spin-orbit coupling can lead to the novel quantum anomalous Hall effect [10], or crystalline magnetoanisotropy [21]. However, here we focus on the regime in which PAMR is most pronounced (θ = π/2). As shown in Fig. 4(a), PAMR exhibits a C2v symme- try due to the interplay between the Bychkov-Rashba and exchange interactions. The expected magnitudes of PAMR are about 1%, similar to what is observed in ferro- magnetic metals [21]. The anisotropic resistivity tensor has non-zero off-diagonal elements due to the presence of exchange and spin-orbit couplings. This leads to the planar Hall effect, shown in Fig. 4(b). The magnitude of the planar Hall effect could reach up to 4 Ω which greater than the typical values studied in metallic ferromagnetic systems [38]. What values can PAMR reach for a reasonable range of proximity parameters? Figure 5 shows two parameter maps, one with the Bychkov-Rashba and intrinsic, the other with the Bychkov-Rashba and exchange couplings. We see two distinct features. (i) First, in Fig. 5(a) a horizontal line around λBR ∼ λex ≈ 10 meV separates two regions. For λBR (cid:46) 10 meV, increasing λI increases PAMR. For λBR (cid:38) 10 meV, PAMR initially increases with increasing λI , reaching a maximum of about 1 % around λI ∼ 7 meV, beyond which PAMR decreases. (ii) Second, in Fig. 5(b) the line λBR = λex marks a sharp crossover between weak and strong PAMR. However, this crossover is not uniform. PAMR is largest for large values of both λex and λBR slightly greater than λex. The reason why this region gives the largest PAMR (more than 1%) is that in this parameter range there is a band crossing between the strongly spin-orbit coupled subbands. In conclusion, we used a realistic transport model to predict magnetotransport anisotropies in graphene with proximity exchange and spin-orbit couplings. We predict marked anisotropies in the magnetoresistance, with sim- ilar values as reached in ferromagnetic metal junctions and slabs. The calculated PAMR depends strongly on (a) (b) the spin-orbit coupling and exchange parameters. We also calculated the magnetoanisotropies of the planar Hall and inverse spin-galvanic effects. All these magne- toanisotropies should be a sensitive tool to probe prox- imity effects in graphene. We thank Denis Kochan, Jonathan Eroms, and Cosimo Gorini for useful discussions. This work was supported by the DFG SFB 689 and the European Union Sev- enth Framework Programme under Grant Agreement No. 604391 Graphene Flagship. ∗ [email protected] [1] M. Gmitra, S. Konschuh, C. Ertler, C. Ambrosch-Draxl, and J. Fabian, Phys. Rev. B 80, 235431 (2009). [2] I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). [3] D. Kochan, M. Gmitra, and J. Fabian, Phys. Rev. Lett. 112, 116602 (2014). [4] J. Bundesmann, D. Kochan, F. Tkatschenko, J. Fabian, and K. Richter, Phys. Rev. B 92, 081403 (2015). [5] D. Van Tuan and S. Roche, Phys. Rev. Lett. 116, 106601 (2016). [6] A. Ferreira, T. G. Rappoport, M. A. Cazalilla, and A. H. Castro Neto, Phys. Rev. Lett. 112, 066601 (2014). [7] J. Wilhelm, M. Walz, and F. Evers, Phys. Rev. B 92, 014405 (2015). [8] M. R. Thomsen, M. M. Ervasti, A. Harju, and T. G. Pedersen, Phys. Rev. B 92, 195408 (2015). [9] H. X. Yang, A. Hallal, D. Terrade, X. Waintal, S. Roche, and M. Chshiev, Phys. Rev. Lett. 110, 046603 (2013). [10] Z. Qiao, W. Ren, H. Chen, L. Bellaiche, Z. Zhang, A. H. MacDonald, and Q. Niu, Phys. Rev. Lett. 112, 116404 (2014). [11] P. Lazi´c, K. D. Belashchenko, and I. Zuti´c, Phys. Rev. B 93, 241401 (2016). [12] C. B. Crook, C. Constantin, T. Ahmed, J.-X. Zhu, A. V. Balatsky, and J. T. Haraldsen, Sci. Rep. 5, 12322 (2015). and J. Fabian, [13] K. Zollner, M. Gmitra, T. Frank, arXiv:1607.08008 (2016). [14] Z. Wang, C. Tang, R. Sachs, Y. Barlas, and J. Shi, Phys. Rev. Lett. 114, 016603 (2015). [15] J. C. Leutenantsmeyer, A. A. Kaverzin, M. Wojtaszek, and B. J. van Wees, arXiv:1601.00995 (2016). [16] P. Wei, S. Lee, F. Lemaitre, L. Pinel, D. Cutaia, W. Cha, F. Katmis, Y. Zhu, D. Heiman, J. Hone, J. S. Moodera, 5 and C.-T. Chen, Nat. Mater. advance online publication, DOI: 10.1038 (2016). [17] W. Han, R. K. Kawakami, M. Gmitra, and J. Fabian, Nat. Nanotechnol. 9, 794 (2014). [18] M. Trushin and J. Schliemann, Phys. Rev. B 75, 155323 (2007). [19] O. Chalaev and D. Loss, Phys. Rev. B 77, 115352 (2008). [20] A. Avsar, J. Y. Tan, T. Taychatanapat, J. Balakrishnan, G. K. W. Koon, Y. Yeo, J. Lahiri, A. Carvalho, A. S. Rodin, E. C. T. O'Farrell, G. Eda, A. H. Castro Neto, and B. Ozyilmaz, Nat. Commun. 5, 4875 (2014). [21] T. Hupfauer, A. Matos-Abiague, M. Gmitra, F. Schiller, and J. Loher, D. Bougeard, C. H. Back, J. Fabian, D. Weiss, Nat. Commun. 6, 7374 (2015). [22] N. M. R. Peres, Rev. Mod. Phys. 82, 2673 (2010). [23] S. Adam, E. H. Hwang, V. M. Galitski, and S. Das Sarma, Proc. Natl. Acad. Sci. U.S.A. 104, 18392 (2007). [24] E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 (2007). [25] E. H. Hwang, S. Adam, and S. Das Sarma, Phys. Rev. Lett. 98, 186806 (2007). [26] E. H. Hwang and S. Das Sarma, Phys. Rev. B 77, 195412 (2008). [27] We discretize the Fermi contour (100 to 500 points usu- ally suffice) and solve the resulting sets of linear equations algebraically. [28] S. D. Ganichev, E. L. Ivchenko, V. V. Bel'kov, S. A. Tarasenko, M. Sollinger, D. Weiss, W. Wegscheider, and W. Prettl, Nature (London) 417, 153 (2002). [29] S. Ganichev and W. Prettl, Intense terahertz excitation of semiconductors (Oxford University, New York, 2006). [30] S. D. Ganichev and L. E. Golub, Phys. Status Solidi B 251, 1801 (2014). [31] A. Dyrda(cid:32)l, M. Inglot, V. K. Dugaev, and J. Barna´s, Phys. Rev. B 87, 245309 (2013). [32] A. Dyrda(cid:32)l, J. Barna´s, and V. K. Dugaev, Phys. Rev. B 89, 075422 (2014). [33] Ka Shen, G. Vignale, and R. Raimondi, Phys. Rev. Lett. 112, 096601 (2014). [34] J. S´anchez, L. Vila, G. Desfonds, S. Gambarelli, J. P. At- tan´e, J. M. De Teresa, C. Mag´en, and A. Fert, Nat. Commun. 4, 2944 (2013). [35] A. Matos-Abiague, M. Gmitra, and J. Fabian, Phys. Rev. B 80, 045312 (2009). [36] A. Matos-Abiague and J. Fabian, Phys. Rev. B 79, 155303 (2009). [37] L. Brey, C. Tejedor, and J. Fern´andez-Rossier, Appl. Phys. Lett. 85, 1996 (2004). [38] H. X. Tang, R. K. Kawakami, D. D. Awschalom, and M. L. Roukes, Phys. Rev. Lett. 90, 107201 (2003). 

1311.5678

1

1311

Electronic and Transport Properties of Molecular Junctions under a Finite Bias: A Dual Mean Field Approach

We show that when a molecular junction is under an external bias, its properties can not be uniquely determined by the total electron density in the same manner as the density functional theory (DFT) for ground state (GS) properties. In order to correctly incorporate bias-induced nonequilibrium effects, we present a dual mean field (DMF) approach. The key idea is that the total electron density together with the density of current-carrying electrons are sufficient to determine the properties of the system. Two mean fields, one for current-carrying electrons and the other one for equilibrium electrons can then be derived.By generalizing the Thomas-Fermi-Dirac (TFD) model to non-equilibrium cases, we analytically derived the DMF exchange energy density functional. We implemented the DMF approach into the computational package SIESTA to study non-equilibrium electron transport through molecular junctions. Calculations for a graphene nanoribbon (GNR) junction show that compared with the commonly used \textit{ab initio} transport theory, the DMF approach could significantly reduce the electric current at low biases due to the non-equilibrium corrections to the mean field potential in the scattering region.

cond-mat.mes-hall

cond-mat

Electronic and Transport Properties of Molecular Junctions under a Finite Bias: A Dual Mean Field Approach Shuanglong Liu1, Yuan Ping Feng1, and Chun Zhang1,2∗ 1Department of Physics and Graphene research centre, National University of Singapore, 2 Science Drive 3, Singapore 117542 2Department of Chemistry, National University of Singapore, 3 Science Drive 3, Singapore 117543 We show that when a molecular junction is under an external bias, its properties can not be uniquely determined by the total electron density in the same manner as the density functional theory (DFT) for ground state (GS) properties. In order to correctly incorporate bias-induced nonequilibrium effects, we present a dual mean field (DMF) approach. The key idea is that the total electron density together with the density of current-carrying electrons are sufficient to determine the properties of the system. Two mean fields, one for current-carrying electrons and the other one for equilibrium electrons can then be derived. Calculations for a graphene nanoribbon (GNR) junction show that compared with the commonly used ab initio transport theory, the DMF approach could significantly reduce the electric current at low biases due to the non-equilibrium corrections to the mean field potential in the scattering region. Since the pioneering work of Aviram and Ratner1, molecular electronics has attracted a great deal of in- terests due to its promise for future electronics technol- ogy. The central topic in theoretical research of molecular electronics is to understand the quantum electron trans- port at the molecular level by relating the electric cur- rent passing through the molecular junction to its intrin- sic electronic properties. The commonly used ab initio approach combines the quantum transport theory based on non-equilibrium Green's function (NEGF) techniques and the computational method based on density func- tional theory (DFT).2,3 The approach has been applied to describe the quantum electron transport through various types of molecular junctions.4 -- 8, and great success has been achieved in understanding quantum electron trans- port and also inspiring novel applications in molecular electronics. Despite its great success, there are still two problems in the current approach: 1) whether or not the DFT is good enough for molecular junctions under a finite bias is questionable, and 2) when quantitatively compared with experiments, in most cases, theoretically calculated elec- tric current is significantly higher. For some molecular junctions, there might be orders of magnitude differences between experiment and theory. 4,5,9 In this paper, we examine in details the bias-induced noneqiulibrium ef- fects in molecular junctions that are neglected in the current approach and propose a new ab initio method to incorporate these nonequilibrium effects. Molecular junctions connected with two reservoirs can be modeled as open systems with the general structure shown in Fig. 1. The system is divided into three re- gions: left, right reservoirs, and the molecular-scale scat- tering region in the middle. Electrons in two reservoirs are assumed to be in their local equilibrium and can be described by single-electron or mean-field Hamiltonians ∗Electronic address: [email protected] Scattering Region Left Reservoir Right Reservoir Molecular Center FIG. 1: A molecular scale junction consists of an interact- ing scattering region and two non-interacting reservoirs. At t = −∞, an infinitely high single-electron barrier potential (the gray line) is applied at the left contact so that no elec- tric current is flowing through. The barrier potential slowly decreases to zero from t = −∞ to t = 0. The system reaches the desired non-equilibrium steady state at t = 0. (HL and HR for left and right reservoirs respectively) so that two constant chemical potentials µl and µr can be defined in left and right reservoirs respectively. The bias voltage across the system can then be defined by the dif- ference between chemical potentials in two reservoirs as Vb = (µl − µr)/e. Electrons in the scattering region are described by an interacting Hamiltonian HS. The total Hamiltonian for such an open system can be written as H = HL + HR + HS + HT where HT is the tunneling term. The goal of the transport theory is to relate the electronic and transport properties of the steady state of the system to the Hamiltonian and the voltage bias, which is extremely difficult (if not impossible) with the presence of the complicated interaction term HS in the scattering region. For simplicity, we set the temperature to be zero throughout the paper. When µl equals µr, the whole sys- tem is in equilibrium; the DFT is applicable, and then the interaction term HS can be replaced by the mean- field DFT Hamiltonian. Using NEGF techniques, the mean-field Schrodinger equation together with the open- system boundary conditions can be solved,10,11 and in turn the electronic properties can be worked out. In practice, due to the problems of existing DFT function- als and also the fact that the conductance calculations rely on single-electron orbital, calculations may not be accurate.12 -- 19 When µl is not the same as µr, the sit- uation however is different. In this case, the scattering region is driven out of equilibrium, and whether or not its electronic structures in principle can be described by DFT needs to be carefully examined. Theoretically, the non-equilibrium steady state of the junction can be achieved by the following adiabatic time- dependent process. At beginning t = −∞, an infinitely high single-electron barrier potential is applied at the left contact as shown in Fig. 1 so that there is no electric cur- rent flowing through the system. The two parts separated by the barrier can be denoted as LB (the part on the left of the barrier) and RB (the part on the right of the bar- rier). The coupling between LB and RB is then gradually turned on by slowly decreasing the barrier potential to zero from t = −∞ to t = 0. The final state at t = 0 is the non-equilibrium steady state we desire. For such a time- dependent system, the Runge-Gross (RG) theorem, the foundation of time-dependent DFT,20 claims that the ex- ternal potential at time t, vext(r, t), can be uniquely de- termined by the time-dependent electron density ρ(r, t) together with the initial state ψ(t = −∞) up to a triv- ial additive time-dependent function c(t). If initially the system is in a stationary ground state, the initial state ψ(t = −∞) itself is a functional of initial electron den- sity ρ(r, t = −∞) according to the first Hohenberg-Kohn (HK) theorem21, and then the initial-state dependence of the external potential in RG theorem can be elimi- nated. As a result, at t = 0, the external potential and in turn the Hamiltonian can be uniquely determined by ρ(r, t = 0), justifying the commonly used ab initio trans- port theory that combines DFT and NEGF. Unfortunately, when the external bias Vb is not zero, initially (t = −∞), the system is in a non-equilibrium state instead of a stationary ground state. The non- equilibrium initial state consists of two separated parts, LB and RB, each of which is in its own equilibrium. From the first HK theorem, we know that the exter- nal potential of each part at t = −∞ can be deter- mined by its electron density up to an arbitrary con- stant. Combining two parts together, the external po- tential of the whole system at t = −∞ is determined by ρ(r, t = −∞) in whole space up to two independent ar- bitrary constants cl (from LB) and cr (from RB), which can be expressed as vext(r, t = −∞) ≡ evext[ρ(r, t = −∞)] + clr∈LB + crr∈RB. Two constants, cl and cr, shift the Hamiltonian and chemical potentials in LB and RB, respectively. Only knowing the electron density, cl and cr are not determined, then the Hamiltonian of the whole system and also the bias voltage Vb cannot be determined, resulting in a fact that the initial Hamil- tonian and in turn the initial state in general can not be uniquely determined by the electron density alone. With given evext[ρ(r, t = −∞)], cl and cr simply shift µl and µr respectively. Without losing generality, the exter- nal potential can also be written as vext(r, t = −∞) ≡ 2 the definition of the bias voltage Vb, we have vext(r, t = eevext[ρ(r, t = −∞)] + µlr∈LB + µrr∈RB. According to −∞) ≡ eevext[ρ(r, t = −∞)] + eVbr∈LB + µrr∈(LB+RB). We see immediately that the initial external potential and in turn the initial state is determined by the ini- tial electron density together with the bias voltage Vb up to a trivial additive constant for the whole system µr. Consequently, the initial-state dependence at t = 0 in RG theorem can be replaced by the voltage depen- dence, leading to an important theorem that forms the basis of the ab initio transport theory: When the system reaches the steady state (t = 0), its external potential, Vext(r, t = 0), is uniquely determined by the steady-state electron density together with the bias voltage Vb up to a trivial additive constant. As a direct consequence, the energy of the steady state can be written as a voltage- dependent density functional E[ρ(r), Vb]. When Vb = 0, the functional goes to the commonly used DFT one. Next, we show that the external parameter Vb can be determined by intrinsic properties of the system. For this purpose, as shown in Eq. 1, we divide the total electron density of the steady state ρt into two parts and name these two parts the equilibrium density ρe and non- equilibrium density ρn, respectively. ρt(r) = −iZ µl ρe(r) = −iZ µr ρn(r) = −iZ µl µr −∞ −∞ G<(r, ǫ)dǫ = ρe + ρn G<(r, ǫ)dǫ G<(r, ǫ)dǫ (1) Here we assume µr < µl. All physical quantities in Eq. 1 are defined for the steady state at t = 0. The term G< is the non-equilibrium lesser Green's function that includes effects of both non-equilibrium distribution and many-body interactions.2,3,10,11 The non-equilibrium density ρn as defined has a physical meaning of the den- sity of current-carrying electrons. Note that according to the definition, ρn can also be computed in reservoirs although reservoirs are assumed to be in equilibrium. With the assumption of the mean-field reservoirs and also the lesser Green's function in non-interacting sys- tems2,3,22, it is straitforward to show that given ρe and ρn, the total electron density ρt and the voltage bias Vb are determined. Considering the fact that ρe and ρn can also be determined by ρt and Vb, we therefore proved the one-to-one correspondence between these two sets of variables. Now we have one of the major results of the paper, the foundation of the proposed ab initio approach: For molecular junctions under a finite bias, the steady- state properties of the system are uniquely determined by the equilibrium and non-equilibrium electron densi- ties, ρe and ρn, as defined in Eq. 1. The aforementioned voltage dependent energy functional can then be written as E[ρe, ρn]. We now try to generalize the existing DFT-based ab initio transport theory2,3 to include bias-induced nonequilibrium effects. For this purpose, we assume a stationary principle for the non-equilibrium steady state: The variation of the the steady-state energy functional is zero, δEδρe ,δρn = 0. By taking the variations of the energy functional with respect to ρe and ρn in the scatter- ing region, we obtained two effective mean field equations (Eq. 2), (cid:18)− (cid:18)− 1 2 1 2 ∂exc ∂ρe (cid:19) φe ∂ρn(cid:19) φn ∂exc ∇2 + vext + VH + j = λe j φe j , ∇2 + vext + VH + j = λn j φn j , (2) where φe and φn are single-electron orbitals that con- tribute to ρe and ρn respectively. In the derivation of above equations, the generalized local density approxi- mation, Exc = R exc[ρe, ρn](r)dr, is used where exc is the exchange-correlation energy density of the uniform electron gas. The Hartree potential VH in the scattering region can be obtained by solving the Poisson equation with correct boundary conditions.10,11 Two coefficients, λe and λn, are energies of corresponding orbitals. These two equations have to be solved self-consistently together with the correct open-system boundary conditions, and NEGF techniques are powerful in matching the required boundary conditions. Defining two mean-field exchange- correlation potentials as V e , Eq. 2 suggests that the non-equilibrium electrons or current- carrying electrons experience a different mean-field ef- fective potential from the equilibrium electrons do. We therefore name the proposed method the dual-mean-field (DMF) approach. The DMF equations (Eq. 2) are key results of the paper, which provide the theoretical basis for the investigation of electronic properties of molecular junctions under a finite bias. xc = ∂exc ∂ρn xc = ∂exc ∂ρe and V n In general, the exchange-correlation energy density exc can be written as the summation of the exchange part and the correlation part, exc = ex + ec. The DMF ex- change energy density can be worked out by general- izing the Thomas-Fermi-Dirac (TFD) model23 to non- equilibrium cases by applying an external bias voltage to the uniform electron gas. By placing the uniform elec- tron gas between two non-interacting reservoirs with dif- ferent chemical potentials, the exchange energy density (Eq. 3) can be analytically derived (The derivation can be found in supporting information), and then two DMF x = ∂ex exchange potentials defined as V e ∂ρn can be computed, x = ∂ex ∂ρe and V n ex(ρt, η) = 1 4 (1 + η) 4/3(cid:20) −(cid:0)1 − η2(cid:1)2 +η4ln (η) + η4 + η3 − η2 + η + (ρt), (3) ln (1 + η) 3 2(cid:21)eT F D x 1 2 where η(r) = ρn(r)/ρt(r) which is called the non- 3 TranSIESTA DMF 18 16 14 12 10 ) A μ ( t n e r r u C 8 6 4 2 0 0 0.1 0.2 0.3 Voltage (Volts) 0.4 0.5 FIG. 2: I-V curves for a GNR junction calculated from both DMF approach and the commonly used ab initio transport theory via the software TranSIESTA. Inset: The atomic structure of the GNR junction and the iso-surface of the difference between the exchange energy potentials of current-carrying electrons calculated from DMF approach and TranSIESTA,δV = V n . The exchange poten- tials were calculated under 0.2 V. The iso-surface value is 15 meV . x − V T ranSIEST A x 1+η(cid:17)1/3 equilibrium index in this paper, and η = (cid:16) 1−η . Ac- cording to the definition, η takes the value between 0 and 1, which measures the extent of the nonequilibrium. When η = 0, the exchange density reduces to the equilib- rium TFD one eT F D . The DMF correlation energy density is however challenging. In this paper, we set the correlation functional to be the same as the DFT one, which may not be a bad approximation for weakly correlated systems under low biases. 4π3 (cid:0)3π2ρt(cid:1)4/3 (ρt) = − 1 x We now are ready to apply the DMF approach to real molecular scale junctions. As a case study, we choose a junction made of two zigzag graphene nanoribbons (Z-GNRs) with different widths as shown in the inset of Fig. 2. The GNRs have been regarded as one of the most promising building blocks for graphene-based electronic devices.24 Since the long-range magnetic order may not be stable under a finite temperature for such a one-dimensional system, we follow previous studies to set the total spin of the system zero.25,26 For Z-GNR based junctions under low biases, It has been well known that the current flows through edge states. We have imple- mented the DMF approach into the SIESTA computa- tional package.27 For comparison, we performed calcu- lations with both the DMF approach and the commonly used DFT based transport method via the function Tran- SIESTA built in SIESTA11 (More computational details can be found in supporting information). In Fig. 2, I-V curves from DMF and TranSIESTA calculations for the GNR junction are presented. When bias is small (< 0.1 V), the DMF approach essentially 0.1 η=0 FIG. 3: Non-equilibrium index η(r) in the scattering region of the GNR junction. The color map was plotted in the plane 2.2 A above the GNR. The system goes from local equilibrium to non-equilibrium when the color changes from red to blue. reproduces the TranSIESTA results. Starting from 0.1 V, significant deviations between two approaches occur, and the current from DMF calculation is always lower than that calculated from TranSIESTA. To understand this, we plot in the inset of Fig. 2 the iso-surface of the difference between the DMF non-equilibrium exchange potential (V n x ) and the DFT exchange potential calcu- lated from TranSIESTA. The potentials were calculated for the bias voltage 0.2 V. We see that the exchange po- tential increases significantly at edges which are places the electric current flows through. For other parts of the system, the non-equilibrium correction to the potential is not that important. The increase of the exchange poten- tial leads to a higher scattering barrier in the scattering region, and in turn, decreases the current as we see in Fig. 2. The non-equilibrium index in the DMF approach, η(r), provides detailed spacial information for the non- equilibrium steady state of the system. To demonstrate this, in Fig. 3, we plot the color contour map of η(r) in 4 the scattering region of the GNR junction under 0.2 V. From the figure, we can see that in the scattering region, the extent of the non-equilibrium at different places are quite different: Edges are far away from equilibrium while electrons in the center of ribbons are still approximately in local equilibrium. In conclusion, we propose a DMF approach to de- scribe electronic and transport properties of molecular- scale junctions under a finite bias. We show that two elec- tron densities, ρe and ρn, are needed to uniquely deter- mine the properties of the steady state of non-equilibrium molecular junctions. Subsequently, two mean fields, one for current-carrying electrons and the other one for equi- librium electrons, can be derived. The transport prop- erties can then be calculated from the mean-field poten- tial that the current-carrying electrons experience. The case study for a GNR junction shows that the DMF ap- proach could lead to significantly lower electric current than the commonly used transport theory. For molec- ular junctions that have localized molecular orbitals in the scattering region, the non-equilibrium corrections to the mean-field potential in the DMF approach will cause significant variations of these localized orbitals and lead to more profound changes in transport properties, which will be discussed in our future studies. We acknowledge the support from Ministry of Educa- tion (Singapore) and NUS academic research grants (R- 144-000-325-112 and R144-000-298-112). Computations were performed at the Graphene Research Centre and Centre for Computational Science and Engineering at NUS. See Supplementary Material Document No.xxxxx for the generalized TFD model and computational de- tails. For information on Supplementary Material, see http://www.aip.org/pubservs/epaps.html. The compu- tational codes for all calculations are available online or from the author. 1 A. Aviram, and M. A. Ratner, Chem. Phys. Lett. 29, 277 11 M. Brandbyge, J. Mozos, P. Ordejon, J. Taylor, and K. (1974) Stokbro, Phys. Rev. B 65, 165401 (2002) 2 J. Taylor, H. Guo, and J. Wang, Phys. Rev. B 63, R121104 12 S. Y. Quek, L. Venkataraman, H. J. Choi, S. Louie, M. S. (2001) Hybertsen, and J. B. Neaton, Nano Lett. 7, 3477 (2007) 3 Y. Xue, S. Datta, and M. A. Ratner, Chem. Phys. 281, 13 S. H. Ke, H. U. Baranger, and W. Yang, J. Chem. Phys. 151 (2002) 126, 201102 (2007) 4 S. M. Lindsay, and M. A. Ratner, Adv. Mater. 19, 23 14 C. Toher, A. Filippetti, S. Sanvito, and K. Burke, Phys. (2007) Rev. Lett. 95, 146402 (2005) 5 J. Tomfohr, and O. F. Sankey, J. Chem. Phys. 120, 1542 15 P. Delaney, and J. C. Greer, Phys. Rev. Lett. 93, 036805 (2004) (2004) 6 C. Zhang, M. H. Du, H. Cheng, X. Zhang, A. Roitberg, 16 M. Koentopp, K. Burke, and F. Evers, Phys. Rev. B 73, and J. L. Krause, Phys. Rev. Lett. 92, 158301 (2004) 121403 (2006) 7 K. Stokbro, J. Taylor, and M. Brandbyge, J. Am. Chem. 17 K. S. Thygesen, and A. Rubio, Phys. Rev. B 77, 115333 Soc. 125, 3674 (2003) 8 S. Barraza-Lopez, M. Kindermann, and M. Y. Chou, Nano Lett. 12, 3424 (2012) (2008) 18 K. S. Thygesen, Phys. Rev. Lett. 100, 166804 (2008) 19 J. B. Neaton, M. S. Hybertsen, and S. G. Louie, Phys. 9 L. Shen, M. G. Zeng, S. W. Yang, C. Zhang, X. F. Wang, Rev. Lett. 97, 216405 (2006) and Y. P. Feng, J. Am. Chem. Soc. 132, 11481 (2010) 10 J. Taylor, H. Guo, and J. Wang, Phys. Rev. B 63, 245407 (2001) 20 E. Runge, and E. K. U. Gross, Phys. Rev. Lett. 52, 997 (1984) 21 P. Hohenberg, and W. Kohn, Phys. Rev. B 136, B864 (1964) 22 C. Zhang, X. Zhang, P. S. Krsti´c, H. P. Cheng, W. H. But- ler, and J. M. Maclaren, Phys. Rev. B 69, 134406 (2004) 23 N. H. March, Adv. in Phys. 6, 1 (1957) 24 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009) 25 Q. Yan, B. Huang, J. Yu, F. Zheng, J. Zang, J. Wu, B. 5 Gu, F. Liu, and W. Duan, Nano Lett. 7, 1469 (2007) 26 Z. Li, H. Qian, J. Wu, B. Gu, and W. Duan, Phys. Rev. Lett. 100, 206802 (2008) 27 J. Solar, E. Artacho, J. Gale, A. Garc´ıa, J. Junquera, P. Ordej´on, and D. S´anchez-Portal, J. Phys.: Condens. Mat- ter 14, 2745 (2002) 

1907.00601

2

1907

Electrically driven spin torque and dynamical Dzyaloshinskii-Moriya interaction in magnetic bilayer systems

Efficient control of magnetism with electric means is a central issue of current spintronics research, which opens an opportunity to design integrated spintronic devices. However, recent well-studied methods are mostly based on electric-current injection, and they are inevitably accompanied by considerable energy losses through Joule heating. Here we theoretically propose a way to exert spin torques into magnetic bilayer systems by application of electric voltages through taking advantage of the Rashba spin-orbit interaction. The torques resemble the well-known electric-current-induced torques, providing similar controllability of magnetism but without Joule-heating energy losses. The torques also turn out to work as an interfacial Dzyaloshinskii-Moriya interaction which enables us to activate and create noncollinear magnetism like skyrmions by electric-voltage application. Our proposal offers an efficient technique to manipulate magnetizations in spintronics devices without Joule-heating energy losses.

cond-mat.mes-hall

cond-mat

Electrically driven spin torque and dynamical Dzyaloshinskii-Moriya interaction in magnetic bilayer systems Akihito Takeuchi,1 Shigeyasu Mizushima,1 and Masahito Mochizuki2, 3 1Department of Physics and Mathematics, Aoyama Gakuin University, Sagamihara, Kanagawa 229-8558, Japan 2Department of Applied Physics, Waseda University, Okubo, Shinjuku-ku, Tokyo 169-8555, Japan 3PRESTO, Japan Science and Technology Agency, Kawaguchi, Saitama 332-0012, Japan Abstract Efficient control of magnetism with electric means is a central issue of current spintronics re- search, which opens an opportunity to design integrated spintronic devices. However, recent well- studied methods are mostly based on electric-current injection, and they are inevitably accompa- nied by considerable energy losses through Joule heating. Here we theoretically propose a way to exert spin torques into magnetic bilayer systems by application of electric voltages through taking advantage of the Rashba spin-orbit interaction. The torques resemble the well-known electric- current-induced torques, providing similar controllability of magnetism but without Joule-heating energy losses. The torques also turn out to work as an interfacial Dzyaloshinskii-Moriya interaction which enables us to activate and create noncollinear magnetism like skyrmions by electric-voltage application. Our proposal offers an efficient technique to manipulate magnetizations in spintronics devices without Joule-heating energy losses. PACS numbers: 9 1 0 2 l u J 2 ] l l a h - s e m . t a m - d n o c [ 2 v 1 0 6 0 0 . 7 0 9 1 : v i X r a 1 Introduction Recent research in spintronics [1 -- 8] is seeking efficient ways to control magnetism in materials and realize high-performance magnetic devices beyond conventional techniques based on classical electromagnetism. To manipulate the magnetization in magnets, we need to exert torques to drive them. Spin-transfer torque is one such torque and, mediated by the exchange interaction, originates with the angular momentum transfer from the spins of the conducting electrons to magnetizations in the magnetic structure [9 -- 11]. Another is the so- called β-term torque originating from the spin relaxation due to the spin-orbit interactions and/or magnetic impurity scatterings, which is perpendicular to the spin transfer torque and thus behaves as a dissipative torque [11 -- 13]. These two torques enable the magnetization to be driven and switched via the injection of electric currents [14 -- 17]. The Rashba-type spin-orbit interaction (RSOI) has recently attracted interest as a means to manipulate conduction-electron spins [5, 18]. This interaction is of relativistic origin and becomes active in systems with broken spatial inversion symmetry such as surfaces and interfaces [5, 18 -- 21]. Because the RSOI mediates mutual coupling between spins and orbital momenta of the conduction electrons, it works as an effective magnetic field acting on their spins. The strength and direction of the effective magnetic field that each electron feels are governed by the momentum of the electron. Therefore, the RSOI can be a source of nontrivial torques acting on the magnetization through its control on the spin polarizations of the conduction electrons [22 -- 27]. The strength of the RSOI can be tuned by applying a gate voltage normal to the surface or interfacial plane [19], which modulates the extent of the spatial inversion asymmetry. Thus, an alternate-current (AC) gate voltage produces a time-varying RSOI, and thereby offers a potential technique to produce electrically an oscillating magnetic texture via the nontrivial Rashba-mediated torques. This technique must be useful for manipulation of magnetic skyrmions in a magnetic bilayer system, which have recently attracted great interest from the viewpoint of possible application to high- performance memory devices [28, 29]. Recent theoretical studies demonstrated that this time-varying RSOI induces an AC spin current [30 -- 32]. Ho et al. derived an expression for the torque T from the AC spin cur- rent [33], finding that this expression does not contain a term corresponding to the β-term torque. It is known that the spin-transfer torque alone cannot drive magnetic domain walls, 2 but other types of torques such as the β-term torque and the spin-orbit torque [34 -- 36] are required to drive them. However, their theory was based on a continuity equation for the conduction electron spins where the contribution of spin relaxation torque is totally ne- glected. The lack of a β-term torque may be a consequence of this crude approximation [37]. Therefore, we reexamine whether the β-term torque in the time-dependent Rashba electron systems is present using a more elaborate theoretical method. In this paper, we derive using the quantum field theory an explicit analytical formula for the torque arising from the AC RSOI in the magnetic bilayer systems. We demonstrate that the appropriate incorporation of the spin relaxation effect leads to a formula including both the spin-transfer torque and the β-term torque given in terms of the spin current proportional to the time derivative of the RSOI. We find that the derived expression may be regarded as a time-dependent Dzyaloshinskii-Moriya interaction (DMI) at the interface and may be exploited to manipulate noncollinear magnetic textures such as domain walls, magnetic vortices, and skyrmions. For a demonstration, we performed a numerical simulation of the electrical activation of a two-dimensional skyrmion lattice, where the effective AC DMI induces a periodic expansion and contraction of the skyrmions. This collective excitation of the skyrmions resembles the breathing mode [38, 39] observed in a skyrmion crystal under out-of-plane microwave magnetic fields. Efficient techniques to control and drive magnetism using electric fields are subject of intensive studies in the recent spintronics research because electric fields in insulators are not accompanied by Joule-heating losses. Note that electric currents and resulting Joule-heating losses can be significantly suppressed by using a ferromagnet/insulator bilayers or a metallic bilayer fabricated on an insulating substrate. Our finding provides a new efficient technique to control magnetism using AC electric fields and means to electrically excite eigenmodes of noncollinear spin textures. Results Model We consider a magnetic bilayer system, that is, a two-dimensional electron system on top of the surface of a magnet (see Fig. 1). This system is fabricated on an insulating substrate to enhance the effects of the electric gate voltage acting on the ferromagnet/heavy-metal 3 interface through preventing the electric-current flow. The total Hamiltonian of this electron system has four contributing terms, H = HK + HR + Hex + Himp with HK = 1 2me Z d2r(cid:12)(cid:12)pψ(r, t)(cid:12)(cid:12) αR(t) 2 − εFZ d2r ψ†(r, t)ψ(r, t), HR = −  Z d2r ψ†(r, t)(p × σ)zψ(r, t), Hex = JexZ d2r m(r) · ψ†(r, t) σψ(r, t), Himp = Z d2r vimp(r)ψ†(r, t)ψ(r, t), (1) (2) (3) (4) where me and p denote the mass and momentum, respectively, of a conduction electron, εF the Fermi energy, σ the Pauli matrices, and ψ† (ψ) the creation (annihilation) operator of a conduction electron. The term HK represents the kinetic energies of the conduction electrons with εF being the Fermi energy, while the term HR describes the time-varying RSOI with αR(t) being the time-dependent coupling coefficient. The term Hex represents the exchange interaction between the conduction-electron spins and the local magnetization with Jex and m being the coupling constant and the normalized local magnetization vector, respectively. The term Himp represents the scattering potentials from spatially distributed nonmagnetic impurities and determines the relaxation time of electrons given by τ . More conretely, we consider the impurity potential given by vimp(r) = uimpPi δ(r − Ri), where uimp denotes the strength of impurity scattering, Ri positions of the impurities, and δ(r) the Dirac delta function. When we take an average over the impurity positions as vimp(r) = 0 and vimp(r)vimp(r′) = nimpu2 impδ(r − r′), the relaxation time of electrons is given by τ = /2πνenimpu2 imp in the first Born approximation. Here, nimp denotes the concentration of impurities and νe = me/2π2 the density of state. In the present study, we focus on the spin torques induced by the time-varying RSOI but neglect those induced by the magnetization dynamics. Note that there should be a feedback effect that the spin states of the conduction electrons are modulated by the magnetization dynamics, but we neglect this subsequent effect on the spin torques. Spin torque arising from time-dependent RSOI The torque induced by the electron spins via the exchange interaction is defined as m × s, T = Jexa2  4 (5) (cid:85)(cid:74)(cid:78)(cid:70)(cid:14)(cid:69)(cid:70)(cid:81)(cid:15)(cid:1)(cid:51)(cid:66)(cid:84)(cid:73)(cid:67)(cid:66)(cid:1)(cid:87)(cid:70)(cid:68)(cid:85)(cid:80)(cid:83) DR(t) (cid:79)(cid:80)(cid:79)(cid:68)(cid:80)(cid:77)(cid:77)(cid:74)(cid:79)(cid:70)(cid:66)(cid:83)(cid:1) (cid:78)(cid:66)(cid:72)(cid:79)(cid:70)(cid:85)(cid:74)(cid:91)(cid:66)(cid:85)(cid:74)(cid:80)(cid:79)(cid:84) m(r) (cid:71) (cid:70) (cid:83) (cid:83) (cid:80)(cid:78) (cid:66) (cid:72) (cid:79) (cid:70) (cid:85) (cid:77) (cid:66) (cid:90) (cid:70) (cid:83) (cid:1) (cid:73) (cid:70) (cid:66) (cid:87) (cid:90) (cid:14)(cid:78) (cid:70) (cid:85) (cid:66) (cid:77) (cid:74) (cid:79) (cid:84) (cid:86) (cid:77) (cid:66) (cid:85) (cid:74) (cid:79) (cid:72) (cid:1) (cid:84) (cid:86) (cid:67) (cid:84) (cid:85) (cid:83) (cid:66) (cid:85) (cid:70) AC (cid:34)(cid:36)(cid:1)(cid:72)(cid:66)(cid:85)(cid:70)(cid:1) (cid:87)(cid:80)(cid:77)(cid:85)(cid:66)(cid:72)(cid:70) FIG. 1: Schematic illustration of the time-dependent Rashba electron system interacting with local magnetizations m(r). The Rashba parameter αR(t) is time-modulated by an external AC electric voltage. The insulating substrate prevents electric-current flows and enhances the effects of the electric voltage acting on the RSOI-hosting interface. where a is the lattice constant, s = hψ† σψi is the conduction electron spin density, and the brackets denote the quantum expectation value. We assume a metallic bilayer system in which the condition Jex < εF usually holds and an adiabatic case with a slowly varying magnetization texture of q ≪ kF where q is the wavenumber of local magnetization and kF is the Fermi wavenumber. We also assume that a weak magnitude of αR (αRkF ≪ εF) and low frequencies Ω (Ω ≪ εF) for the time-dependent RSOI. In this perturbation regime, we obtain the analytical formula of the torque in the form T = T 1+T 2+T 3 = − a  D1(m×∇)zm+ a  D2(m×∇)zm− a  βRD2m×h(m×∇)zmi, (6) where each coefficient is defined in the condition with εF ≫ /τ , Jex ≫ /τ and εF − Jex ≫ /τ as (η = /2τ ) D1(t) = νea 2πτ (cid:20) εF εF − Jex(cid:19) − 2(cid:21)αR(t), ln(cid:18) εF + Jex Jex ex − η2) J 2 ex(J 2 ex + η2)2 (J 2 dαR(t) dt , D2(t) = νeaεFτ βR = 2Jexη ex − η2 . J 2 (7) (8) (9) We note that D1 vanishes in the clean limit with τ → ∞. Note that D1 does not vanish in the three-dimensional case or in the half-metallic case with Jex > εF [40] even in the clean limit. 5 In Eq. (6), the first two contributions, T 1 + T 2, describe an effective DMI [41, 42], which is given in the continuum form as HDMI = D1 − D2 a ǫαβz Z d2r (m × ∇αm)β. (10) (Note that m × (a2/)δHDMI/δm leads to T 1 + T 2). The contribution D1 (∝ αR) appears even in the steady Rashba system [40, 43, 44]. Recent experiments indeed demonstrated voltage-induced variations of the interfacial DMI [45, 46], which were ascribed to this steady Rashba contribution. Quite recently, an experimental observation of gigantic variation of the DMI that reaches 130 % has been reported for the Ta/FeCoB/TaOx multilayer system [46]. In contrast, the contribution D2 (∝ ∂tαR) appears only in a driven Rashba system with a time-dependent RSOI. This interfacial DMI may be tuned by an electric gate voltage via the RSOI, and, more interestingly, an oscillating DMI may be achieved by applying an AC gate voltage. The Rashba coefficient αR(t) in the driven Rashba system is a sum of steady and time-dependent components, αR(t) = α0 + αext(t) with αext(t) = αext sin (Ωt). For metallic (semiconducting) bilayer systems, the strength of this Rashba-induced DMI is roughly estimated to be D1 ∼ 0.1 meV (6 × 10−6 meV) and D2 ∼ 5 × 10−3 meV (2 × 10−6 meV). Here we assume typical parameter values [19 -- 21], i.e., a = 5 A, εF = 4 eV (10 meV), kF = 1 A−1 (0.01 A−1), Jex/εF = 0.25 (0.5), τ = 10−14 s (10−12 s), α0 = 2 eV·A (0.07 eV·A), αext/α0 = 0.1, and Ω/2π = 1 GHz. The strength of the Rashba-mediated DMI is relatively strong (weak) in metal (semiconductor) systems. We also note that the magnitude of D2 being proportional to ∂tαext(t) may be tuned by changing the amplitude and frequency of the AC gate voltage. The relative strength of D2 or the ratio D2/D1 tends to be small. Note that the ratio D2/D1 is approximately given by ΩεFτ 2/2π, which takes ∼ 10−4 (10−2) for metallic (semiconducting) bilayer systems when a typical frequency of Ω=1 GHz is assumed. Namely, the ratio D2/D1 tends to be larger for the semiconducting system, whereas the absolute value of D2 tends to be larger for the metallic system. We should choose an appropriate system depending on the target phenomena or the experiments. In Eq. (6), the last two terms proportional to D2 may be rewritten as T 2 + T 3 ∝ (js · ∇)m − βRm × (js · ∇)m. (11) by adopting a definition of the spin current js ≡ (e/a)D2 z × m. We find that these terms 6 have equivalent forms with the spin-transfer torque and the nonadiabatic torque associ- ated with the spin current js, respectively. With an AC-dependent αR(t), the spin current js ∝ ∂tαR(t) z × m gives rise to AC torques. In the clean limit with /Jexτ ≪ 1, the coefficient βR is reduced to /Jexτ . Finally, the second term is exactly identical in form to the conventional current-induced nonadiabatic torque although τ corresponds to a differ- ent time scales, specifically, the relaxation time of the conduction electrons (spins) in the present (current-induced) case [11]. Assuming the above-mentioned material parameters for the metallic bilayer systems, we evaluate the values of js = (e/a)D2 and βR as ∼ 2 A/m and ∼ 0.07, respectively. These values are large enough to induce the magnetization dynamics. Application of oscillating DMI The AC torques in the driven Rashba system may be exploited to activate resonances of the magnetic textures. For a demonstration, we numerically show that the breathing mode of a skyrmion crystal can be excited by the RSOI-mediated AC torques. We start with the continuum limit of the spin model for a two-dimensional magnet, H = Z d2r J 2 (∇m)2 − µBµ0 a2 Z d2r m · H. (12) The model contains the ferromagnetic exchange interaction and the Zeeman interaction with an external magnetic field H = H z. We adopt typical material parameters of a = 5 A, J = 1 meV and µ0H = 34 mT. The dynamics of magnetizations obeys the Landau-Lifshitz- Gilbert equation m = −γmm ×(cid:18)− a2 γm δH δm(cid:19) + αGm × m + T 1, (13) where γm is the gyromagnetic ratio, and αG (= 0.04) is the Gilbert-damping coefficient. We incorporate T 1 as the Rashba-mediated torque whereas T 2 and T 3 are neglected because they are much smaller than T 1 in the metallic bilayer systems. that T 1 (T 2 and T 3) originates from D1 (D2), and the ratio D2/D1 is 10−12-10−14 as men- (Equation (6) indicates tioned above.) The coupling coefficient D1(t) is composed of a steady component D0 and the time-dependent component Dext(t) as D1(t) = D0 + Dext(t). We solve this equation numerically using the fourth-order Runge-Kutta method for a system of 140 × 162 nm2, including N = 280 × 324 magnetizations, applying a periodic boundary condition. Without the AC electric voltage, the Rashba-mediated torque has a steady component only with 7 D1 = D0(= 0.09 meV). In this situation, the torque stabilizes the skyrmion crystal with hexagonally packed N´eel-type skyrmions (Fig. 2(a)) at low temperatures by effectively work- ing as a static DMI, as exemplified by Eq. (10). Next, we simulated the dynamics of this skyrmion crystal in the presence of AC gate volt- ages by switching on the Rashba-mediated AC torque with Dext(t) 6= 0. The time evolution of the numerical simulation is performed for every 0.01 ps. We first calculate the dynami- cal magnetoelectric susceptibility χ(Ω) to identify the resonance frequency of this skyrmion crystal. We trace a time profile of the net magnetization M (t) = (1/Na2)R d2r m(r, t) and ∆M (t) = M (t) − M (0) after applying a short electric-field pulse by switching on the time-dependent component of the torque Dext(t) = 0.05D0 for an interval of 0 ≤ t ≤ 1. We obtain its spectrum (Fig. 2(b)) by calculating the Fourier transform of ∆M (t). From this spectrum, we find that the skyrmion crystal has a resonance at Ω/2π = 1.72 GHz. Note that the resonance frequency (the time period) of the skyrmion eigenmode is determined by the spin-wave gap which is proportional to D2 0 (D−2 0 ). We then apply a microwave electric field to this skyrmion-hosting magnetic bilayer sys- tem by switching on the AC component of the torque Dext(t) = 0.05D0 sin (Ωt). We ex- amined both resonance with Ω/2π = 1.72 GHz and off-resonance with Ω/2π = 1 GHz. For the former, we observe a breathing-type mode where all the skyrmions constituting the skyrmion crystal uniformly contracted and expanded periodically (Fig. 2(c)). For the latter, the induced breathing oscillation of the skyrmions is small. We find that the periodically modulated skyrmion diameter is nearly proportional to the time-dependent DMI coefficient. The present simulation demonstrated the electrical activation of a skyrmion resonance. The induced dynamical DMI is also expected to realize the electrical creation of skyrmions [48 -- 53]. The electrical writing of skyrmions with the RSOI-mediated DMI in the field polarized ferromagnetic state and even the helical state should be established in the future study. Discussion In summary, we have theoretically derived a precise formula of the spin torque in a time-varying Rashba electron system driven by the AC gate voltage. The obtained formula contains not only the spin-transfer torque but also the β-term torque associated with the AC spin current that is proportional to the time derivative of the RSOI. These AC torques can 8 FIG. 2: Eigenmodes of N´eel-type skyrmion crystal activated by an AC DMI. (a) Skyrmion crys- tal with hexagonally packed N´eel-type skyrmions. In-plane and out-of-plane components of the magnetizations are shown by arrows and color map, respectively. (b) Imaginary part of the dy- namical magnetoelectric susceptibility. (c) Snapshots of the electrically activated breathing motion at t = 0, π/2Ω, π/Ω and 3π/2Ω for the resonant condition with Ω/2π = 1.72 GHz, and (d) those for an off-resonant condition with Ω/2π = 1 GHz. excite resonance of magnetic textures through acting as an interfacial AC DMI. Indeed, we have numerically demonstrated that the effective AC DMI can activate the breathing mode of magnetic skyrmions. We have confirmed that not only the crystallized magnetic skyrmions (skyrmion crystal) but also isolated skyrmions in ferromagnets can be excited resonantly by application of a microwave field, which offers a better experimental feasibility because a lot of ferromagnet/heavy-metal bilayer systems turned out to host magnetic skyrmions as topological defects [28, 54 -- 56]. Recent theoretical studies revealed that activation of the skyrmion breathing mode under application of a magnetic field inclined from the vertical direction induces translational motion of the skyrmions [57 -- 59], which provides a means to 9 drive magnetic skyrmions electrically with a low energy consumption. Our finding provides a promising technique to manipulate noncollinear magnetic textures with a great efficiency that have potential applications in memory, logic, and microwave devices. There are several types of devices to realize the proposed effects. The devices must have two important features, i.e., (1) an interface that hosts the SOI due to the broken inversion symmetry and (2) insulating nature to prevent the electric-current flow. One possible type of device is a ferromagnet/insulator bilayer system [45, 46]. On the contrary, we proposed another type of device with ferromagnet/heavy-metal bilayer fabricated on an insulating substrate. In the latter system, we expect much stronger SOI because of the heavy-metal layer. In this case, the required insulating nature is taken up by the insulating substrate. The issue which system is appropriate is left for future study. The RSOI is originally very strong in the latter system, but its electric tunability may be low because the applied electric field mainly acts on the heavy-metal/insulator layer but not on the ferromagnet/heavy-metal interface. In addition, due to the short screening length in metal, the electric field decays quickly and may hardly reach the interface that hosts the RSOI. On the other hand, the ferromagnet/insulator bilayer system can originally have a weak RSOI only, but its electric tunability can be large because the applied electric field directly acts on the ferromagnet/insulator interface that hosts the RSOI. It should be also mentioned that several types of bilayer systems with interfacial DMI have been intensively studied recently, where driven spin torques and generations of skyrmion-type noncollinear magnetic textures using the interfacial DMI have been experimentally demon- strated not only for ferromagnet/heavy-metal systems but also for ferromagnet/transition- metal-dichalcogenide systems [60] and ferromagnet/topological-insulator systems [61]. In the latter two cases, the SOI is much more complicated than the simple Rashba model for the ferromagnet/heavy-metal system considered in the present study [62], and it is unclear which of many terms emerged at the interfaces of these systems can be modulated by AC gate voltage in reality [63]. These problems require further investigations, which are left for future researches. In experiments, we need to take care of magnetic anisotropies in the fer- romagnetic layer because the stability and the resonance modes of skyrmions are sensitively affected by them [64]. Our results will provide a firm basis and a good starting point for future experimental and theoretical studies. 10 Methods Diagonalization of exchange interaction We need to calculate the electron spin density vector s to obtain the spin torque. However, as the exchange coupling Jex is generally large compared with other energy scales such as the kinetic energy and the RSOI, we cannot regard it as a perturbation. Instead, we need to perform a diagonalization of the exchange interaction term [11, 33]. For this purpose, we employ a 2 × 2 unitary matrix U ≡ n · σ with n = (sin (θ/2) cos φ, sin (θ/2) sin φ, cos (θ/2)). Here, θ and φ are the polar and azimuthal angles of the local magnetization vector m = (sin θ cos φ, sin θ sin φ, cos θ). As relation m· U † σ U = σz holds, the exchange interaction term may be diagonalized using a new operator of the conduction electron Ψ expressed in the local coordinates rotating along with the spatially modulated magnetization vectors. Hereafter, this coordinate system is referred to as the rotated spin frame. The new operator Ψ is related to the original operator ψ defined in the global coordinates via a unitary transformation as ψ = U Ψ. Finally, the total Hamiltonian is rewritten with the Ψ operators as H = 1 2me Z d2r(cid:12)(cid:12)(cid:2)p + eAα(r, t)σα(cid:3)Ψ(r, t)(cid:12)(cid:12) +JexZ d2r Ψ†(r, t)σzΨ(r, t) 2 − εFZ d2r Ψ†(r, t)Ψ(r, t) meα2 R(t) 2 − Z d2r Ψ†(r, t)Ψ(r, t) +Z d2r vimp(r)Ψ†(r, t)Ψ(r, t), where −e (< 0) is the electron charge. The non-Abelian gauge potential A appears as a by-product of the diagonalization of the exchange interaction. This gauge potential contains two contributions [33], denoted A = A(ex) + A(so). The first term A(ex) is a gauge potential originating from the spatial variation of the magnetization structure whereas the second term A(so) comes from the RSOI. Each gauge potential is defined as A(ex)α µ σα ≡ − A(so)α µ σα ≡ − i e meαR e ǫµβz U † σβ U = − meαR e ǫµβzRαβ σα, U †∇µ U = (n × ∇µn)ασα,  e with Rαβ σα ≡ U † σβ U = (2nαnβ − δαβ)σα. Here Rαβ is an element of a 3 × 3 orthogonal matrix. The symbols δαβ and ǫαβγ denote the Kronecker delta and the Levi-Civita antisym- metric tensor, respectively. In the rotated spin frame, the Ψ-electron spin density is given by 11 Sα = hΨ†σαΨi. The spin density S in the rotated spin frame is related to the spin density s in the original frame via R as sα = RαβSβ. Therefore, the spin torque T is rewritten using S± = Sx ± iSy, T α = Jexa2 2 ǫαβγmβ Xσ=± (Rγx − iσRγy)Sσ. In the calculation, we consider the impurity potential given by vimp(r) = uimpXi δ(r − Ri), where uimp denotes the strength of impurity scattering, Ri the positions of the impurities, and δ(r) the Dirac delta function. When we take an average over the impurity positions as vimp(r) = 0, vimp(r)vimp(r′) = nimpu2 impδ(r − r′) the relaxation time of electrons is given by τ = /2πνenimpu2 imp in the first Born approxi- mation. Here, nimp denotes the concentration of impurities and νe = me/2π2 the density of state. Calculation of spin torque arising from AC RSOI The Ψ-electron spin density S± is written in terms of the path-ordered Green function S±(r, t) = −iTrhσ± G<(r, t; r, t)i, where σ± = σx ± iσy and Tr signifies the trace over the spin indices. The lesser component of the path-ordered Green function [47] is represented by G<(r, t; r′, t′) = (i/)hΨ†(r′, t′)Ψ(r, t)i. In the present system, the Dyson equation is given as G(r, t; r′, t′) = g(r − r′, t − t′) +ZC where C denotes the Keldysh contour and V is defined as dt′′ Z d2r′′ g(r − r′′, t − t′′) V (r′′, t′′) G(r′′, t′′; r′, t′), V (r′′, t′′) = − ie 2me(cid:20) ∂ ∂r′′ µ Aα µ(r′′, t′′) + Aα µ(r′′, t′′) + e2 2me Aα µ(r′′, t′′)Aα µ(r′′, t′′)I − 12 ∂ ∂r′′ µ(cid:21)σα R(t′′)I + vimp(r′′)I. me 2 α2 Here g denotes the noninteracting Green function given in the Fourier space as gk,ω = G< is the identity matrix. (1/2)Pσ=±(I + σσz)gk,ω,σ where I bol < represents the relation [RC dt′′ G1(t, t′′) G2(t′′, t′)]< = R ∞ 1 (t, t′′) Ga 2 (t′′, t′) + 2(t′′, t′)] [47]. The retarded, advanced, and lesser Green functions (gr, ga and k,ω,σ) where fω is the Fermi distribu- k,ω,σ)∗ = 1/(ω − εk + εFσ + iη) where εk = 2k2/2me and εFσ = εF − σJex. The spin density S± tion function. The retarded (advanced) Green function is defined as gr g<) are mutually related by g< k,ω,σ = fω(ga k,ω,σ − gr −∞ dt′′ [ Gr 1(t, t′′) G< The superscript sym- k,ω,σ = (ga can be obtained by iteration of this equation. The dominant contributions are given by the first-order perturbation expansions in A(so) and up to first order in A(ex). After some algebra, this equation is reduced to (see Supplementary Materials for details) S± = − 2e JexahD1 − (1 ∓ iβR)D2i(cid:16)m × A(ex)±(cid:17)z , where A(ex)± = A(ex)x ±iA(ex)y. Substituting this result into the definition of the spin torque and using the relations Pσ=±(Rαx − iσRαy)A(ex)σ µ iσRαy)iσA(ex)σ µ = −(/e)(m × ∇µm)α and Pσ=±(Rαx − = (/e)∇µmα, we thus have obtained the result given in Eqs. (6-9). Note on the Numerical Simulations Our numerical simulation with the LLG equation corresponds to the micromagnetic sim- ulation based on the continuum spin model with the exchange stiffness A = 4.0 × 10−14 [Jm−1], the continuum DM parameter B = 1.44 × 10−5 [Jm−2], the magnetic field µ0Hz=34 mT, and the saturation magnetization Ms = 3.5 × 104 [Am−1] for a system size of 140 nm × 162 nm × 2 nm. This simulation can be performed with commercial or free softwares such as OOMMF and mumax. In the present study, the continuum spin model is mapped to the lattice spin model by dividing the continuum space into identical rectangular cells. More concretely, dividing the continuum space into identical rectangular cells of 0.5 nm×0.5 nm×2 nm, we obtain the lattice spin model with the normalized magnetization vectors mi where the exchange coupling J=1 meV, the DM parameter D0/J=0.09, and the magnetic field µBµ0Hz/J=0.004. [1] Ohno, H. Making nonmagnetic semiconductors ferromagnetic. Science 281, 951 (1998). 13 [2] Wolf, S. A., Awschalom, D. D., Buhrman, R. A., Daughton, J. M., von Moln´ar, S., Roukes, M. L., Chtchelkanova, A. Y. & Treger, D. M. Spintronics: a spin-based electronics vision for the future. Science 294, 1488 (2001). [3] Zuti´c, I., Fabian, J. & Das Sarma, S. Spintronics: fundamentals and applications. Rev. Mod. Phys. 76, 323 (2004). [4] Chappert, C., Fert, A. & Van Dau, F. N. The emergence of spin electronics in data storage. Nat. Mater. 6, 813 (2007). [5] Manchon, A., Koo, H. C., Nitta, J., Frolov, S. M., & Duine, R. A. New perspectives for Rashba spin-orbit coupling. Nat. Mat. 14, 871 (2015). [6] Endoh, T., & Honjo, H. A recent progress of spintronics devices for integrated circuit appli- cations. J. Low Power Electron. Appl. 8, 44 (2018). [7] Jungwirth, T., Sinova, J., Manchon, A., Marti, X., Wunderlich, J., & Felser, C. The multiple directions of antiferromagnetic spintronics. Nat. Phys. 14, 200 (2018). [8] Baltz, V., Manchon, A., Tsoi, M., Moriyama, T., Ono, T., & Tserkovnyak, Y. Antiferromag- netic spintronics. Rev. Mod. Phys. 90, 015005 (2018). [9] Slonczewski, J. C. Current-driven excitation of magnetic multilayers. J. Magn. Magn. Mater. 159, L1 (1996). [10] Berger, L. Emission of spin waves by a magnetic multilayer traversed by a current. Phys. Rev. B 54, 9353 (1996). [11] Tatara, G., Kohno, H. & Shibata, J. Microscopic approach to current-driven domain wall dynamics. Phys. Rep. 468, 213 (2008). [12] Zhang, S. & Li, Z. Roles of nonequilibrium conduction electrons on the magnetization dynam- ics of ferromagnets. Phys. Rev. Lett. 93, 127204 (2004) [13] Thiaville, A., Nakatani, Y., Miltat, J. & Suzuki, Y. Micromagnetic understanding of current- driven domain wall motion in patterned nanowires. Eyrophys. Lett. 69, 990 (2005). [14] Klaui, M., Vaz, C. A. F., Bland, J. A. C., Wernsdofer, W., Faini, G., Cambril, E. & Hey- derman, L. J. Domain wall motion induced by spin polarized currents in ferromagnetic ring structures. Appl. Phys. Lett. 83, 105 (2003). [15] Tsoi, M., Fontana, R. E. & Parkin, S. S. P. Magnetic domain wall motion triggered by an electric current. Appl. Phys. Lett. 83, 2617 (2003). [16] Yamaguchi, A., Ono, T., Nasu, S., Miyake, K., Mibu, K. & Shinjo, T. Real-space observation of 14 current-driven domain wall motion in submicron magnetic wires. Phys. Rev. Lett. 92, 077205 (2004). [17] Yamanouchi, M., Chiba, D., Matsukura, F. & Ohno, H. Current-induced domain-wall switch- ing in a ferromagnetic semiconductor structure. Nature 428, 539 (2004). [18] Rashba, E. I. Properties of semiconductors with an extremum loop. 1. Cyclotron and combi- national resonance in a magnetic field perpendicular to the plane of the loop. Sov. Phys. Solid State 2, 1109 (1960). [19] Nitta, J., Akazaki, T., Takayanagi, H. & Enoki, T. Gate control of spin-orbit interaction in an inverted In0.53Ga0.47As/In0.52Al0.48As heterostructure. Phys. Rev. Lett. 78, 1335 (1997). [20] Ast, C. R., Henk, J., Ernst, A., Moreschini, L., Falub, M. C., Pacil'e, D., Bruno, P. Kern, K. & Grioni, M. Giant spin splitting through surface alloying. Phys. Rev. Lett. 98, 186807 (2007) [21] Nakagawa, T., Ohgami, O., Saito, Y., Okuyama, H., Nishijima, M. & Aruga, T. Transition be- tween tetramer and monomer phases driven by vacancy configuration entropy on Bi/Ag(001). Phys. Rev. B 75, 155409 (2007). [22] Obata, K. & Tatara, G. Current-induced domain wall motion in Rashba spin-orbit system. Phys. Rev. B 77, 214429 (2008). [23] Manchon, A. & Zhang, S. Theory of nonequilibrium intrinsic spin torque in a single nanomag- net. Phys. Rev. B 78, 212405 (2008). [24] Manchon, A. & Zhang, S. Theory of spin torque due to spin-orbit coupling. Phys. Rev. B 79, 094422 (2009). [25] Kim, K.-W., Seo, S.-M., Ryu, J., Lee, K.-J. & Lee, H.-W. Magnetization dynamics induced by in-plane currents in ultrathin magnetic nanostructures with Rashba spin-orbit coupling. Phys. Rev. B 85, 180404(R) (2012). [26] Miron, I. M., Gaudin, G., Auffret, S., Rodmacq, B., Schuhl, A., Pizzini, S., Vogel, J. & Gambardella, P. Current-driven spin torque induced by the Rashba effect in a ferromagnetic metal layer. Nat. Mater. 9, 230 (2010). [27] Miron, I. M., Moore, T., Szambolics, H., Buda-Prejbeanu, L. D., Auffret, S., Rodmacq, B., Pizzini, A., Vogal, J., Bonfim, M., Schuhl, A. & Gaudin, G. Fast current-induced domain-wall motion controlled by the Rashba effect. Nat. Mater. 10, 419 (2011). [28] Fert, A., Cros, V. & Sampaio, J. Skyrmions on the track. Nat. Nanotech. 8, 152 (2013). [29] Soumyanarayanan, A., Reyren, N., Fert, A. & Panagopoulos, C. Emergent phenomena induced 15 by spin-orbit coupling at surfaces and interfaces. Nature 539, 509 (2016). [30] Mal'shukov, A. G., Tang, C. S., Chu, C. S. & Chao, K. A. Spin-current generation and detection in the presence of an ac gate. Phys. Rev. B 68, 233307 (2003). [31] Zhang, S.-f. & Zhu, W. The limit spin current in a time-dependent Rashba spin-orbit coupling system. J. Phys.: Condens. Matter 25, 075302 (2013). [32] Ho, C. S., Jalil, M. B. A. & Tan, S. G. Spin force and the generation of sustained spin current in time-dependent Rashba and Dresselhaus systems. J. Appl. Phys. 115, 183705 (2014). [33] Ho, C. S., Jalil, M. B. A. & Tan, S. G. Gate-control of spin-motive force and spin-torque in Rashba SOC system. New J. Phys. 17, 123005 (2015). [34] Ando, K., Takahashi, S., Harii, K., Sasage, K., Ieda, J., Maekawa, S., & Saitoh, E. Electric manipulation of spin relaxation using the spin Hall effect. Phys. Rev. Lett. 101, 036601 (2008). [35] Emori, S., Bauer, U., Ahn, S.-M., Martinez, E., & Beach, G. S. D. Current-driven dynamics of chiral ferromagnetic domain walls. Nat. Mat. 12, 611 (2013). [36] Ryu, K.-S., Thomas, L., Yang, S.-H., & Parkin, S. Chiral spin torque at magnetic domain walls. Nat. Nanotech. 8, 527 (2013). [37] Tatara, G. & Entel, P. Calculation of current-induced torque from spin continuity equation. Phys. Rev. B 78, 064429 (2008). [38] Mochizuki, M. Spin-wave modes and their intense excitation effects in skyrmion crystals. Phys. Rev. Lett. 108, 017601 (2012). [39] Onose, Y., Okamura, Y., Seki, S., Ishiwata, S. & Tokura, Y. Observation of magnetic excita- tions of skyrmion crystal in a helimagnetic insulator Cu2OSeO3. Phys. Rev. Lett. 109, 037603 (2012). [40] Ado, I. A., Qaiumzadeh, A., Duine, R. A., Brataas, A. & Titov, M. Asymmetric and symmetric exchange in a generalized 2D Rashba ferromagnet. Phys. Rev. Lett. 121, 086802 (2018). [41] Dzyaloshinskii, I. E. Thermodynamics theory of "weak" ferromagnetism in antiferromagnetic substances. Sov. Phys. JETP 5, 1259 (1957). [42] Moriya, T. Anisotropic superexchange interaction and weak ferromegnetism. Phys. Rev. 120, 91 (1960). [43] Kim, K.-W., Lee, H.-W., Lee, K.-J. & Stiles, M. D. Chirality from interfacial spin-orbit coupling effects in magnetic bilayers. Phys. Rev. Lett. 111, 216601 (2013). [44] Kikuchi, T., Koretsune, T., Arita, R. & Tatara, G. Dzyaloshinskii-Moriya interaction as a 16 consequence of a Doppler shift due to spin-orbit induced intrinsic spin current. Phys. Rev. Lett. 116, 247201 (2016). [45] Nawaoka, K., Miwa, S., Shiota, Y., Mizuochi, N. & Suzuki, Y. Voltage induction of interfacial Dzyaloshinskii-Moriya interaction in Au/Fe/MgO artificial multilayer. Appl. Phys. Express 8, 063004 (2015). [46] Srivastava, T. et al. Large-voltage tuning of Dzyaloshinskii-Moriya interactinos: a route toward dynamic control of skyrmion chirality. Nano Lett. 18, 4871 (2018). [47] Haug, H. & Jauho, A. P. Quantum Kinetics in Transport and Optics of Semiconductors (Springer, 2007). [48] Mochizuki, M., & Watanabe, Y. Writing a skyrmion on multiferroic materials. Appl. Phys. Lett. 107, 082409 (2015). [49] Mochizuki, M., Creation of skyrmions by electric field on chiral lattice magnetic insulators. Adv. Electron. Mater. 2, 1500180 (2016). [50] Schott, M. et al. The skyrmion switch: Turning magnetic skyrmion bubbles on and off with an electric field. Nano Lett. 17, 3006 (2017). [51] Huang, P. et al. In situ electric field skyrmion creation in magnetoelectric Cu2OSeO3. Nano Lett. 18, 5167 (2018). [52] Kruchkov, A. J. et al. Direct E field control of the skyrmion phase in a magnetoelectric insulator. Sci. Rep. 8, 10466 (2018). [53] Wang, L. et al. Ferroelectrically tunable magnetic skyrmions in ultrathin oxide heterostruc- tures. Nat. Mat. 17, 1087 (2018). [54] Woo, S. et al. Observation of room-temperature magnetic skyrmions and their current-driven dynamics in ultrathin metallic ferromagnets. Nat. Mater. 15, 501 (2016). [55] Yu, G. et al. Room-Temperature Skyrmion Shift Device for Memory Application. Nano Lett. 17, 261 (2017). [56] Jiang, W. et al. Blowing magnetic skyrmion bubbles. Science 349, 283 (2015). [57] Wang, W., Beg, M., Zhang, B., Kuch, W. & Fangohr, H. Driving magnetic skyrmions with microwave fields. Phys. Rev. B 92, 020403(R) (2015). [58] Takeuchi, A. & Mochizuki, M. Selective activation of an isolated magnetic skyrmion in a ferromagnet with microwave electric fields. Appl. Phys. Lett. 113, 072404 (2018). [59] Ikka, M., Takeuchi, A. & Mochizuki, M. Resonance modes and microwave-driven translational 17 motion of a skyrmion crystal under an inclined magnetic field. Phys. Rev. B 98, 184428 (2018). [60] Liu, R. H., Lim, W. L. & Urazhdin, S. Control of current-induced spin-orbit effects in a ferromagnetic heterostructure by electric field. Phys. Rev. B 89, 220409(R) (2014). [61] Lv, W. et al. Electric-Field Control of Spin-Orbit Torques in WS2/Permalloy Bilayers. ACS Appl. Mater. Interfaces 10, 2843 (2018). [62] Belashchenko, K. D., Kovalev, A. A. & van Schilfgaarde, M. First-principles calculation of the spin-orbit torques in a Co/Pt bilayer. arXiv:1810.11003. [63] Gmitra, M., Kochan, D., Hogl, P. & Fabian, J. Trivial and inverted Dirac bands and the emergence of quantum spin Hall states in graphene on transition-metal dichalcogenides. Phys. Rev. B 93, 155104 (2016). [64] Nakamura, M. et al. Emergence of Topological Hall Effect in Half-Metallic Manganite Thin Films by Tuning Perpendicular Magnetic Anisotropy. J. Phys. Soc. Jpn. 87, 074704 (2018). Acknowledgements This work was supported by JSPS KAKENHI (Grant No. 17H02924 and No. 16H06345), Waseda University Grant for Special Research Projects (Project Nos. 2017S-101, 2018K- 257), and JST PRESTO (Grant No. JPMJPR132A). 18 Supplementary Material for "Electrically driven spin torque and dynamical Dzyaloshinskii-Moriya interaction in magnetic bilayer systems" Calculation of Ψ-electron spin density S± We provide details of the derivation of the Ψ-electron spin density S±. In Fig. 3, we present the Feynman diagrams associated with the Ψ-electron spin density induced by the time-dependent RSOI. Here the vertex corrections due to the nonmagnetic impurity scat- terings are not considered because they are negligible in the present case. A(so) S A(ex) FIG. 3: Diagrammatic representations of the dominant contributions of the Ψ-electron spin density S. The solid, wavy, and dotted lines represent the Green function, the RSOI A(so), and the exchange interaction A(ex) in the rotated spin frame, respectively. Expanding the lesser component with respect to q and Ω, the spin density S±(q, Ω) is 19 written S±(q, Ω) = (q′, Ω)A(ex)z ν (q − q′) + A(so)z µ (q′, Ω)A(ex)± ν (q − q′)i(cid:27) 2e µ  Xq′ hA(so)± k,ω,σ)2 − gr k,ω,−σ(ga µ ie4 m2 e (cid:26) ± qνA(so)± σXk Xω ×Xσ=± Ω(cid:20) ± qνA(so)± ie4 2m2 e + µ (q, Ω) − fωkµkνhga k,ω,−σ(gr k,ω,σ)2i A(so)± µ (q′, Ω)A(ex)z ν (q − q′)(cid:21) k,ω,σ(ga k,ω,−σ)2 − (gr k,ω,σ)2ga k,ω,−σ (q, Ω) − 2e  Xq′ dω(cid:20)gr kµkν(cid:26) dfω k,ω,−σ)ga k,ω,−σ + 1 2 gr k,ω,σ(gr k,ω,σ − gr k,ω,−σ)gr k,ω,−σ(cid:21) ×Xσ=± (σ ∓ 1)Xk Xω k,ω,σ − ga ga k,ω,σ(ga + 1 2 −fωhga k,ω,−σgr −gr µ ie23 m2 e + A(so)z ×Xσ=± ΩXq′ (σ ∓ 1)Xk Xω k,ω,σ(ga k,ω,σ − ga ga −fωh(ga k,ω,−σ)2(ga 1 2 − k,ω,−σga k,ω,σ(ga k,ω,σ − ga k,ω,−σ)2 + (ga k,ω,−σ)2(ga k,ω,σ)2 k,ω,σ(gr k,ω,σ − gr k,ω,−σ)2 − (gr k,ω,−σ)2(gr k,ω,σ)2i(cid:27) (q′, Ω)A(ex)± (q − q′) ν kµkν(cid:26) dfω dω(cid:20)gr k,ω,σga k,ω,−σga k,ω,σ − gr k,ω,−σgr k,ω,σga k,ω,−σ k,ω,−σ)ga k,ω,−σ − 1 2 gr k,ω,σ(gr k,ω,σ − gr k,ω,−σ)gr k,ω,−σ(cid:21) k,ω,σ)2 − (gr k,ω,−σ)2(gr k,ω,σ)2i(cid:27). (14) , (15) Using the following relation, ± i∇νA(so)± µ + 2e  the spin density S±(r, t) is rewritten as A(so)± µ A(ex)z ν = 2e  A(so)z µ A(ex)± ν S±(r, t) = 8e23 m2 e A(so)z µ (r, t)A(ex)± ν + 4e23Jex m2 e (r, t) µ ∂A(so)z ∂t A(ex)± ν ±iIm(cid:20)σ dfω dω gr k,ω,σ(ga k,ω,−σ)2ga (r)ImXσ=± σXk Xω (r)Xσ=±Xk Xω k,ω,σ + 2Jexfω(ga fωkµkνga k,ω,−σ(ga k,ω,σ)2 kµkν(cid:26)dfω dω k,ω,−σ)3(ga k,ω,σ)3(cid:21)(cid:27). Re(ga k,ω,σ − gr k,ω,σ)(ga k,ω,−σ)2ga k,ω,σ (16) Here we neglect contributions from the products of ga to calculate terms proportional to the first order in Ω as they only give small corrections compared with those from the products 20 of gr and ga. Therefore, the spin density reduces to S±(r, t) = 8e23 m2 e − Jex 2 ∂A(so)z µ A(ex)± ν (r)(cid:20)A(so)z µ (r, t)ImXσ=± σXk Xω (Re ∓ iσIm)Xk Xω 2 hC (2) (r, t) − Jex µ (r, t) ∂t Xσ=± (r)(cid:26)C (1) µν A(so)z = 8e2 me A(ex)± ν fωkµkνga k,ω,−σ(ga k,ω,σ)2 dfω dω kµkνgr k,ω,σ(ga k,ω,−σ)2ga k,ω,σ(cid:21) µν ∓ iC (3) µ µν i ∂A(so)z ∂t (r, t) (cid:27). (17) The summations over k and ω are performed in the following manner, (ω − ε + εF,−σ − iη)(ω − ε + εF,σ − iη)2 ε 2 me νe 2π C (1) µν = = = = 4πJ 2 ex 4πJ 2 ex fωkµkνga k,ω,−σ(ga k,ω,σ)2 νe νe 0 σXk Xω ImXσ=± δµνImXσ=±Z ∞ δµνImXσ=± δµνImXσ=± δµνImZ Jex δµνImZ Jex δµν(cid:26) εF ex − η2 Jex νe νe −Jex −Jex 4πJ 2 ex 4πJ 2 ex νeη 16πτ Jex F − J 2 ε2 Jexη ε ε − σJex −∞ −∞ −∞ −∞ −∞ dω dω dω dε Z 0 dε Z 0 dε Z 0 σZ 0 σZ −σJex dε Z 0 dε εh ln (ε + εF − iη) + iπi ln(cid:20) (εF + Jex)2 + η2 (εF − Jex)2 + η2(cid:21) − 2 εF + Jex(cid:19) − tan−1(cid:18) ω + ε + εF − iη dω −∞ η ε (cid:20) tan−1(cid:18) = − = − = − + ω + ε + εF,σ − iη ω + ε + εF − iη η εF − Jex(cid:19)(cid:21)(cid:27), (18) 21 C (2) µν = = = 2 me νe 2π ReXσ=±Xk Xω δµνReXσ=±Z 0 δµνReXσ=± −∞ νe 8πJex dfω dω kµkνgr k,ω,σ(ga k,ω,−σ)2ga k,ω,σ dε (ε + εF,σ + iη)(ε + εF,−σ − iη)2(ε + εF,σ − iη) ε σZ 0 −∞ dε(cid:20) 1 σεFJex i2η(σJex − iη)2(cid:18) ε + εF − σJex − iη(cid:19) − 1 − − 1 ε + εF + σJex − iη(cid:19) (ε + εF − σJex + iη)2(cid:21) 1 εF − σJex + iη σJex − iη ε + εF − σJex + iη − 1 i2η(cid:18) 1 ε + εF − σJex + iη = νe 8πJex δµνReXσ=± σ(cid:26) σεFJex i2η(σJex − iη)2(cid:20) ln(cid:18)εF − σJex + iη εF + σJex − iη(cid:19) − i2π(cid:21) σJex − iη(cid:27) 1 − = − − − 1 εF − σJex − iη(cid:19) − i2π(cid:21) + i2η(cid:20) ln(cid:18) εF − σJex + iη ex − η2) ex + η2) ex + η2)2 δµν(cid:26)1 − ex − η2) η νeτ εF(J 2 2(J 2 1 η(J 2 πεF(J 2 2π(cid:20) tan−1(cid:18) ex + η2)2 (J 2 εF + Jex(cid:19) + tan−1(cid:18) ex − η2)(cid:20) tan−1(cid:18) 2πεFJex(J 2 η εF − Jex(cid:19)(cid:21) + ηJex ex − η2) 2π(J 2 ln(cid:20) (εF + Jex)2 + η2 (εF − Jex)2 + η2(cid:21) η εF + Jex(cid:19) − tan−1(cid:18) η εF − Jex(cid:19)(cid:21)(cid:27), (19) C (3) µν = 2 me ImXσ=± σXk Xω δµνImXσ=±Z 0 −∞ νe 8πJex = dfω dω kµkνgr k,ω,σ(ga k,ω,−σ)2ga k,ω,σ dε(cid:20) εF(σ2Jex − iη) σ2Jex(σJex − iη)2(cid:18) 1 ε + εF − σJex + iη − 1 ε + εF + σJex − iη(cid:19) εF − σJex + iη σJex − iη 1 (ε + εF − σJex + iη)2(cid:21) − 1 σ2Jex(cid:18) = νe 8πJex δµνImXσ=± ε + εF − σJex + iη 1 1 − ε + εF − σJex − iη(cid:19) − σ2Jex(σJex − iη)2(cid:20) ln(cid:18)εF − σJex + iη (cid:26) εF(σ2Jex − iη) εF − σJex − iη(cid:19) − i2π(cid:21) + 1 1 σ2Jex(cid:20) ln(cid:18) εF − σJex + iη εF + σJex − iη(cid:19) − i2π(cid:21) σJex − iη(cid:27) ex + η2) 8πJ 3 ex ln(cid:20) (εF + Jex)2 + η2 (εF − Jex)2 + η2(cid:21) νeεFJex ex + η2)2 δµν(cid:26)1 − η(J 2 ex + η2) 2πεFJ 2 ex + η(3J 2 η εF + Jex(cid:19) + tan−1(cid:18) η εF − Jex(cid:19)(cid:21) 2(J 2 1 2π(cid:20) tan−1(cid:18) ex + η2)2 4πεFJ 3 (J 2 ex (cid:20) tan−1(cid:18) − = − − − η εF + Jex(cid:19) − tan−1(cid:18) η εF − Jex(cid:19)(cid:21)(cid:27). (20) 22 

1808.05960

2

1808

Dynamical density response and optical conductivity in topological metals

Topological metals continue to attract attention as novel gapless states of matter. While there by now exists an exhaustive classification of possible topologically nontrivial metallic states, their observable properties, that follow from the electronic structure topology, are less well understood. Here we present a study of the electromagnetic response of three-dimensional topological metals with Weyl or Dirac nodes in the spectrum, which systematizes and extends earlier pioneering studies. In particular, we argue that a smoking-gun feature of the chiral anomaly in topological metals is the existence of propagating chiral density modes even in the regime of weak magnetic fields. We also demonstrate that the optical conductivity of such metals exhibits an extra peak, which exists on top of the standard metallic Drude peak. The spectral weight of this peak is transferred from high frequencies and its width is proportional to the chiral charge relaxation rate.

cond-mat.mes-hall

cond-mat

Dynamical density response and optical conductivity in topological metals Department of Physics and Astronomy, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada (Dated: October 23, 2018) A.A. Burkov Topological metals continue to attract attention as novel gapless states of matter. While there by now exists an exhaustive classification of possible topologically nontrivial metallic states, their observable properties, that follow from the electronic structure topology, are less well understood. Here we present a study of the electromagnetic response of three-dimensional topological metals with Weyl or Dirac nodes in the spectrum, which systematizes and extends earlier pioneering studies. In particular, we argue that a smoking-gun feature of the chiral anomaly in topological metals is the existence of propagating chiral density modes even in the regime of weak magnetic fields. We also demonstrate that the optical conductivity of such metals exhibits an extra peak, which exists on top of the standard metallic Drude peak. The spectral weight of this peak is transferred from high frequencies and its width is proportional to the chiral charge relaxation rate. I. INTRODUCTION Topological metal (TM) is a recently discovered new phase of matter.1 -- 18 It is characterized by topological in- variants, defined on the Fermi surface,19 -- 22 rather than in the whole Brillouin zone (BZ), as in topological insulators (TI). Such Fermi surface topological invariants arise as a consequence of monopole-like singularities in the elec- tronic structure, Weyl nodes, whose significance was em- phasized early on by Volovik and by Murakami.19,22 Perhaps the most interesting feature of TM is that their electronic structure topology leads not only to spec- troscopic manifestations in the form of edge states,5 a feature they share with TI, but also to nontrivial re- sponse. This novel response is usually described as being a consequence of the chiral anomaly,23 which may be un- derstood in the following way. While the appearance of gapless Weyl nodes in the spectrum has a topological ori- gin, it also leads to an emergent symmetry, or an emer- gent conservation law, namely conservation of the chiral charge. This conservation law becomes increasingly more precise as the Fermi energy of the TM approaches the Weyl nodes. However, this apparent low-energy conser- vation law is violated when the system is coupled to an electromagnetic field. The origin of this violation lies in the fact that the chiral symmetry can never be an exact symmetry of a (3+1)-dimensional Dirac fermion on a lat- tice, as first pointed out by Nielsen and Ninomiya,24 as a single (or, more generally, an odd number) Dirac point in the BZ is topologically incompatible with the chiral symmetry. Thus, while the chiral symmetry appears to be present when one focuses only on states at the small Fermi surface, enclosing the Weyl points, the global lack of chiral symmetry manifests in the electromagnetic re- sponse of the system. This property is of great interest both because it has a topological origin and because it is contrary to one of the fundamental postulates of the standard theory of metals, which states that anything of observable consequence in a metal involves only states on the Fermi surface. The chiral anomaly in TM has numerous predicted ob- servable consequences, which include negative longitudi- nal magnetoresistance (LMR),25,26 giant planar Hall ef- fect (PHE),27,28 and anomalous Hall effect.29 While most of these have already been observed experimentally in various TM materials,18,30 -- 34 none of these phenomena by themselves may be regarded as smoking-gun manifes- tations of the chiral anomaly, in the sense that all of them may in principle arise from unrelated sources, and these sources all have to be ruled out before the chiral anomaly origin may be claimed. An excellent discussion of these issues in the case of the negative LMR may be found in Ref. 35. As first discussed by Altland and Bagrets,36 a truly unique feature of the chiral anomaly is the highly un- usual dependence of the transport properties, such as the sample conductance, on the relevant length (and time or frequency, as will be shown in this paper) scales. In an ordinary three-dimensional (3D) metal the conductance scales linearly with the sample size L G(L) = σL, (1) where the Drude conductivity σ is related to the density of states at the Fermi energy g and the diffusion constant D by the Einstein relation σ = e2gD. (2) Corrections to Eq. (1) are small in good metals, the small parameter being 1/kF (cid:96), where kF is the Fermi momen- tum (¯h = c = kB = 1 units are used henceforth) and (cid:96) is the mean free path; the corrections arise only at very low temperatures as a result of quantum interference phe- nomena. The scaling of Eq. (1) is partly a consequence of the fact that, in an ordinary metal in the diffusive trans- port regime, i.e. at length scales, longer than the mean free path (cid:96) and time scales longer than the momentum relaxation time τ , no intrinsic hydrodynamic (i.e. long) length scales remain, besides the sample size L. However, as discussed in Ref. 36, in a TM two addi- tional hydrodynamic length scales emerge. These are the chiral charge diffusion length (cid:112) Lc = Dτc, (3) 8 1 0 2 t c O 2 2 ] l l a h - s e m . t a m - d n o c [ 2 v 0 6 9 5 0 . 8 0 8 1 : v i X r a G(L) = e2Nφ 2π , (5) where where τc (cid:29) τ is the chirality relaxation time, and La = D/Γ, (4) where Γ = eB/2π2g and B is the applied magnetic field. La is a new purely quantum mechanical magnetic-field- √ related length scale, which is distinct from the magnetic eB and which arises from the chiral length (cid:96)B = 1/ anomaly. It is related to the magnetic length as La ∼ (cid:96)(kF (cid:96)B)2 and is thus much longer than the mean free path in the weak-field (quasiclassical) regime kF (cid:96)B (cid:29) 1, which we will be interested in here. Transport properties of TM may then be shown to depend strongly on the interplay of the three length scales: L, Lc, and La.27,36 In particular, the strength of the negative LMR and the PHE depends on the parameter Lc/La, getting stronger as this ratio increases. Particularly striking phenomena arise when La < L < c/La,36 which is an extended and accessible range when L2 Lc/La (cid:29) 1. In this regime the sample conductance is given by where Nφ = L2/2π(cid:96)2 B is the number of magnetic flux quanta, piercing the sample with cross-section area L2. This means that in the regime La < L < L2 c/La the sample transports electric current as Nφ one-dimensional (1D) conduction channels and the conduction is ballistic and dissipationless [of course Eq. (5) only represents the dominant part of the conductance and ordinary dissipa- tive Ohmic conduction is also present]. This is striking because it arises in a 3D metal with a Fermi surface and in the weak magnetic field regime kF (cid:96)B (cid:29) 1. The ex- istence of such ballistic quasi-1D transport regime is a smoking-gun manifestation of the chiral anomaly in 3D TM. In this paper we further elaborate on this striking prop- erty of TM and consider their related dynamical proper- ties. In particular, we demonstrate that the quasi-1D transport regime manifests in dynamics as chiral propa- gating density modes, which exist in a range of wavevec- tor values given by La/L2 c < q < 1/La. (6) This "one-dimensionalization" of the electron dynamics is a unique property of TM, related to the chiral anomaly. We also demonstrate that related phenomena exist in frequency-dependent properties of TM. In particular we demonstrate that the frequency dependence of the op- tical conductivity of TM has a non-Drude form, where an extra narrow peak exists at low frequencies, whose width scales as 1/τc while height is a function of the ratio Lc/La. The spectral weight of this extra peak is transferred from high frequencies. The rest of the paper is organized as follows. In Sec- tion II we calculate the full density response function of a simple model of a TM in an external magnetic field. 2 We analyze the eigenmode structure of the density re- sponse function and demonstrate the presence of chiral propagating density modes when La/L2 c < q < 1/La. In Section III we relate the existence of these propagating chiral modes to observable transport properties of TM. We also demonstrate that similar phenomena exist in the frequency domain: we analyze the frequency dependence of the optical conductivity and point out its non-Drude nature. We conclude in Section IV with a brief discussion of the main results. II. DENSITY RESPONSE FUNCTION OF A TOPOLOGICAL METAL We start from the simplest model of a TM, which con- tains the necessary ingredients to capture the physics we want to describe. The simplest such model is the follow- ing model of a lattice Dirac fermion H = tγ0γµ sin kµ + ∆(k)γ0, ∆(k) = t(3 − cos kx − cos ky − cos kz), (7) (8) and γµ are Dirac gamma matrices in, for example, the Weyl representation γ0 = τ x, γi = −iτ yσi, i = 1, 2, 3. (9) This model describes two Weyl nodes of opposite chi- rality at the Γ-point in the BZ (the effects we will be discussing do not depend on the momentum-space sep- aration between the Weyl nodes). Since a single Dirac point in the BZ is incompatible with the chiral symme- try, Eq. (7) also has an essential property, shared by all real Weyl and Dirac semimetals, that the chiral symme- try (chiral charge conservation) is only an approximate low-energy symmetry of Eq. (7), which emerges when H is expanded to linear order in k near the Γ-point. In this case we have H = tγ0γµkµ, (10) and the chirality operator γ5 = iγ0γ1γ2γ3 = τ z com- mutes with H, which is no longer true once nonlinear terms are included. This gives a finite (but small) chiral charge relaxation rate, which is an essential property of a Weyl or Dirac semimetal. We add a uniform magnetic field in the z-direction B = B z, and choose the Landau gauge for the vector potential A = xB y. We will ignore the Zeeman effect for simplicity. To find the eigenstates of H in the presence of the magnetic field, we expand to first order in kx,y, while keeping the full kz dependence. This is a good ap- proximation in the regime of weak magnetic fields when kF (cid:96)B (cid:29) 1, which we will be interested in. For computa- tional convenience we also make the following canonical transformation in the original Weyl representation of the gamma-matrices: τ x,y → σzτ x,y, σx,y → τ zσx,y. (11) This brings the Hamiltonian to the form H = t(σxπx + σyπy) + m(kz)σz, (12) where π = −i∇ + eA is the canonical momentum and m(kz) = tτ z sin kz + ∆(0, 0, kz)τ x. (13) Diagonalizing Eq. (12), we find the eigenstate wavefunc- tions n, s, p, ky, kz(cid:105) = zsp + zsp n↑τ (kz)n − 1, ky, kz,↑, τ(cid:105) n↓τ (kz)n, ky, kz,↓, τ(cid:105), (14) where n is an integer Landau level index, ↑,↓ label the two eigenvalues of σz, τ = ± are the two eigenval- ues of τ z, and s, p = ±. Here and throughout sums over repeated indices will be implicit. The amplitudes zsp nστ (kz) may be regarded as components of an eigenvec- tor zsp (cid:115) (cid:32)(cid:115) n (kz)(cid:105) ⊗ up(kz)(cid:105), where n (kz)(cid:105) = vsp (cid:115) (cid:32)(cid:115) t sin kz m(kz) 1√ 2 1 + p 1 − p , p up n(kz)(cid:105) = (cid:33) (cid:33) , t sin kz m(kz) vsp n (kz)(cid:105) = 1√ 2 1 + sp m(kz)
n(kz) , s 1 − sp m(kz)
n(kz) . (15) The corresponding energy eigenvalues are given by (cid:113) 2ω2 Bn + m2(kz),
nsp(kz) = s
n(kz) = s (16) where m(kz) = 2t sin kz, and ωB = t/(cid:96)B, for all n ≥ 1. The lowest Landau level (LLL), corresponding to n = 0, is special: it does not have the s label and its eigenenergy and the corresponding eigenvector are given by and
0p(kz) = −pm(kz), 0(kz)(cid:105) = (0, 1). vp (17) (18) We add to the Hamiltonian Eq. (12) random impurity potential V (r), which we take to be of the Gaussian white noise form with (cid:104)V (r)(cid:105) = 0 and (cid:104)V (r)V (r(cid:48))(cid:105) = γ2δ(r − r(cid:48)). (19) We take the impurity potential to be independent of the spin and orbital pseudospin indices. Physically this means that the impurities are taken to be nonmagnetic and the potential is smooth enough that its spatial vari- ation on the scale of the unit cell of the crystal is neg- ligible. The last assumption is not essential, but does simplify the subsequent calculations. 3 We will evaluate the density response for the above model of a TM using the self-consistent Born approxi- mation (SCBA) and the ladder approximation to perform the impurity averaging. This is a conserving approxima- tion, meaning it preserves exact conservation laws and sum rules, and amounts physically to neglecting quantum interference effects. This is justified in the quasiclassical transport regime, which we will confine ourselves to: we assume that we are interested in the density response at length scales much longer than the inverse Fermi mo- mentum and time scales much longer than the inverse Fermi energy; the impurity scattering is taken to be weak enough, so that kF (cid:96) (cid:29) 1 and, as already mentioned, magnetic field is also assumed to be weak, which means kF (cid:96)B (cid:29) 1. Finally, we will assume that the Fermi energy is close to the Dirac point
F (cid:28) t (but
F τ (cid:29) 1), which defines the regime of a TM. The last condition ensures the near conservation of the chiral charge, as will be seen explicitly below. The calculation of the SCBA impurity self-energy in a similar model has already been discussed in detail in Ref. 26. We will thus omit the details of this calculation here and simply quote the result. One obtains that in the quasiclassical transport regime the impurity scattering rate is independent of both the Landau level index n and the longitudinal momentum component kz and is given by the standard SCBA expression 1 τ = πγ2g 2 , (20) where the density of states at the Fermi energy is given by g =
F πt2 Θ[
F − m(kz)], dkz 2π (21) (cid:90) π −π Θ(x) being the Heaviside step function. We evaluate the density response function by summing the impurity ladder diagrams. We start from the most general retarded density matrix response function, de- fined as χα1α2,α3α4(r, tr(cid:48) (cid:48)) , t (cid:48))(cid:104)[
α1α2 † (r, t),
α3α4 = −iΘ(t − t (r(cid:48) (cid:48))](cid:105), , t (22) where the density matrix is defined as
α1α2(r, t) = Ψ† α2 (r, t)Ψα1 (r, t), (23) and α = (στ ) is a composite index, which encodes both the spin and orbital pseudospin labels. The standard procedure to find the real-time response function Eq. (22) is to start from the corresponding imaginary-time response function χα1α2,α3α4(r, τr(cid:48) (cid:48)) , τ , τ − τ (cid:48))Gα4α2(r(cid:48) = −Gα1α3 (r, r(cid:48) (24) where Gαα(cid:48)(r, r(cid:48), τ − τ(cid:48)) is the exact imaginary-time Green's function, which depends on both r and r(cid:48) sepa- rately due to both the lack of translational symmetry in , r, τ (cid:48) − τ ), 4 FIG. 1. Diagrammatic representation of (a) SCBA Green's function. Thin line represents the bare Green's function, thick line is the SCBA impurity-averaged Green's function and the dashed line represents the disorder potential correla- tor (cid:104)V (r)V (r(cid:48))(cid:105) = γ2δ(r − r(cid:48)). (b) Density response function χ. (c) Bethe-Salpeter equation for the diffusion vertex D. The physical meaning of the two contributions to the density response function, χI and χII , is that χI arises from states on the Fermi surface, while all filled states contribute to χII . χII thus represents equilibrium part of the response and is easily shown to be a diagonal matrix, with the nonzero matrix elements equal to −g. On the other hand, χI represents the dynamical nonequilibrium part of the density response and is given by χI (q, Ω) = − iΩ 2πγ2I(q, Ω)D(q, Ω), (33) d3r d3r (cid:48) e −iq·(r−r(cid:48)) Iα1α2,α3α4 (q, Ω) = , Ω)GA α1α3 γ2 L3 (r(cid:48) α4α2 , r, 0), × GR (r, r(cid:48) GR,A being the retarded and advanced real-time impurity-averaged SCBA Green's functions. They are explicitly given by (cid:88) αα(cid:48) (r, r(cid:48) GR,A (cid:104)r, αn, s, p, ky, kz(cid:105)(cid:104)n, s, p, ky, kzr(cid:48), α(cid:48)(cid:105) (34) , ω) ω − ξnsp(kz) ± i/2τ , = nspkykz (35) where ξnsp(kz) ≡
nsp(kz) −
F . For a general direction of the wavevector q, the eval- uation of I(q, Ω) is severely complicated by the fact that contributions of different Landau levels are mixed in Eq. (34). This is not the case only when q = qz, when translational symmetry in the xy-plane leads to decoupling of the individual Landau level contributions. D = 1 + ID, where I ≡ γ2P 0. The solution of this equation is (29) where (cid:90) (cid:88) the presence of a random impurity potential, and the lack of gauge invariance in the presence of an external mag- netic field. One then performs impurity averaging, which restores translational invariance in the density response function and gives χα1α2,α3α4 (q, iΩ) = 1 β where Pα1α2,α3α4(q, iω, iω + iΩ), iω (25) Pα1α2,α3α4(r − r(cid:48) = −(cid:104)Gα1α3 (r, r(cid:48) , iω, iω + iΩ) , iω + iΩ)Gα4α2 (r(cid:48) , r, iω)(cid:105), (26) is the impurity-averaged generalized polarization bubble, and β = 1/T is the inverse temperature. In the quasiclas- sical regime we are interested in, P may be evaluated by summing all the SCBA diagrams for the impurity self- energy and the ladder vertex corrections, as shown in Fig. 1. The result of this diagram summation may be written in a shorthand matrix notation as P = P 0D, (27) where P 0 is the bare polarization bubble, in which only the self-energy corrections are included P 0 α1α2,α3α4 = −Gα1α3 (r, r(cid:48) is (r − r(cid:48) , iω, iω + iΩ) , iω + iΩ)Gα4α2 (r(cid:48) Gαα(cid:48)(r, r(cid:48), iω) here the disorder-averaged SCBA Green's function, which still depends on r and r(cid:48) sepa- rately since it is a gauge-dependent quantity in the pres- ence of an external magnetic field. The vertex part D, which is also known as the diffusion propagator, or dif- fuson, satisfies the following Bethe-Salpeter equation , r, iω). (28) D = (1 − I)−1. (30) (cid:90) ∞ To obtain the real-time retarded response function we analytically continue to real frequency iΩ → Ω+iη, which gives −∞ d
2πi χ(q, Ω) = nF (
) [P(q,
+ iη,
+ Ω + iη) − P(q,
− iη,
+ Ω + iη) + P(q,
− Ω − iη,
+ iη) − P(q,
− Ω − iη,
− iη)] . (31) In the low-frequency limit, when Ω (cid:28)
F , this simplifies to χ(q, Ω) = − iΩ 2π (cid:90) ∞ P(q,−iη, Ω + iη) d
nF (
)ImP(q,
+ iη,
+ Ω + iη) − 1 π ≡ χI (q, Ω) + χII (q, Ω). −∞ (32) =+=+<latexit sha1_base64="TwKQu5I6W5bc9HOXn7rGty6B3x4=">AAAB63icbVBNSwMxEJ3Ur1q/qh69BIvgqeyKoMeiF48VbC20S8mm2W5okl2SrFCW/gUvHhTx6h/y5r8x2+5BWx8MPN6bYWZemApurOd9o8ra+sbmVnW7trO7t39QPzzqmiTTlHVoIhLdC4lhgivWsdwK1ks1IzIU7DGc3Bb+4xPThifqwU5TFkgyVjzilNhCGtCYD+sNr+nNgVeJX5IGlGgP61+DUUIzyZSlghjT973UBjnRllPBZrVBZlhK6ISMWd9RRSQzQT6/dYbPnDLCUaJdKYvn6u+JnEhjpjJ0nZLY2Cx7hfif189sdB3kXKWZZYouFkWZwDbBxeN4xDWjVkwdIVRzdyumMdGEWhdPzYXgL7+8SroXTd9r+veXjdZNGUcVTuAUzsGHK2jBHbShAxRieIZXeEMSvaB39LForaBy5hj+AH3+AP+rjjI=</latexit><latexit sha1_base64="TwKQu5I6W5bc9HOXn7rGty6B3x4=">AAAB63icbVBNSwMxEJ3Ur1q/qh69BIvgqeyKoMeiF48VbC20S8mm2W5okl2SrFCW/gUvHhTx6h/y5r8x2+5BWx8MPN6bYWZemApurOd9o8ra+sbmVnW7trO7t39QPzzqmiTTlHVoIhLdC4lhgivWsdwK1ks1IzIU7DGc3Bb+4xPThifqwU5TFkgyVjzilNhCGtCYD+sNr+nNgVeJX5IGlGgP61+DUUIzyZSlghjT973UBjnRllPBZrVBZlhK6ISMWd9RRSQzQT6/dYbPnDLCUaJdKYvn6u+JnEhjpjJ0nZLY2Cx7hfif189sdB3kXKWZZYouFkWZwDbBxeN4xDWjVkwdIVRzdyumMdGEWhdPzYXgL7+8SroXTd9r+veXjdZNGUcVTuAUzsGHK2jBHbShAxRieIZXeEMSvaB39LForaBy5hj+AH3+AP+rjjI=</latexit><latexit sha1_base64="TwKQu5I6W5bc9HOXn7rGty6B3x4=">AAAB63icbVBNSwMxEJ3Ur1q/qh69BIvgqeyKoMeiF48VbC20S8mm2W5okl2SrFCW/gUvHhTx6h/y5r8x2+5BWx8MPN6bYWZemApurOd9o8ra+sbmVnW7trO7t39QPzzqmiTTlHVoIhLdC4lhgivWsdwK1ks1IzIU7DGc3Bb+4xPThifqwU5TFkgyVjzilNhCGtCYD+sNr+nNgVeJX5IGlGgP61+DUUIzyZSlghjT973UBjnRllPBZrVBZlhK6ISMWd9RRSQzQT6/dYbPnDLCUaJdKYvn6u+JnEhjpjJ0nZLY2Cx7hfif189sdB3kXKWZZYouFkWZwDbBxeN4xDWjVkwdIVRzdyumMdGEWhdPzYXgL7+8SroXTd9r+veXjdZNGUcVTuAUzsGHK2jBHbShAxRieIZXeEMSvaB39LForaBy5hj+AH3+AP+rjjI=</latexit><latexit sha1_base64="TwKQu5I6W5bc9HOXn7rGty6B3x4=">AAAB63icbVBNSwMxEJ3Ur1q/qh69BIvgqeyKoMeiF48VbC20S8mm2W5okl2SrFCW/gUvHhTx6h/y5r8x2+5BWx8MPN6bYWZemApurOd9o8ra+sbmVnW7trO7t39QPzzqmiTTlHVoIhLdC4lhgivWsdwK1ks1IzIU7DGc3Bb+4xPThifqwU5TFkgyVjzilNhCGtCYD+sNr+nNgVeJX5IGlGgP61+DUUIzyZSlghjT973UBjnRllPBZrVBZlhK6ISMWd9RRSQzQT6/dYbPnDLCUaJdKYvn6u+JnEhjpjJ0nZLY2Cx7hfif189sdB3kXKWZZYouFkWZwDbBxeN4xDWjVkwdIVRzdyumMdGEWhdPzYXgL7+8SroXTd9r+veXjdZNGUcVTuAUzsGHK2jBHbShAxRieIZXeEMSvaB39LForaBy5hj+AH3+AP+rjjI=</latexit>=(a)(b)(c)5 near conservation is a defining property of a TM, as dis- cussed above. Thus we may project the original 16 × 16 matrix onto the 2 × 2 subspace, describing the coupled transport of the electric and the chiral charge, which is accomplished as Iab(q, Ω) = Iα1α2,α3α4(q, Ω)Γb (36) Γa α3α4 , α2α1 1 4 where a, b = 0, 5, corresponding to the electric (0) or chi- ral (5) charges, and Γa,b are the corresponding operators, i.e. Γ0 = τ 0σ0 = 1, Γ5 = τ zσ0 = τ z. (37) After a tedious, but straightforward, calculation, we ob- tain Fortunately, this is in fact the case of primary interest to us, since the chiral anomaly leads to unusual transport phenomena in the direction of the magnetic field. Thus we will take q = qz henceforth. In this case the evaluation of I(q, Ω) is relatively straightforward, particularly in the weak magnetic field regime kF (cid:96)B (cid:29) 1 that we are interested in. An addi- tional simplification arises from the fact that we are not interested in the whole 16 × 16 matrix I, which contains a lot of unnecessary information. We are interested only in the response of conserved, or nearly conserved, quan- tities, which will always dominate everything else at long times and long distances. In a generic TM, we expect only two such quantities to exist: the electric charge, which is strictly conserved, and the chiral charge, whose (cid:18) 1 − iΩτ − iq(cid:96) (cid:19) − i(
F /t)2(1 − iΩτ )4 1 − iΩτ + iq(cid:96) , ln 8(q(cid:96))5 i 2q(cid:96) (cid:20) i I00(q, Ω) = I55(q, Ω) = I05(q, Ω) = I50(q, Ω) = 2q(cid:96) (cid:21) ln (cid:18) 1 − iΩτ − iq(cid:96) 1 − iΩτ + iq(cid:96) (cid:19) − (
F /t)2(1 − iΩτ ) 12(q(cid:96))2 + (
F /t)2(1 − iΩτ )3 4(q(cid:96))4 , i 2(kF (cid:96)B)2 q(cid:96) (1 − iΩτ )2 + (q(cid:96))2 . (38) Substituting this into Eq. (33), we obtain the dynami- cal nonequilibrium contribution to the density response χI (q, Ω), while the equilibrium contribution is a diagonal matrix given by 55(q, Ω) = −g, (39) χII 00(q, Ω) = χII as already mentioned above. A comment is in order here. As can be seen from Eq. (38), only the off-diagonal matrix element I05 de- pends on the magnetic field. This is true in the quasi- classical limit kF (cid:96)B (cid:29) 1 only, and is a consequence of the fact that in this limit we may ignore the effect of the magnetic field on the density of states. Summation over the Landau level index n, which arises when evaluat- ing Eq. (38), may in this case be replaced by integration and the magnetic field dependence disappears to lead- ing order in 1/kF (cid:96)B. In contrast, the off-diagonal matrix element I05 arises entirely from the contribution of the n = 0 Landau level. This contribution is proportional to 1/(kF (cid:96)B)2, but leads to large effects at long length scales and long times, as will be seen below, provided τc/τ (cid:29) 1. Eqs. (32), (33), (38) and (39) give a general expression for the density response function of a TM in the quasi- classical regime χ(q, Ω) = −g[iΩτI(q, Ω)D(q, Ω) + 1]. (40) This expression is valid in either diffusive Ωτ, q(cid:96) (cid:28) 1 or ballistic Ωτ, q(cid:96) (cid:29) 1 limits and may be used, in particular, to study the ballistic-diffusive crossover regime. We will start by analyzing the two limits. A. Ballistic regime In this regime all components of the matrix I are small and thus D ≈ 1. Physically this means that we are look- ing at short length and time scales at which the impurity scattering may be ignored. While the response function χ(q, Ω) is a 2 × 2 matrix, only its χ00(q, Ω) component describes observable density response. Taking the limit Ωτ, q(cid:96) → ∞ in Eqs. (38), (40) we obtain (cid:18) Ω − qt + iη (cid:19)(cid:21) Ω + qt + iη . (41) χ00(q, Ω) = −g 1 + Ω 2qt ln (cid:20) This is just the familiar Lindhard function (in the limit q (cid:28) kF and Ω (cid:28)
F ), describing the density response of a clean Fermi liquid with the Fermi velocity t. The imaginary part of χ00(q, Ω), which is determined by the branch cuts of the Lindhard function Imχ00(q, Ω) = g Θ(qt − Ω), πΩ 2qt (42) describes the excitation spectrum of the Fermi liquid, which forms a particle-hole continuum. Thus in the bal- listic regime and in the weak magnetic field limit chi- ral anomaly has no effect on the density response of a TM [its effects appear only at order 1/(kF (cid:96)B)2, which is negligible compared to Eq. (41)]. This of course will no longer be true if we tune the Fermi energy to zero (i.e. to the ideal Weyl or Dirac semimetal limit), but this is a fine-tuned, non-generic situation, and is of somewhat less interest for this reason. B. Diffusive regime The situation is much more interesting in the diffusive limit Ωτ, q(cid:96) (cid:28) 1. In this case I ≈ 1, and multiple impu- rity scattering needs to be taken into account. Expanding in Taylor series in Ωτ and q(cid:96), we obtain I00(q, Ω) ≈ 1 + iΩτ − Dq2τ, I05(q, Ω) = I50(q, Ω) ≈ iΓqτ, I55(q, Ω) ≈ 1 + iΩτ − τ /τc − Dq2τ. (43) Here D = t2τ /3 = t(cid:96)/3 is the diffusion constant, Γ = eB 2π2g = t 2(kF (cid:96)B)2 , (44) is a new transport coefficient, which describes the chiral- anomaly-induced coupling between the electric and the chiral charge densities, and 1 τc =
2 F 20 t2τ , (45) is the chiral charge relaxation rate. Note that the fact the chiral charge relaxation rate vanishes in the limit
F → 0 is a consequence of our assumption that the impurity potential is diagonal in the spin and orbital indices and thus commutes with the chiral charge operator γ5 = τ z. In general this is not the case and we can expect some residual chiral charge relaxation even in the
F → 0 limit. In the diffusive regime the dynamics of the density re- sponse is determined by the poles of the diffusion propa- gator D, instead of the branch cuts of the response func- tion, as in the ballistic limit. From Eq. (43), the inverse diffusion propagator is given by D−1(q, Ω) = (cid:18) −iΩτ + Dq2τ −iΓqτ (cid:19) −iΩτ + τ /τc + Dq2τ −iΓqτ , 6 which gives the following explicit expression for the 00 component of the matrix response function, which corre- sponds to the observable electric charge density response (cid:20) Ω(Ω + i/τc + iDq2) (Ω − Ω+)(Ω − Ω−) (cid:21) − 1 . (51) χ00(q, Ω) = g We now note that the frequency Ω0 is purely imaginary at the smallest momenta when where we have introduced two new length scales q < 1 2Γτc = La 2L2 c ≡ 1 L∗ , (cid:112) Lc = Dτc, (52) (53) which has the meaning of the chiral charge diffusion length and La = D Γ = 2 3 (cid:96)(kF (cid:96)B)2. (54) La is a magnetic-field-related length scale, distinct from the magnetic length, which arises from the chiral anomaly. It is a long hydrodynamic length scale in the weak magnetic field regime, in the sense that La (cid:29) (cid:96), but it may still be much smaller that either the chiral charge diffusion length Lc or the sample size L. In fact, the ra- tio Lc/La quantifies the strength of the chiral-anomaly- related density response phenomena, as will be seen be- low. Thus when q < 1/L∗ the eigenfrequencies of the diffu- sion propagator are purely imaginary, which corresponds to ordinary diffusion (nonpropagating) modes. However, when q > 1/L∗ (which may be a very small momentum when the ratio Lc/La is large), Ω0 is real, which signals the emergence of a pair of propagating modes in this regime. The modes are only weakly damped as long as (46) Ω0 ≈ Γq > Dq2, (55) The zeros of the determinant of this matrix determine the eigenmode frequencies where Ω± = ±Ω0 − i(Dq2 + 1/2τc), Ω0 =(cid:112)Γ2q2 − 1/4τ 2 c . (47) (48) which defines the upper limit on the wavevector q = 1/La, above which the propagating modes disappear. The propagating modes thus exist in the interval 1/L∗ < q < 1/La. (56) This interval is significant when Lc/La (cid:29) 1. Within this interval of q the density response function The diffusion propagator itself may then be written as takes the following approximate form D(q, Ω) = 1/τ (Ω − Ω+)(Ω − Ω−) (cid:18) iΩ − Dq2 − 1/τc × −iΓq (49) Taking into account that in the diffusive regime I ≈ 1, iΩ − Dq2 −iΓq . we obtain from Eq. (40) χ(q, Ω) ≈ −g[iΩτD(q, Ω) + 1], (50) (cid:19) χ00(q, Ω) = g Ω2 0 (Ω + iDq2)2 − Ω2 0 , (57) where Ω0 = Γq. This is the density response func- tion of an effective 1D system with the Fermi velocity Γ = t/2(kF (cid:96)B)2 (cid:28) t. Note that this is very different from the 1D response one would obtain in a TM in the quantum limit kF (cid:96)B < 1, when only the lowest n = 0 Landau level contributes to the density response. In this case one gets Nφ 1D modes, which correspond to Nφ or- bital states within the LLL. The Fermi velocity of these 1D modes is equal to the microscopic Fermi velocity t. In our case, while the ultimate origin of the 1D dynamics is still the LLL, its emergence is only possible in the diffu- sive regime and thus requires multiple impurity scatter- ing. The corresponding Fermi velocity Γ is proportional to the applied magnetic field and is much smaller than t in the quasiclassical weak-field regime. Such effectively 1D density response, with propagating rather than diffu- sive density dynamics, which exists in a 3D metal with a large Fermi surface (kF (cid:96) (cid:29) 1) in a weak magnetic field (kF (cid:96)B (cid:29) 1) is a truly unique feature of TM and should be regarded as their true smoking-gun characteristic. On the other hand, when La > Lc, propagating modes do not exist for any q and one obtains a pair of standard diffusion modes Ω+ = −iDq2, Ω− = −iDq2 − i/τc, (58) which correspond to independent diffusion of the electric and the chiral charge densities. III. TRANSPORT IN TOPOLOGICAL METALS It is very useful to also look at the transport properties, which follow from the density response, described in Sec- tion II. In addition to providing further insight into the physical meaning of the results, discussed in the previous section, this will also allow us to calculate experimentally measurable physical quantities, such as the frequency- and scale-dependent conductivity. A. Scale-dependent conductance It is easy to show that Eqs. (49) and (50) for the dif- fusion propagator and the generalized density response function are equivalent to the following transport equa- tion in real space and time, that the electric n0 and chiral n5 charge densities must satisfy ∂n0 ∂t ∂n5 ∂t = D∇2(n0 + gV0) + Γ · ∇(n5 + gV5), = D∇2(n5 + gV5) − n5 + gV5 τc + Γ · ∇(n0 + gV0), (59) where V0 and V5 are external electric and chiral poten- tials correspondingly and we have generalized to an ar- bitrary magnetic field direction, which is why the coeffi- cient Γ ∝ B has become a vector. The chiral potential V5 may arise, for example, in a situation when the inver- sion symmetry is broken, in which case the Weyl nodes of different chirality will generally be located at differ- ent energies, V5 being precisely this energy difference. Otherwise this should simply be regarded as a fictitious potential, which couples linearly to the chiral charge nc. Indeed, Fourier transforming Eq. (59) we obtain 7 (cid:19) . (60) (cid:19) (cid:18) n0 n5 = −g(cid:2)iΩτ + D−1(q, Ω)(cid:3)(cid:18) V0 (cid:19) (cid:18) V0 V5 = −g[iΩτD(q, Ω) + 1] , (61) V5 D−1(q, Ω) This gives(cid:18) n0 (cid:19) n5 which is equivalent to Eq. (50). Solving Eq. (59) in the steady state, assuming a uni- form sample of linear size L, attached to normal metal leads (in which the chiral electrochemical potential n5 + gV5 = 0) in the z-direction (i.e. the current flows along the magnetic field), one obtains the following expression for the scale-dependent sample conductance27,36 G(L) = e2Nφ 2π F (L/La, L/Lc), (62) where the scaling function F (x, y) is given by (cid:112)1 + y2/x2 + tanh (1 + y2/x2)3/2 (cid:112)1 + y2/x2 (cid:16) x 2 (cid:17) . (63) F (x, y) = y2 2x This scaling function exhibits crossover behaviors which exactly match the corresponding crossovers in the wavevector dependence of the diffusion modes, described in Section II. Indeed, when x (cid:28) y, which means La (cid:29) Lc, we have F (x, y) ≈ 2/x, which gives G(L) ≈ e2gDL = σL, (64) which is simply the standard Ohmic conductance, with a small magnetic-field dependent correction, which goes as (Lc/La)2, and which we have ignored here for the sake of brevity.27 This corresponds to the regime, in which we have two independent diffusion modes, given by Eq. (58), corresponding to independent diffusion of the electric and the chiral charges. On the other hand, when La (cid:28) Lc, or x (cid:29) y, we obtain F (x, y) ≈ 1 y2/2x + tanh(x/2) . (65) This exhibits a regime of quasiballistic conductance with G(L) ≈ e2Nφ 2π , which is realized when La < L < L∗. (66) (67) This corresponds precisely to the range of the wavevec- tors q in Eq. (56), for which propagating modes exist when La (cid:28) Lc. Thus, one of the observable manifesta- tions of the existence of quasi-1D propagating modes in a TM is the quasiballistic conductance, given by Eq. (66). It is instructive to see what the quasiballistic con- ductance regime corresponds to directly in terms of the transport equations Eq. (59). In this regime both the sec- ond derivative D∇2n0,5 and the relaxation n5/τc terms may be ignored and we obtain ∂n0 ∂t ∂n5 ∂t = Γ = Γ ∂n5 ∂z ∂n0 ∂z , . (68) Introducing the left- and right-handed charges as nR,L = (n0 ± n5)/2 we obtain ∂nR ∂t ∂nL ∂t , ∂nR = Γ ∂z = −Γ ∂nL ∂z . (69) Eq. (69) describes two chiral bosonic density modes, which propagate along and opposite to the direction of the applied magnetic field. Such "bosonization" of the electron dynamics, which occurs in a 3D metal in a weak quasiclassical magnetic field, is a characteristic smoking- gun feature of a TM. Eq. (69) means, in particular, that a density distur- bance, created in a TM in magnetic field, with split into two chiral modes, which will propagate ballistically in opposite directions, spatially separating electrons of dif- ferent chirality. It might be possible to detect this effect optically.37 B. Optical conductivity Optical conductivity of TM has been studied before, with a focus mostly on the interband transition ef- fects.38 -- 42 Here we will demonstrate that low-frequency intraband optical conductivity is qualitatively affected by the chiral anomaly, which has not been noticed before. From the general expression for the density response function Eq. (40) we may easily obtain the frequency- dependent conductivity. Indeed, electric charge conser- vation requires that σzz(Ω) = −e2 lim q→0 iΩ q2 χ00(q, Ω). (70) A straightforward calculation then gives σzz(Ω) = σ 1 − iΩτ 1 − iΩτc + (Lc/La)2 1 − iΩτc , (71) where σ = e2gD is the zero-field DC conductivity. Eval- uating the real part, one obtains (cid:34) (cid:18) Lc (cid:19)2 1 − Ω2τ τc La 1 + Ω2τ 2 c (cid:35) Re σzz(Ω) = σ 1 + Ω2τ 2 1 + . (72) In this paper we have studied density response in TM and the corresponding experimentally observable phe- nomena. We have argued that one of the truly unique 8 FIG. 2. (Color online) Frequency-dependent conductivity for Lc/La = 1 (solid line) and Lc/La = 0 (dashed line), and τ /τc = 0.04. (cid:90) ∞ 0 Eq. (72) is one of the main new results of this paper. The prefactor in Eq. (72) is the standard Drude expres- sion for the optical conductivity of a metal. The part in the square brackets is a correction that arises in a TM as a consequence of the chiral anomaly. This correc- tion represents transfer of the spectral weight from high frequencies into a new low-frequency peak, whose width scales with the chiral charge relaxation rate 1/τc, while height is proportional to the ratio (Lc/La)2. Importantly, Eq. (72) satisfies the exact f -sum rule dΩ Re σzz(Ω) = πσ 2τ , (73) which means that the appearance of the new low- frequency peak indeed represents spectral weight trans- fer, as it should, see Fig. 2. It is instructive to examine the high-frequency limit of τ τc. In this limit √ Eq. (72), namely when Ω > 1/τc, 1/ we obtain Re σzz(Ω) ≈ σ 1 + Ω2τ 2 1 − 1 3 . (74) (cid:34) (cid:19)2(cid:35) (cid:18) (cid:96) La The negative second term in the square brackets ex- presses the reduction of the spectral weight at high fre- quencies, induced by the chiral anomaly. We note that while formally the whole expression may become negative for La (cid:28) (cid:96), this would be outside of the regime of validity of our theory, which assumes weak magnetic field regime kF (cid:96)B (cid:29) 1 and thus La (cid:29) (cid:96). Within this regime, the real part of the optical conductivity is always positive, as it should be. IV. DISCUSSION AND CONCLUSIONS 0.51.01.52.0Ωτ0.51.01.52.0Reσzz(Ω)/σ9 features of TM is the existence of propagating density modes, which are induced by the combined effect of the chiral anomaly and impurity scattering. The modes exist only in the diffusive limit and disappear in the ballistic regime. We have demonstrated that one of the observ- able manifestations of the existence of such propagating modes is the highly nontrivial scaling of the conductance of a TM with the sample size, first pointed out by Altland and Bagrets.36 We have also demonstrated an entirely new phenomenon, namely a nontrivial frequency depen- dence of the optical conductivity, which exhibits trans- fer of the spectral weight from high frequencies, greater than 1/ τ τc, into a new non-Drude low-frequency peak of width 1/τc. The existence of this new narrow peak in the optical conductivity is a smoking-gun consequence of the chiral anomaly in TM. √ One issue we have not touched upon in this pa- per is the effect of the electron-electron, in particular long-range Coulomb, interactions. One might worry that the Coulomb interactions could push the linearly- dispersing sound-like mode Eq. (55) to the plasma fre- quency, as happens in the case of the ordinary elec- tronic zero sound mode, if short-range interactions are replaced by Coulomb interactions. This does not happen in our case, however, since the existence of the sound- like mode has nothing to do with the electron-electron interactions. Its physical origin lies in the effective "one- dimensionalization" of the electron dynamics in a dirty TM in the presence of even a weak magnetic field. What this means is that the LLL dominates the density re- sponse at long times and long distances even when many higher Landau levels are occupied since the dynamics is ballistic in the LLL while it is diffusive in the higher Landau levels. This picture has nothing to do with the electron-electron interactions and will not be signifi- cantly modified by them, just as the ordinary low-energy particle-hole continuum in a clean Fermi liquid is not sig- nificantly affected by the interactions. The frequency of e2t2g is not significantly af- fected by a weak applied magnetic field43,44 and is much larger than the frequency of the low-energy chiral den- sity mode Ω0 = Γq, which arises within the low-energy particle-hole continuum of the clean metal. This means that the two modes do not interact with each other in any significant way. However, the issue of collective plas- mon modes in a dirty TM is interesting in its own right and will be addressed in a future publication. the plasmon modes ΩP ∼(cid:112) ACKNOWLEDGMENTS We thank Xi Dai for a useful discussion. Financial sup- port was provided by Natural Sciences and Engineering Research Council (NSERC) of Canada. 1 N. P. Armitage, E. J. Mele, and A. Vishwanath, Rev. (2014). Mod. Phys. 90, 015001 (2018). 2 M. Z. Hasan, S.-Y. Xu, I. Belopolski, and S.-M. Huang, Annual Review of Condensed Matter Physics 8 (2017). 3 B. Yan and C. Felser, Annual Review of Condensed Matter Physics 8 (2017). 4 A. A. Burkov, Annual Review of Condensed Matter Physics 9, 359 (2018). 5 X. Wan, A. M. Turner, A. Vishwanath, Savrasov, Phys. Rev. B 83, 205101 (2011). and S. Y. 6 A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). 7 A. A. Burkov, M. D. Hook, and L. Balents, Phys. Rev. B 84, 235126 (2011). 8 G. Xu, H. Weng, Z. Wang, X. Dai, and Z. Fang, Phys. Rev. Lett. 107, 186806 (2011). 9 S. M. Young, S. Zaheer, J. C. Y. Teo, C. L. Kane, E. J. Mele, and A. M. Rappe, Phys. Rev. Lett. 108, 140405 (2012). 10 Z. Wang, Y. Sun, X.-Q. Chen, C. Franchini, G. Xu, H. Weng, X. Dai, and Z. Fang, Phys. Rev. B 85, 195320 (2012). 11 Z. Wang, H. Weng, Q. Wu, X. Dai, and Z. Fang, Phys. Rev. B 88, 125427 (2013). 12 Z. K. Liu, B. Zhou, Y. Zhang, Z. J. Wang, H. M. Weng, D. Prabhakaran, S.-K. Mo, Z. X. Shen, Z. Fang, X. Dai, Z. Hussain, and Y. L. Chen, Science 343, 864 (2014). 13 M. Neupane, S.-Y. Xu, R. Sankar, N. Alidoust, G. Bian, C. Liu, I. Belopolski, T.-R. Chang, H.-T. Jeng, H. Lin, A. Bansil, F. Chou, and M. Z. Hasan, Nat. Commun. 5 14 S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.-C. Lee, S.- M. Huang, H. Zheng, J. Ma, D. S. Sanchez, B. Wang, A. Bansil, F. Chou, P. P. Shibayev, H. Lin, S. Jia, and M. Z. Hasan, Science 349, 613 (2015). 15 B. Q. Lv, N. Xu, H. M. Weng, J. Z. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, C. E. Matt, F. Bisti, V. N. Strocov, J. Mesot, Z. Fang, X. Dai, T. Qian, M. Shi, and H. Ding, Nat Phys 11, 724 (2015). 16 B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, Z. Fang, X. Dai, T. Qian, and H. Ding, Phys. Rev. X 5, 031013 (2015). 17 L. Lu, Z. Wang, D. Ye, L. Ran, L. Fu, J. D. Joannopoulos, and M. Soljaci´c, Science 349, 622 (2015). 18 E. Liu, Y. Sun, L. Muechler, A. Sun, L. Jiao, J. Kroder, V. Suss, H. Borrmann, W. Wang, W. Schnelle, S. Wirth, S. T. B. Goennenwein, and C. Felser, ArXiv e-prints (2017), arXiv:1712.06722 [cond-mat.mtrl-sci]. 19 G. Volovik, The Universe in a Helium Droplet (Oxford: Clarendon, 2003). 20 F. D. M. Haldane, Phys. Rev. Lett. 93, 206602 (2004). 21 G. E. Volovik, in Quantum Analogues: From Phase Tran- sitions to Black Holes and Cosmology, Lecture Notes in Physics, Vol. 718, edited by W. Unruh and R. Schtzhold (Springer Berlin Heidelberg, 2007). 22 S. Murakami, New Journal of Physics 9, 356 (2007). 23 A. A. Zyuzin and A. A. Burkov, Phys. Rev. B 86, 115133 (2012). 24 H. Nielsen and M. Ninomiya, Physics Letters B 130, 389 (1983). and N. P. Ong, Phys. Rev. X 8, 031002 (2018). 36 A. Altland and D. Bagrets, Phys. Rev. B 93, 075113 25 D. T. Son and B. Z. Spivak, Phys. Rev. B 88, 104412 (2016). 10 (2013). 26 A. A. Burkov, Phys. Rev. B 91, 245157 (2015). 27 A. A. Burkov, Phys. Rev. B 96, 041110 (2017). 28 S. Nandy, G. Sharma, A. Taraphder, and S. Tewari, Phys. Rev. Lett. 119, 176804 (2017). 29 A. A. Burkov, Phys. Rev. Lett. 113, 187202 (2014). 30 J. Xiong, S. K. Kushwaha, T. Liang, J. W. Krizan, M. Hirschberger, W. Wang, R. J. Cava, and N. P. Ong, Science 350, 413 (2015). 31 Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosic, A. V. Fedorov, R. D. Zhong, J. A. Schneeloch, G. D. Gu, and T. Valla, Nat Phys 12, 550 (2016). 32 H. Li, H. Wang, H. He, J. Wang, and S.-Q. Shen, ArXiv e-prints (2017), arXiv:1711.03671 [cond-mat.mes-hall]. 33 N. Kumar, C. Felser, and C. Shekhar, ArXiv e-prints (2017), arXiv:1711.04133 [cond-mat.mes-hall]. 34 Y. J. Wang, J. X. Gong, D. D. Liang, M. Ge, J. R. Wang, (2018), and C. J. Zhang, ArXiv e-prints W. K. Zhu, arXiv:1801.05929 [cond-mat.mtrl-sci]. 35 S. Liang, J. Lin, S. Kushwaha, J. Xing, N. Ni, R. J. Cava, 37 Q. Ma, S.-Y. Xu, C.-K. Chan, C.-L. Zhang, G. Chang, Y. Lin, W. Xie, T. Palacios, H. Lin, S. Jia, P. A. Lee, P. Jarillo-Herrero, and N. Gedik, Nature Physics 13, 842 EP (2017). 38 C. J. Tabert and J. P. Carbotte, Phys. Rev. B 93, 085442 (2016). 39 S. Ahn, E. J. Mele, and H. Min, Phys. Rev. B 95, 161112 (2017). 40 B. Roy, V. Jurici´c, and S. Das Sarma, Scientific Reports 6, 32446 EP (2016). 41 Y. Sun and A.-M. Wang, Phys. Rev. B 96, 085147 (2017). 42 D. Neubauer, A. Yaresko, W. Li, A. Lohle, R. Hubner, M. B. Schilling, C. Shekhar, C. Felser, M. Dressel, and A. V. Pronin, ArXiv e-prints (2018), arXiv:1803.09708 [cond-mat.mes-hall]. 43 I. Panfilov, A. A. Burkov, and D. A. Pesin, Phys. Rev. B 89, 245103 (2014). 44 J. Zhou, H.-R. Chang, and D. Xiao, Phys. Rev. B 91, 035114 (2015). 

1105.5209

2

1105

Gapless interface states between topological insulators with opposite Dirac velocities

The Dirac cone on a surface of a topological insulator shows linear dispersion analogous to optics and its velocity depends on materials. We consider a junction of two topological insulators with different velocities, and calculate the reflectance and transmittance. We find that they reflect the backscattering-free nature of the helical surface states. When the two velocities have opposite signs, both transmission and reflection are prohibited for normal incidence, when a mirror symmetry normal to the junction is preserved. In this case we show that there necessarily exist gapless states at the interface between the two topological insulators. Their existence is protected by mirror symmetry, and they have characteristic dispersions depending on the symmetry of the system.

cond-mat.mes-hall

cond-mat

Gapless interface states between topological insulators with opposite Dirac velocities Ryuji Takahashi1 and Shuichi Murakami1, 2 1Department of Physics, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro-ku, Tokyo 152-8551, Japan 2PRESTO, Japan Science and Technology Agency (JST), Kawaguchi, Saitama 332-0012, Japan (Dated: November 14, 2018) The Dirac cone on a surface of a topological insulator shows linear dispersion analogous to optics and its velocity depends on materials. We consider a junction of two topological insulators with different velocities, and calculate the reflectance and transmittance. We find that they reflect the backscattering-free nature of the helical surface states. When the two velocities have opposite signs, both transmission and reflection are prohibited for normal incidence, when a mirror symmetry normal to the junction is preserved. In this case we show that there necessarily exist gapless states at the interface between the two topological insulators. Their existence is protected by mirror symmetry, and they have characteristic dispersions depending on the symmetry of the system. PACS numbers: 73.20.-r, 73.40.-c,73.43.-f,75.70.Tj Recently physical phenomena originating from the Dirac cones of electrons have been studied, in the context of graphene sheet [1] or the topological insulator (TI) [2 -- 5]. In a graphene sheet, novel transport phenomena are predicted theoretically in p-n junction systems: for ex- ample the Klein paradox [6], and the negative refraction [7]. The TI in three dimensions (3D) [4, 5], such as Bi2Se3 [8, 9] and Bi2Te3 [10], has a single Dirac cone in its sur- face states, as observed by angle-resolved photoemission spectroscopy. Unlike graphene, the states on the Dirac cone on the surface of the TI are spin filtered; they have fixed spin directions for each wave number k. Because the state at k and that at −k have the opposite spins, the perfect backscattering from k to −k is forbidden. Such linear dispersion is similar to photons. The ve- locity of the Dirac cone on the surface of 3D TI depends on materials. For example, the velocity for Bi2Te3 is about 4 × 105m/s [10] depending on the direction of the wave vector, and that for Bi2Se3 is approximately 5 × 105 m/s [8]. Therefore, when two different TIs are attached together, the refraction phenomenon similar to optics is expected at the junction. In this Letter, we theoreti- cally study the refraction of electrons at the junction be- tween the surfaces of two TIs [Fig. 1(a)]. The resulting transmittance and reflectance are different from optics, reflecting prohibited perfect backscattering. In addition, we show that when the velocities of the two TIs have op- posite signs, neither refraction nor reflection is allowed for the incident electron normal to the junction. In this case, we can show that there necessarily exist gapless interface states between the two TIs and the incident surface elec- trons totally go into the interface states. As long as the mirror symmetry with respect to the yz plane Myz is preserved, the interface gapless states exist. These gap- less states are formed at the interface between the same Z2 nontrivial materials. As a result, these interface states do not come from the Z2 topological number, but come from the mirror Chern number[11], and are protected by the mirror symmetry Myz. The effective Dirac Hamiltonian of the surface states on the xz plane is represented as H = −iv[σx∂z − σy∂x], (1) where σx, σy are the Pauli matrices, and v is the Fermi velocity. From the Hamiltonian one can obtain the linear energy E = svk where k = k, and s = +1(−1) corresponds to the upper (lower) cone, provided v > 0. We consider a refraction problem between the two TIs, which we call TI1 and TI2, with the incidence angle θ, the transmission angle θ′, and the reflection angle θR [Fig. 1(a)]. As in optics, the momentum conservation requires θR = θ, and the wave functions are written as ψI (x, z) = 1√2 eik(x sin θ+z cos θ)(1, e−iθ)t, ψT (x, z) = 1√2 = 1√2 are the wave numbers on TI1 and TI2, respectively, and we consider the Fermi energy EF > 0 (i.e above the Dirac point), giving s = +1 for both of the TIs. Let v1 and v2 denote the velocities of the two TIs. eik′(x sin θ ′+z cos θ ′)(1, e−iθ ′ eik(x sin θ−z cos θ)(1, −eiθ)t, where and k′ ψR(x, z) )t, k 2 1 sin θ = v−1 We first assume v1 and v2 to be positive. The con- servation of the momentum and the energy yields Snell's law: k′ sin θ′ = k sin θ, v−1 sin θ′. Let r and t denote the amplitude of the reflected and transmitted wave, compared with the incident wave. The current conservation in the z direction is written as R + T = 1, where R ≡ r2 and T ≡ v2 cos θ ′ v1 cos θ t2 are the reflectance and the transmittance, respectively. We note that the wavefunction should eventually be discontinuous at the junction when the velocities are different, as has been studied in the context of graphene [12, 13]. The reason is the following. Therefore, the current conservation at the interface requires v1zψ12 = v2zψ22, where v1,2 is a velocity, and the subscripts 1 and 2 represent TI1 and TI2, respectively. Because in our case v1z 6= v2z, we have ψ12 6= ψ22 at the junction, and the continuity 2 FIG. 2: (Color online) Transport at the junction between the surfaces of two TIs, whose velocities have different signs. (a) Linear dispersion at kx = 0. The incident wave (I) is perpendicular to the junction. Both the transmission (T) and reflection (R) are prohibited due to spin conservation. (b) Normal incidence. TI1 (red) and TI2 (blue) have the velocities of opposite signs. The purple region represents the interface. two TIs. These interface states arise from hybridization between the two surface states from the two TIs. To show the existence of gapless interface states, we first write down the effective Hamiltonian at the interface from the two Dirac cones with hybridization: H = (cid:18)H1 V V † H2(cid:19) . (3) FIG. 1: (Color online) (a) Schematic of the refraction of the surface states at the junction between the two TIs, TI1 and TI2. (b)(c): Reflectance (red) and transmittance (blue) for the ratios of the velocities of the two TIs: (b)v2/v1 = 0.6 and (c)v2/v1 = 1.4. The solid curves are the results for the junction between two TIs, while the dotted curves show the results for optics with p and s polarizations. of the wavefunction is violated. The proper way is to set the Hamiltonian to be Hermitian also at the bound- Here H1(2) is the effective surface Hamiltonian for the surface of TI1 (TI2) at the interface: ary, i.e. H = −i(cid:2) 1 where v(z) is the velocity dependent on z. The resulting coefficients are 2 [v(z)σx∂z + σx∂zv(z)] − v(z)σy∂x(cid:3) , r = i sin θ ′−θ cos θ+θ ′ 2 2 e−iθ, t = r v1 v2 cos θ cos θ+θ ′ 2 ei θ′ −θ 2 . (2) They satisfy the current conservation. The results are plotted as the solid curves in Figs. 1(b)(c). The dotted curves represent corresponding results for optics. Unlike optics, for normal incidence (θ = 0), the perfect transmis- sion (T = 1, R = 0) occurs, which reflects the prohibited backscattering on the surface of the TI. This is similar to graphene [6, 12, 13] but the transmittance in our case monotonically decreases with the incidence angle. Next, we consider the case where the velocities of the two TIs have opposite signs, where we can no longer use the above approach. One might think that it is similar to the negative refraction in optics [14, 15], but it is not true because the Fermi energy is above the Dirac point for the two TIs. Furthermore, both reflection and transmission are prohibited for normal incidence, because the incident wave has no way to conserve its momentum kx along the interface and spin simultaneously (see Fig. 2). Thus it is a paradox what happens for normal incidence. Our answer to this question is that gapless states exist at the interface between the two TIs (the purple region in Fig. 2(b)). The normally incident wave goes along the surface of one TI, then into the interface between the H1 = v1(σ × k)z, H2 = −v2(σ × k)z (4) and V is the hybridization at the interface. For simplic- ity, we retain only the lowest order in k. In the expression of H2, there is an extra minus sign; on the surface of TI2 in Fig. 2, the mode going in the +z direction evolves from that going in the −y direction, whereby the extra sign necessarily appears. We explain the reason for justifying our model in Eqs. (3), (4). For simplicity we assumed that the surface states on the xz surface for TI1 and TI2 are described by the Dirac cone. Generic surface states with non-Dirac types are covered in the later discussion using the mirror Chern number[11]. We used here the fact that the Dirac velocities for each TI have the same signs for the xy and xz surfaces. It is because the signs of the Dirac veloci- ties are determined by the mirror Chern number which is the bulk quantity [11]. We also set the Dirac cones to be isotropic for simplicity; the following results turn out to be unaltered by anisotropy in the Dirac cones. We henceforth impose the mirror symmetry with respect to the yz plane Myz, because this symmetry preserved by H1 and H2 sets the spins parallel to the x axis for the nor- mally (kz) incident wave. By imposing this mirror sym- metry Myz and time-reversal symmetry, V is expressed as V = (cid:18) g ih ih g(cid:19) , where g and h are real constants repre- senting the hybridization between the two surface states. 3 rotational symmetry around the z axis, and is lifted when it is broken by adding higher order terms in k, e.g. the warping term in Bi2Se3 [16]. As seen in Fig. 3(b), the dispersion becomes a collection of Dirac cones. There- fore, for this example Hamiltonian, we could show that there are gapless states at the interface when the system has the mirror symmetry Myz. We note that this method is generic, because the anal- ysis is based only on the symmetry. The only assumption is that the gapless point is near k = 0, and we can expand the Hamiltonian in terms of k. To complement this argu- ment, we show the existence of gapless interface states on generic grounds. Because these gapless states are gener- ated between two TIs with the same Z2 topological num- bers, they are not protected in the same sense as the surface states of three-dimensional TIs. In the following we show that these gapless interface states are protected by the mirror symmetry and the time-reversal symme- try. Each TI with mirror symmetry is characterized by the mirror Chern number [11]. When the system has the mirror symmetry Myz, the surface modes are labeled with the mirror eigenvalues M = ±i at kx = 0, corre- sponding to the spin along −x and +x, respectively. The mirror Chern number is obtained as nM = (n+i − n−i)/2 where n±i are the Chern numbers [17, 18] for the sub- space of states with mirror eigenvalues M = ±i. We have n+i = −n−i by the time-reversal symmetry. In our case where the two surface Dirac cones have opposite ve- locities, the mirror Chern numbers for the two TIs are different. TI1 has n(1) −i = +1, M and TI2 has n(2) −i = −1 at kx = 0 M plane. For the M = +i (Sx < 0) subspace, this corre- sponds to the junction of two systems with Chern num- bers n(1) +i = −2, it gives rise to two left-going chiral modes in the y di- rection. On the other hand, for M = −i it also gives two right-going chiral modes in the y direction. These modes are schematically shown in Fig. 3(d). Therefore it is natural to generate the two Dirac cones in the junc- tion. Thus these gapless states are protected by the mir- ror symmetry. If the mirror symmetry is not preserved, the gapless states do not exist in general. This discussion is generic, and is complementary to our discussion by the surface Dirac Hamiltonian. Therefore, we conclude that the gapless interface states exist for the generic cases with mirror symmetry, even with e.g. lattice mismatch at the interface. In real materials, mirror symmetry may be lost by disorder in principle; nevertheless, if the the sample is relatively clean, the gapless interface states are expected to survive and can be measured experimentally. +i = +1; because n(1) +i = −1 and n(2) +i = −1, n(1) +i = +1, n(2) = −1, i.e. n(1) = 1, i.e. n(2) +i − n(2) The distance between the Dirac cones of the gapless interface states are proportional to the magnitude of the hybridization between the two TIs at the interface. When the hybridization becomes as strong as the bandwidth, the spacing between the interface Dirac cones is of the FIG. 3: (Color online) (a)(b) Dispersion on the interface be- tween the two TIs in Eq. (5) with velocities v1 = 1, v2 = −2. In (a), the hybridization is g = 2, h = 1. There are two Dirac points (kx, ky) = (0, ±p5/2) where the gap closes. In (b), the hybridization is g = 2, h = 0 and the warping term λ(k3 −)σz with λ = 0.4 added to H1 and H2. There ap- pear six Dirac cones. (c) Illustration of the interface mode. The surface current goes into the interface. (d) Schematic of the dispersion of interface states on kx = 0. + + k3 From the Hamiltonian [Eq. (3)], the eigenvalues are cal- culated as E = ±r∆k ±q∆2 k − η, ∆k = g2 + h2 + 1 + v2 v2 2 2 k2,(5) (6) η = v2 1v2 2k4 + ∆2 0 − 2v1v2k2(h2 cos 2α − g2), where we set kx+iky = keiα, and α is real. The condition for existence of gapless interface states is v2 1v2 2k4 − 2v1v2k2(h2 cos 2α − g2) + (g2 + h2)2 = 0. (7) To solve this equation, we note that g is nonzero, whereas h can become zero when additional symmetries such as rotational symmetry with respect to the z axis are im- posed. Then we can see that for v1v2 > 0 (the two veloc- ities with the same signs), the interface states are gapped by the hybridization. Only when two velocities have opposite signs (v1v2 < 0), are there gapless states on the interface. Disper- sion of the gapless states depends on whether h 6= 0 or h = 0. When h 6= 0, the solutions are (kx, ky) = (cid:16)0, ±p(g2 + h2)/v1v2(cid:17) and there are gapless states on the interface. The interface states have two Dirac cones (Fig. 3 (a)). On the other hand, when h = 0 due to rotational symmetry with respect to the z axis, the gap closing points form a circle k2 . This de- generacy on the circle in k space is due to the continuous y = g2 v1v2 x + k2 order of inverse of the lattice spacing. In that case the transport properties will be like the graphene, having two Dirac cones at K and K' points. We note that in graphene there are spin-degenerate Dirac cones, whereas in the present case the interface Dirac cones are not spin degenerate. From Fig. 3(d), when the wave number k goes around one of the Dirac point, the spin direction also rotates around the z axis (normal to the interface). In the similar way as in graphene, one can consider the valley degree of freedom as a pseudospin, and develop val- leytronics [19, 20] similar to graphene. These interface states can be measured via transport; for this purpose one should suppress the surface transport by attaching ferromagnets on the surface. From the spin-resolved angle-resolved photoemission spectra, all the TIs observed so far, such as Bi1−xSbx [21, 22], Bi2Se3[9], and Bi2Te3[9], have nM = −1. To realize the protected interface states in experiments discussed in this Letter, one needs to find a TI with nM = +1, i.e., the surface Dirac cone with negative velocity, and the spins on the upper cone is in the counterclockwise direction in the k space. It is an interesting issue to search for such TIs. The Dirac velocity v is nothing but the coefficient λ in the Rashba spin-splitting term λ(σ × k)z in the Hamiltonian. The Rashba coefficient λ originates from an integral of a sharply peaked function near the nuclei, which rapidly varies between positive and negative values [23, 24]. Therefore, we expect that it can change sign in principle. The sign of the mirror Chern number nM is also related with the mirror chirality of the bulk Dirac Hamiltonian describing the bands near the bulk gap [11]. Because the mirror chirality governs the sign of the g- factor which can be negative or positive as a result of the spin-orbit coupling, one may well expect that in some materials nM can become +1. In conclusion, we study refraction phenomena on the junction between the two TI surfaces with different veloc- ities. The resulting reflectance and transmittance reflect the backscattering-free nature of the surface states of TIs. When the velocities of the TI surface states for the two TIs have different signs, we show that the gapless states appear on the interface. The existence of the gapless states is shown by using the mirror Chern number, and thus is topologically protected by the mirror symmetry. 4 The authors are grateful to T. Oguchi for discus- sions. This research is supported in part by Grant-in- Aid for Scientific Research (No. 21000004 and 22540327) from the MEXT, Japan and by Kurata Grant from Ku- rata Memorial Hitachi Science and Technology Founda- tion. R.T also acknowledges the financial support from the Global Center of Excellence Program by MEXT, Japan through the "Nanoscience and Quantum Physics" Project of Tokyo Institute of Technology. [1] K. S. Novoselov et al., Science 306, 666 (2004). [2] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005); ibid. 95, 146802 (2005). [3] B. A. Bernevig and S.-C. Zhang, Phys. Rev. Lett. 96 (2006) 106802. [4] L. Fu, C. L. Kane and E. J. Mele, Phys. Rev. Lett. 98,106803(2007). [5] J. E. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2007). [6] M. I. Katsnelson et al., Nature Physics 2, 620 - 625 (2006). [7] V. V. Cheianov et al., Science 315, 1252 (2007). [8] Y. Xia et al., Nature Phys. 5, 398 (2009). [9] D. Hsieh et al., Nature (London) 460, 1101 (2009). [10] Y. L. Chen et al., Science 325, 178 (2009). [11] J. C. Y. Teo, L. Fu, C. L. Kane, Phys. Rev. B 78, 045426 (2008). [12] A. Raoux, M. Polini, R. Asgari, A. R. Hamilton, R. Fazio, and A. H. MacDonald, Phys. Rev. B81, 073407 (2010) [13] A. Concha and Z. Tesanovi´c Phys. Rev. B82, 033413 (2010) [14] V. G. Veselago, Sov. Phys. Usp. 10, 509 (1968). [15] J. B. Pendry, Phys. Rev. Lett. 85, 3966 (2000). [16] L. Fu, Phys. Rev. Lett. 103, 266801 (2009) [17] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [18] M. Kohmoto, Ann. Phys. N.Y. 160, 343 (1985). [19] O. Gunawan et al., Phys. Rev. Lett. 97, 186404 (2006). [20] D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 99, 236809 (2007). [21] D. Hsieh et al., Science 323, 919 (2009). [22] A. Nishide et al., Phys. Rev. B 81, 041309(R) (2010). [23] M. Nagano, A. Kodama, T. Shishidou and T. Oguchi, J. Phys.: Conden. Matter 21, 064239 (2009). [24] G. Bihlmayer et al., Surf. Sci. 600, 3888 (2006). 

1608.06436

1

1608

Dynamic dependence to domain wall propagation through artificial spin ice

Domain wall propagation dynamics have been studied in nanostructured artificial kagome spin ice structures. A stripline circuit has been used to provide localised pulsed magnetic fields within the artificial spin ice structure. This provides control of the system through electrically assisted domain wall nucleation events. Synchronisation of the pulsed fields with additional global magnetic fields and the use of a focussed magneto-optical Kerr effect magnetometer allows our experiments to probe the domain wall transit through an extended ASI structure. We find that the propagation distance depends on the driving field revealing field driven properties of domain walls below their intrinsic nucleation field.

cond-mat.mes-hall

cond-mat

a Dynamic dependence to domain wall propagation through artificial spin ice Department of Physics, Imperial College London, London SW7 2BZ, United Kingdom D.M. Burn, M. Chadha, and W.R. Branford (Dated: November 10, 2018) Domain wall propagation dynamics have been studied in nanostructured artificial kagome spin ice structures. A stripline circuit has been used to provide localised pulsed magnetic fields within the artificial spin ice structure. This provides control of the system through electrically assisted domain wall nucleation events. Synchronisation of the pulsed fields with additional global magnetic fields and the use of a focussed magneto-optical Kerr effect magnetometer allows our experiments to probe the domain wall transit through an extended ASI structure. We find that the propagation distance depends on the driving field revealing field driven properties of domain walls below their intrinsic nucleation field. Magnetic meta-materials such as artificial spin ice show behaviour arising from complex geometrical struc- turing in addition to the original material properties.1,2 In these systems it is the combination of magnetic charge interactions and topological constraints determine the magnetisation behaviour of the system. Artificial spin ice structures consisting of arrays of magnetic nano- bars provide a 2D analogue to explore frustrated mag- netic phenomena.3,4 These systems are of fundamen- tal scientific interest5 -- 12 and have even been identified as potentials for novel neural network or processing technologies.3,13,14 Magnetisation reversal in artificial spin ice structures composed of interconnected magnetic bars can be de- scribed by an ensemble of magnetic domain wall (DW) processes. The creation, annihilation and propagation of these DWs throughout the system leads to magneti- sation reversal within the bars as well as the transport of both magnetic and topological charges throughout the system. The conservation of both magnetic and topo- logical charge provides constraints on the creation and annihilation of DWs in the system. This reveals the physical significance of the finer details of the micromag- netic DW structure such as its chirality or topological makeup when the DW interacts with a complex magnetic structure.8,9,15 -- 19 The majority of our understanding of the magnetisa- tion behaviour in artificial spin ice systems is based on ex- periments combining thermal and quasi-static magnetic fields applied to the entire system.10,12,20 -- 22 The role of DWs have been typically investigated based on their nat- ural occurrence,23 when an applied field exceeds the nu- cleation field which is typically lower at the edges of the structures. This approach is therefore limited in that we can only investigate the internal behaviour of the system once a process related to the edge of the system takes place. In this investigation the localised injection of DWs along the length of a lithographically patterned mi- crostrip is employed.24,25 Here the pulsed field DW in- jection technique allows control over the DW nucleation location within the system leading to significant experi- mental advantages as the DW nucleation process can be separated from a global applied field. Firstly, this allows the behaviour of DWs in the system to be investigated in a wider field range, even at lower fields than their nu- cleation field. Secondly, this allows the magnetisation dynamics in the system to be explored. This is of great interest in the artificial spin ice system and understand- ing of the behaviour in a dynamic context is necessary for any future technological applications. Our understanding of the propagation path of a DW through a series of vertex structures can be explained through topological considerations. Figure 1 shows a two-vertex section of an artificial spin ice structure with arrows representing the magnetisation orientation in each bar and with topological defects pinned to the edges of the structure associated with both the DWs and the ver- tices. Figure 1 shows the evolution of a down-chirality DW incident upon a vertex with initial magnetisation sat- urated to the left. During the interaction the −1/2 topological defect initially belonging to the DW becomes pinned on the upper edge of the vertex. The +1/2 from the DW follows the edge of the structuring and pairs with the −1/2 initially associated with the vertex. This new defect pair corresponds to a DW which is able propagate along the lower branch at the vertex. The similar +ve magnetic charges of the initial DW and the vertex pro- vide a repulsive force which means there is an energy bar- rier associated with this process which can be overcome through the application of an applied magnetic field. In figure 1(b) the -ve charge of the second vertex now provides an attractive force on the positively charged DW. Here the topological charges on the lower side of the nanobar are opposite and therefore unwind when they FIG. 1. Simple model of DW progation and annihilation based on topological constraints in multiple vertices. meet. The DW annihilates resulting in the two-in one-out state illustrated in figure 1(c) where just a −1/2 topo- logical defect from the incident DW now remains pinned at the vertex. In this model, the DW no longer exists in the system. Figure 1(d) shows how a new DW can be injected into the lower horizontal nanobar based on the state shown in figure 1(c). The lower edge of the vertex contains zero topological charge which is separated to form two defects of +1/2 and −1/2 respectively. The +1/2 forms a pair with the pre-existing −1/2 defect on the upper edge of the vertex and represents a DW which can propagate along the nanobar whilst the remaining −1/2 defect re- mains at the lower edge of the vertex. This process in- volves the separation of two opposite magnetic charges which gives rise to an energy barrier which needs to be overcome to complete this process. The series of interactions illustrated in figure 1 shows how an initial DW with down-chirality can interact with two verticies in an artificial spin ice structure resulting in a down-chirality DW in a subsequent nanobar with similar geometry. The repeat of this process is consis- tent with reversal events following from one another. A similar series of interactions would also take place for up-chirality DW which would propagate along the upper branch at the first vertex. This would be followed by an equivalent annihilation event at the second vertex and the availability to nucleate a further up-chirality DW in the final horizontal nanobar. In all cases, a sizable mag- netic field is required to supply the energy to overcome the energy barriers associated with moving like charges towards one another, and separating zero charge into a positive and negative charge pair. This current understanding is based on the system which maintains a minimum energy micromagnetic spin configuration. In this quasi-static regime energy must be supplied to overcome energy barriers associated with the transitions which nucleate the DW in new nanobars throughout the system. In this work we consider the higher energy processes involved with dynamic propagat- ing DWs and show deviations from our understanding of these processes in the quasi-static regime. By varying the bias field that drives DW motion we investigate the importance of DW dynamics during prop- agation through an artificial spin ice structure with fields applied along the armchair geometry. Combining lo- calised DW injection with a MOKE measurements with a localised magnetisation probe we also infer on the length- scales of propagation through the system at these fields. Critically, our experiments probe the DW propagation behaviour in fields below the intrinsic DW nucleation field for these structures. Artificial spin ice structures consisting of intercon- nected NiFe nanowires were fabricated in a kagome geom- etry using electron-beam lithography and thermal evapo- ration. The bars were 700 nm x 150 nm in dimension and were 10 nm thick. Further details about the patterning of the magnetic structures can be found elsewhere26. A 2 2 µm wide Cr(5 nm)/Au(50 nm) microstrip was added in a second lithographic process and is shown in figure 2 along with the simulated field profile expected from the microstrip.27 The magnetisation reversal in the system was investi- gated using magneto-optical Kerr effect (MOKE) magne- tometry in the longitudinal geometry. Here, a focussed laser spot with a ∼ 10 µm elongated footprint provided a localised probe of the magnetisation reversal as illus- trated in figure 2. The combination of quasi-static and pulsed magnetic fields, supplied from external coils and from the microstrip respectively, were used to drive the magnetisation reversal in the sample. The Kerr signal was averaged over 50 field cycles in each measurement. By introducing a 5◦ angular offset between ASI struc- turing and the applied field direction the magnetisation reversal associated with DW nucleation events and DW propagation through the system can be distinguished by the reversal field.26 Additionally, by varying the contri- bution from the quasi-static and pulsed fields, behaviour from quasi-static energy dependent magnetisation re- versal processes and time-dependent magnetisation pro- cesses have been investigated.24,28 Initially the magnetisation behaviour was investigated with a 1 Hz sinusoidal quasi-static applied magnetic field and is shown in figure 3(a). Two transitions in the mag- netisation occur at two distinct and relatively sharp re- versal fields despite averaging over 50 field cycles and over multiple nanowires within the illuminated laser footprint. The two steps indicate the combination of two reversal processes occurring during the magnetisation reversal as- sociated with the nucleation field of a DW from a vertex and the field required for a pre-existing DW to propagate through a vertex.26 By introducing additional pulsed magnetic fields, fig- ures 3(b) and (c) show modified hysteresis loops where the arrows indicate the triggering of the pulsed field FIG. 2. Image of a interconnected array of nanowires with a Au stripline for applying localised pulsed magnetic fields. 700 nm x 150 nm within the quasi-static field cycle resulting in an increase in the magnetisation. At this point, the combination of pulsed and quasi-static fields locally overcome the rever- sal field leading to the injection of magnetic DWs at the stripline. The additional pulsed field induced reversal results in a reduced magnetisation reversal at the lower quasi-static field and the higher field quasi-static reversal remains unchanged. The magnetisation reversal combines multiple rever- sal processes which can be modelled as a summation of tanh functions.29 The lines in figure 3 show a model fit to the data where each transition is parameterised by a reversal field, the relative change in magnetisation and a parameter representing the shape of that transition. The quasi-static reversal fields are symmetric with increasing and decreasing field and share fitting parameters. The shape of the hysteresis loop representing the in- crease in magnetisation at the lower reversal field also shows some broadening when a pulsed field is present. This could represent a modified reversal field due to a partially-reversed magnetisation state following the pulse. This will be discussed in more detail later. The pulsed-field induced magnetisation reversal de- pends on both the pulsed field voltage and the trigger- ing point within the quasi-static field cycle. Figure 4(a) shows the minimum pulse voltage required to result in the additional pulsed-field induced magnetisation rever- sal steps in figure 3 (b and c). This is plotted as a function of the quasi-static field at the point of pulsed field triggering which can be considered as a static bias field on the timescales of the pulsed field. The linear de- crease represents the contributions to the total field at the stripline where an increase in pulsed field amplitude from a greater pulsed voltage allows magnetisation rever- sal to take place at a lower quasi-static field. A linear fit 3 to this data provides a calibration of the stripline which produces 10.8 ± 0.3 Oe/V. For quasi-static fields greater than 122 Oe the magnetisation reversal is driven purely by the quasi-static field. Therefore the effect of the pulsed field in this field regime cannot be distinguished. Figure 4(a) also compares the difference in behaviour when the laser spot is positioned 0, 10 and 20 µm away from the stripline. All the points fall on the same line indicating the reversal process at the stripline is not af- fected by the measurement position. However, when measurements are performed with greater separation, magnetisation reversal is only observed when the pulsed field triggering occurs at larger quasi-static fields. This feature allows us to probe the motion of the DWs through an extended region of the ASI system. For all measurement positions the combination of pulsed and quasi-static fields still result in magnetisation reversal at the stripline. This reversal is due to the lo- caslied injection of magnetic DWs at the stripline which propagate along the nanobars reversing the magnetisa- tion near the stripline. At greater distances from the stripline only the quasi-static field drives the DW propa- gation and the magnetisation reversal represents the dy- namic behaviour of DWs in fields below their nucleation field. Pulse lengths of 150 ns and 20 ns had little influence on the DW injection process so the quasi-static bias field dependence on DW propagation has been further investi- gated with 20 ns long, 7.5 V pulses which are sufficient to inject DWs when biased with a field greater than 108 Oe. Figure 4(b) shows the pulsed-field-induced magnetisation reversal contribution as a function of the quasi-static bias field for various measurement positions. This is found from the ratio in magnetisation change from the pulsed and quasi-static fields in the hysteresis loops. n o i t a s i t e n g a m d e s i l a m r o N (a) (b) (c) 1 0 -1 1 0 -1 1 0 -1 -250 -200 -150 -100 0 -50 50 Applied field (Oe) 100 150 200 250 ) V ( d e l i f d e s u P l 9 8 7 6 d e c u d n i l d e i f d e s u P l ) % ( l a s r e v e r n o i t a s i t e n g a m 1.0 0.5 0.0 100 Distance to stripline 0 m 5 m 10 m 20 m Distance to stripline 0 m 5 m 10 m 20 m 105 110 115 120 Quasi-static field (Oe) FIG. 3. MOKE hysteresis loops showing the magnetisation reversal driven by (a) a quasi-static magnetic field and (b) and (c) with the addition of a pulsed magnetic field with triggering indicated by the arrows. The loops show the behaviour 9V 20ns pulses with the laser at the stripline. The fitted lines show a model fit to the data. FIG. 4. a) Minimum pulsed voltage required to lead to mag- netisation reversal when pulses are triggered at different QS Bias fields. b)Pulsed field induced magnetisation reversal as a function of quasi-static bias field measured at various sepa- rations from the DW injection stripline. Pulses are 20 ns and 7.5 V i ( s a b c i t ) e O a t s - i s a u Q 120 115 110 105 100 -20 Pulsed field induced magnetisation reversal (%) 0 10 20 30 40 50 60 70 80 -10 0 10 20 Laser position ( m) FIG. 5. Pulsed field induced magnetisation reversal measured as a function of quasi-static bias field and measurement posi- tion from the stripline. 20 ns Pulses at 7.5 V Measurements at the stripline location show a large pulsed-field induced magnetisation change which is most significant when a large quasi-static field is used to drive the DW propagation. Here, the result represents DWs which reverse the magnetisation in nanobars near where they are nucleated. When the quasi-static bias field is reduced below 115 Oe the magnetisation change decreases. This indi- cates that the proportion of pulsed field induced reversal events within the region probed by the laser spot is re- duced. This can be explained by DWs which are not able to propagate so far through the structure in the lower fields. Measurements at greater distances from the stripline show magnetisation reversal driven purely by the quasi- static field (see field profiles in figure 2). Here a lower magnetisation change is found as injected DWs must propagate through a greater number of nanobars and ver- tices before reaching the probed region. This means that DW pinning is more likely, but at high quasi-static bias fields, DW propagation up to 15 µm is still observed. The DW propagation distance through the structure is more clearly shown in in figure 5 where the pulsed-field- induced magnetisation reversal is plotted as a function of the measurement position and the quasi-static field. Here a strong reversal is centered around the position of the stripline at 0 µm which becomes more significant with greater quasi-static fields. Again, as the fields approach 120 Oe the pulsed-field-induced reversal becomes indis- tinguishable from the quasi-static field driven reversal. With the large quasi-static fields, the distance over 4 which the reversal can be detected is much greater than for the smaller quasi-static fields resulting in a triangular shape in figure 5. This shows how the propagation of a DW through the artificial spin ice structure depends on the driving field applied to the DW. With low fields the propagation distance is limited as multiple nanobars and interconnecting verticies are encountered and pro- vide pinning sites. However, with greater fields the pin- ning from these become less significant allowing the DW propagation over a greater number of nanobars and ver- tices. The simplified model of the DW path illustrated in figure 1 relies on the external driving field exceeding the nucleation field for a DW in these structures. However, our results reveal that DWs are able to propagate below this field with a driving field dependence to their propa- gation length through the system. The propagation at a reduced field can be explained by considering the energetics associated with a propagating DW. When considering the annihilation process between the DW and vertex in figure 1(b) and (c) the energy associated with the DW can be used to assist with the nucleation of the DW in figure 1(d). This additional energy would mean less is required externally from the driving field. The propagation length dependence can also be ex- plained in terms of the energetics of the dynamically propagating DWs. With a greater quasi-static driving field the DWs travel with a greater energy. This means that they can encounter, and overcome a greater number energy barriers associated with the verticies before be- coming pinned. The lower energy DWs become pinned after fewer verticies and therefore travel a reduced dis- tance through the artificial spin ice structuring. In conclusion, we have probed the magnetisation rever- sal in artificial spin ice systems through focussed MOKE magnetometry. The combination of pulsed and quasi- static magnetic fields allow for the injection of DWs and the study of their dynamic interactions with the geo- metrical structuring for a range of DW driving fields. Our results demonstrate control over the location of in- jected DWs within artificial spin ice structures and how DW propagation distance depends on the external driv- ing field. Existing quasi-static models based on the manipula- tion of magnetic and topological charges throughout the system do not predict a length dependence to the propa- gation. Our results, probing DWs in the dynamic regime with a range of driving fields suggest changes for DWs arriving at vertices in a higher energy state. 1 C. Nisoli, R. Moessner and P. Schiffer. Reviews of Modern Condensed Matter 25, 363201 (2013). Physics 85, 1473 -- 1490 (2013). 2 L.J. Heyderman and R.L. Stamps. Journal of Physics: 3 R.F. Wang, C. Nisoli, R.S. Freitas, J. Li, W. McConville, B.J. Cooley, M.S. Lund, N. Samarth, C. Leighton, V.H. Crespi and P. Schiffer. Nature 439, 303 (2006). 4 E. Mengotti, L.J. Heyderman, A.F. Rodr´ıguez, F. Nolting, R.V. Hugli and H.B. Braun. Nature Physics 7, 68 (2010). 5 S. Ladak, D.E. Read, G.K. Perkins, L.F. Cohen and W.R. Branford. Nature Physics 6, 359 (2010). 6 R.V. Hugli, G. Duff, B. O'Conchuir, E. Mengotti, A.F. Rodr´ıguez, F. Nolting, L.J. Heyderman and H.B. Braun. Philosophical transactions A 370, 5767 (2012). 7 A. Farhan, P.M. Derlet, A. Kleibert, A. Balan, R.V. Chopdekar, M. Wyss, J. Perron, A. Scholl, F. Nolting and L.J. Heyderman. Physical Review Letters 111, 057204 (2013). 8 W.R. Branford, S. Ladak, D.E. Read, K. Zeissler and L.F. Cohen. Science 335, 1597 -- 600 (2012). 9 N. Rougemaille, F. Montaigne, B. Canals, M. Hehn, H. Ri- ahi, D. Lacour and J.C. Toussaint. New Journal of Physics 15, 10 (2013). 10 Yi Qi, T. Brintlinger and J. Cumings. Physical Review B 77, 094418 (2008). 11 S. Zhang, I. Gilbert, C. Nisoli, G-W. Chern, M.J. Erickson, L. O'Brien, C. Leighton, P.E. Lammert, V.H. Crespi and P. Schiffer. Nature 500, 553 (2013). 12 U.B. Arnalds, A. Farhan, R.V. Chopdekar, V. Kapaklis, A. Balan, E.T. Papaioannou, M. Ahlberg, F. Nolting, L.J. Heyderman and B. Hjorvarsson. Applied Physics Letters 101, 112404 (2012). 13 S. Ladak, D.E. Read, T. Tyliszczak, W.R. Branford and L.F. Cohen. New Journal of Physics 13, 023023 (2011). 14 P.E. Lammert, Xianglin Ke, Jie Li, C. Nisoli, David M. Garand, Vincent H. Crespi and P. Schiffer. Nature Physics 6, 786 -- 789 (2010). 15 S.A. Daunheimer, O. Petrova, O. Tchernyshyov and J. Cumings. Physical Review Letters 107, 167201 (2011). 16 O. Tchernyshyov and G-W. Chern. Physical Review Letters 5 95, 197204 (2005). 17 A. Pushp, T. Phung, C. Rettner, B. P. Hughes, S-H. Yang, L. Thomas and S.S.P. Parkin. Nature Physics 9, 505 (2013). 18 S.K. Walton, K. Zeissler, D.M. Burn, S. Ladak, D.E. Read, T. Tyliszczak, L.F. Cohen and W.R. Branford. New Jour- nal of Physics 17, 013054 (2015). 19 P. Mellado, O. Petrova, Y. Shen and O. Tchernyshyov. Physical Review Letters 105, 187206 (2010). 20 F Montaigne, D. Lacour, I.A. Chioar, N. Rougemaille, D Louis, S Mc Murtry, H. Riahi, B.S. Burgos, T.O. Mente, A. Locatelli, B. Canals and M. Hehn. Scientific reports 4, 5702 (2014). 21 Z. Budrikis, J.P. Morgan, J. Akerman, A. Stein, P. Politi, S. Langridge, C.H. Marrows and R.L. Stamps. Physical Review Letters 109, 037203 (2012). 22 C. Nisoli, Jie Li, Xianglin Ke, D. Garand, P. Schiffer and Vincent H. Crespi. Physical Review Letters 105, 047205 (2010). 23 S. Ladak, S.K. Walton, K. Zeissler, T. Tyliszczak, D.E. Read, W.R. Branford and L.F. Cohen. New Journal of Physics 14, 045010 (2012). 24 M. Hayashi, L. Thomas, Y. B. Bazaliy, C. Rettner, R. Moriya, X. Jiang and S.S.P. Parkin. Physical Review Letters 96, 197207 (2006). 25 A. Himeno, T. Ono, S. Nasu, K. Shigeto, K. Mibu and T. Shinjo. Journal of Applied Physics 93, 8430 (2003). 26 D.M. Burn, M. Chadha and W.R. Branford. Physical Re- view B 92, 214425 (2015). 27 T J Silva, C S Lee, T M Crawford and C T Rogers. Journal of Applied Physics 85, 7849 -- 7862 (1999). 28 D.M. Burn, E. Arac and D. Atkinson. Physical Review B 88, 104422 (2013). 29 B.K. Middleton, M.M. Aziz and J.J. Miles. IEEE Trans- actions on Magnetics 36, 404 (2000). 

1708.02628

1

1708

What is the best planar cavity for maximizing coherent exciton-photon coupling

We compare alternative planar cavity structures for strong exciton$-$photon coupling, where the conventional distributed Bragg reflector (DBR) and three unconventional types of cavity mirrors$-$ air/GaAs DBR, Tamm $-$ plasmon mirror and sub$-$wavelength grating mirror. We design and optimize the planar cavities built with each type of mirror at one side or both sides for maximum vacuum field strength. We discuss the trade$-$off between performance and fabrication difficulty for each cavity structure. We show that cavities with sub$-$wavelength grating mirrors allow simultaneously strongest field and high cavity quality. The optimization principles and techniques developed in this work will guide the cavity design for research and applications of matter$-$light coupled semiconductors, especially new material systems that require greater flexibility in the choice of cavity materials and cavity fabrication procedures.

cond-mat.mes-hall

cond-mat

What is the best planar cavity for maximizing coherent exciton-photon coupling Zhaorong Wang,1 Rahul Gogna,2 and Hui Deng3, a) 1)EECS Department, University of Michigan, Ann Arbor, MI 48109, USA 2)Applied Physics Department, University of Michigan, Ann Arbor, MI 48109, USA 3)Physics Department, University of Michigan, Ann Arbor, MI 48109, USA We compare alternative planar cavity structures for strong exciton-photon coupling, where the conventional distributed Bragg reflector (DBR) and three unconventional types of cavity mirrors – air/GaAs DBR, Tamm-plasmon mirror and sub-wavelength grating mirror. We design and optimize the planar cavities built with each type of mirror at one side or both sides for maximum vacuum field strength. We discuss the trade-off between performance and fabrication difficulty for each cavity struc- ture. We show that cavities with sub-wavelength grating mirrors allow simultaneously strongest field and high cavity quality. The optimization principles and techniques developed in this work will guide the cavity design for research and applications of matter-light coupled semiconductors, especially new material systems that require greater flexibility in the choice of cavity materials and cavity fabrication procedures. 7 1 0 2 g u A 8 ] l l a h - s e m . t a m - d n o c [ 1 v 8 2 6 2 0 . 8 0 7 1 : v i X r a a)Electronic mail: [email protected] 1 Strong coupling between semiconductor quantum well (QW) excitons and cavity photons leads to new quasi-particles – microcavity polaritons. Since their discovery1, planar micro- cavity polaritons have become a fruitful ground for research on fundamental cavity quantum electrodynamics, macroscopic quantum coherence, and novel device applications2–4. Crucial for polariton research is a cavity with a strong vacuum field strength E at the QW plane and a high quality factor Q. A strong vacuum field leads to stronger exciton-photon cou- pling and thus a larger vacuum Rabi splitting between the polariton modes. A high quality factor leads to long lifetime and coherence time of the cavity photon and correspondingly the polariton. They together enable robust polariton modes, thermodynamic formation of quantum phases, and polariton lasers with lower density threshold at higher operating temperatures3. Given the importance of a strong cavity field, there has been a few decades of effort to improve the cavity field strength. Most commonly used in polariton research are planar Fabry-Perot (FP) cavities formed by two distributed Bragg reflectors (DBRs), each DBR consisting of either epitaxially-grown, closely lattice-matched alloys or amorphous dielectric layers. This type of cavity can reach Q of tens of thousands in III-As based systems5,6, but only about a thousand for III-N7,8 and even lower for other material systems9,10. The field penetrates many wavelengths into the DBR, hence the field strength is not optimal11,12. To achieve a stronger field, different cavity structures have been developed with reduced effective cavity length, including using a metal mirror13 and using Al-oxide14–16 or air/GaAs DBRs13,17 with larger index contrast. These structures showed greater polariton splitting but unfortunately worse Q compared to AlGaAs-based DBRs; polariton lasing have not been possible in these structures. Recently, cavities using a high index-contrast sub-wavelength grating (SWG)18 were demonstrated for polariton lasers19. Not only does SWG allow greater design flexibility20,21 and compatibility with unconventional materials, it is only a fraction in thickness compared to typical DBRs, making it promising for shorter effective cavity length. Yet the demonstrated SWG-based polariton cavity was not optimized for strong cavity field. Here we demonstrate how to optimize these different types of cavity for the strongest field, and compare their polariton splitting, quality factors and practicality for fabrication. In particular, we find SWG cavities may allow simultaneously stronger field and high Q. The cavity field relevant to the exciton-photon coupling is the vacuum fluctuation field. 2 Its strength E is normalized to the zero-point photon energy by Z ǫ(r)E(r)2dV = 1 2 ωc, (1) where ǫ is the permittivity, and ωc is the cavity resonance energy. The integral is evaluated in the entire space with a spatial-dependent ǫ. For a planar cavity, we consider the field being confined in the z-direction but unbounded in the x and y directions. The integration is thus evaluated in a quantization volume (Lx, Ly, Lz), where Lx and Ly are set to be much larger than the wavelength, and Lz is set at some cutoff point where the field has decayed significantly (∼1%). The maximum field strength Emax serves as our main figure of merit for cavities designed for strong coupling. Specifically, for a QW placed at the field maximum, the vacuum Rabi splitting Ω is directly proportional to d · Emax, for d the QW exciton dipole moment. We note that the Emax is closely related to the mode volume commonly used to characterize photonic crystals cavity22,23 as well as the effective cavity length Lc for planar cavities11. We choose Emax over Lc because the vacuum Rabi splitting is determined by not only Lc but also ǫ as Ω ∝ (ǫLc)−1/2,12 while ǫ can vary by an order of magnitude in different cavities. Therefore the maximum vacuum field strength is the most unambiguous quantity to characterize the cavity for strong coupling. Also, we focus on high-Q cavities but do not optimize for Q. This is because for cavities designed to have sufficiently high Q, the experimentally achievable Q depends mainly on practical constrains such as chemical purity and structural integrity of the fabricated structure. In the following, we optimize the maximum vacuum field strength Emax for different types of planar cavities. Based on Eq. 1, the Emax is enhanced in cavities with (i) a tightly confined field profile E(r) and (ii) materials of low refractive indices. A tightly confined field profile effectively reduces the range of integration, enhancing the field maximum. For a planar cavity, this means using the shortest possible cavity length (half wavelength) and using mirrors with shorter field penetration length. Equally important, lower refractive indices lead to higher E(r), especially in the cavity layer where most of the field resides. Vacuum or air has the lowest refractive index. Although not applicable to crystalline QWs that needs mechanical support and surface protection, air-cavities can greatly enhance field strength for QWs formed by two-dimensional (2D) van der Waals materials. We use transfer matrix method for calculations of DBR-based cavities, and rigorous 3 coupled-wave analysis (RCWA) for SWG-based cavities. We obtain the cavity resonance from the reflection spectrum of the cavity, using the real-valued bulk permittivity ǫ∞ for the QW layer. Then we compute complex field distribution at the resonance and normalize the field using Eq. 1. To include the exciton-photon coupling, we use a linear dispersion for the QW layer24 modeled by a Lorentz oscillator25: ǫ(e) = n2(e) = ǫ∞ + f q22 mǫ0Lz 1 e2 0 − e2 − iγe , (2) where f is the oscillator strength per unit area, q and m are the charge and mass of the electron respectively, Lz is the QW thickness, e0 is the exciton energy, and γ is the exciton linewidth. For a fair comparison of the different types of cavities, we focus on 2D half- wavelength cavities with a single QW placed at the field maximum.26 We use III-As systems as an example, because polariton cavities of highest qualities are all based on III-As systems. The cavity is made of AlxGaAs alloys where the different Al content gives different refraction index, with AlAs the lowest (n = 3.02), GaAs the highest (n = 3.68), and AlxGaAs a linear interpolation (n = 3.68 − 0.66x). For the QW, we consider either a 12 nm GaAs/AlAs QW or a 7 nm InGaAs/GaAs QW. For simplicity and ease of comparison, we assume both to have the same exciton energy of 1.550 eV (800 nm)19, oscillator strength of f = 6 × 10−4A−2 27 and γ = 0.8 meV. As a benchmark, we first describe the performance of the most widely used, monolithic DBR-DBR cavities. DBRs are typically made of multiple pairs of high- and low-index layers, all with an optical path length (OPL) of λ/4, for λ the cavity resonance wavelength. High reflectance is achieved by maximal constructive interference among multi-reflections from the layer interfaces. Light decays in a distance inversely proportional to the refraction index contrast of the DBR pair. So a high index contrast is preferred for the DBR pair. For a monolithic III-As cavity with a GaAs QW, we use AlAs (n = 3.02) for the low-index layer and Al0.15Ga0.85As (n = 3.58) for the high index layer, where 15% AlAs alloy is included instead of pure GaAs to avoid absorption at the QW exciton resonance. We consider a bottom DBR with 20 pairs on GaAs substrate, and a top DBR with 16.5 pairs, matching the reflectance of the bottom DBR. An experimental cavity Q of a few thousands can be readily achieved with this structure. More DBR pairs will lead to a higher Q, but will have negligible effect on the field confinement or the vacuum Rabi splitting. To show the enhancement of the cavity vacuum field strength by reducing cavity length, 4 8 7 6 5 4 3 V e m SWG-SWG cavity Tamm-DBR hybrid Air DBR both-side optimized SWG-DBR hybrid Air DBR both-side from Ref[17] Air DBR one-side /2-cavity Conventional DBR Cavities with different cavity lengths 8 -cavity 2 1 1.5 2 2.5 E V LxLy 3 10-6 FIG. 1. Calculated vacuum Rabi splitting Ω and maximum vacuum field strength maxeE for various optimized cavities. All cavities are based on III-As material system. Single QW is embedded at the field maximum. The vacuum field strength is calculated at the cavity resonance energy with QW dispersion turned off, while the Rabi splitting is measured from the two reflection dips when the QW dispersion is turned on. The dash-dotted line is a linear fit through all points. we vary the cavity layer OPL from λ/2 to 8λ. As shown in Fig. 1, Emax increases with decreasing cavity length, and the polariton splitting scales linearly with Emax. This confirms the Emax as an appropriate figure of merit for optimizing exciton-photon coupling. The field distribution of the best-performing λ/2 cavity is shown in Fig. 2(a). The maximum field strength is Emax = 1.89 × 10−6V/pLxLy at the QW, but the field extends many wavelengths into the DBRs due to the relatively small index contrast (< 1.2 : 1) of the DBR layers. The reflectance spectrum shows a polariton splitting of 4.54 meV (Fig. 2(b)). This conventional cavity can be fabricated by mature epitaxial growth technology as a monolithic crystalline structure with few impurities or defects. However, the requirement of lattice-matched materials, with resultantly small index contrast, leads to limited field confinement. For the second type of cavity, we increase the index contrast of the DBR layer by replacing the AlAs layer by air or vacuum (n = 1), which we call an air-DBR. It can be created via selective wet etching13,17. It represents the the highest possible refraction index contrast for a DBR. The GaAs QW still needs to be embedded inside an AlAs cavity (as opposed to an air cavity) to avoid detrimental surface effects. Each material layer – both the cavity layer and the Al0.15GaAs layers in the DBRs, need to be sufficiently thick for mechanical stability. 5 (c) 10-6 (a) 10-6 1.5 1 0.5 0 1600 1800 2000 2200 2400 2600 z (nm) (b) 1 e c n a t c e l f e R 0.8 0.6 0.4 0.2 0 1.544 (e) 10-6 2.5 2 1.5 1 0.5 0 QW dispersion turned on QW dispersion turned off 1.547 1.55 1.553 1.556 Energy (eV) 0 1000 2000 3000 z (nm) 2 1.5 1 0.5 0 1 (f) e c n a t c e l f e R 0.8 0.6 0.4 0.2 0 1.544 (d) 1 e c n a t c e l f e R 0.8 0.6 0.4 0.2 QW dispersion turned on QW dispersion turned off 0 500 1000 1500 2000 z (nm) 0 1.544 1.547 1.55 1.553 1.556 Energy (eV) QW dispersion turned on QW dispersion turned off 1.547 1.55 1.553 1.556 Energy (eV) FIG. 2. Vacuum field profile and Rabi splitting for conventional DBR cavity (a-b) and air DBR cavity (c-f). The color stripes represent different materials– red is Al0.15Ga0.85As, lighter red is the lower index AlAs, white region is air, and the thin stripe at the field maximum is a GaAs QW. All three structures sit on GaAs substrate, which is not shown in (a) and (c). The vacuum field strength has a unit of Volt/(normalization length in meters). The experimentally realized structure used a minimum OPL of 3λ/4 for each suspended layer17, which we also adopt. Based on this, Fig. 2(c) shows the optimal structure we obtained. The top mirror consists of 3 pairs of Al0.15Ga0.85As/air layers. The air layer has λ/4 OPL, the Al0.15Ga0.85As layer has 3λ/4 OPL (∼ 168 nm). The bottom DBR is the same as in the DBR-DBR cavity (Fig. 2(a)). The maximum field strength is increased to Emax = 2.29 × 10−6V/pLxLy. The corresponding vacuum Rabi splitting is Ω = 5.51 meV, 21% larger than the conventional DBR cavity. Further increase of the field confinement can be obtained by replacing the bottom DBR also by an air-DBR, as shown in Fig. 2(e). An experimentally fabricated air-DBR cavity17 used a 7 nm In0.13GaAs QW embedded in a GaAs high-index λ-cavity. For this structure, we calculated Emax = 2.47 × 10−6V/pLxLy and Ω = 6.32 meV (Fig. 1), which is 39% improvement over the conventional DBR cavity. The structure can be further optimized by using a cavity layer of low refraction index. However, an AlAs-cavity is incompatible with the selective wet-etching required to create the bottom air-DBR; instead a Ga-rich layer is needed. Hence, for the cavity, we use a Al0.4Ga0.6As (nr = 3.33) layer of λ/2 OPL sandwiched between two Al0.15Ga0.85As (nr = 3.58) layers of λ/4 OPL, for a total OPL of λ. Similarly the 3/4λ-OPL high-index DBR 6 layer can also be replaced by such sandwich structure to improve the index contrast. The wet etching selectivity of AlAs over Al0.4Ga0.6As has been shown to be 107.28 A 12 nm GaAs QW is placed at the center of the cavity. We obtain Emax = 2.91 × 10−6V/pLxLy and Ω = 7.02 meV, 55% larger than conventional DBR cavity. This structure represents the best Rabi splitting in realistic DBR-based III-As cavities. To obtain even tighter field confinement, we consider another two types of cavities where the DBR is replaced by mirrors with shorter penetration depth – a metal mirror and a sub-wavelength grating mirror. Metal mirror, with a typical penetration depth of less than 100 nm, can be used to form Tamm-plasmon polaritons29. The optimal metal-DBR cavity we find consists of 45 nm thick gold layer as the top mirror (Fig. 3(a)), the same bottom DBR as in the conventional DBR-DBR cavity is used, and a λ/2 AlAs cavity layer slightly adjusted to compensate for the change of the reflection phase from π. In this structure, the field decays rapidly in the gold layer, resulting in a much shorter effective cavity length (Fig. 3(a)), therefore a much higher vacuum field strength. We obtain 30 and Ω = 6.51 meV, 43% larger than conventional DBR Emax = 2.47 × 10−6V/pLxLy cavity. This is an impressive improvement considering only one side of the cavity is replaced by a metal mirror. The structure was used to create the first organic polariton31,32. In III-As cavities, Grossmann et al13 additionally replaced the bottom DBR with an air DBR to further improve the vacuum Rabi splitting to about double that of the conventional DBR cavity.33 However, the drawback of this structure is the low cavity Q due to metal loss. The cavity linewidth is comparable to the Rabi splitting; the system remains in the intermediate coupling regime rather than the strong coupling regime (Fig. 3(b))34. The large loss in metal also makes it difficult to achieve quantum degeneracy or low threshold laser. To maintain both a high cavity Q and short cavity length, we use a III-As subwavelength grating (SWG) for the fourth type of cavity18,19. The SWG is a thin layer of high-index dielectric grating suspended in air, acting as the high-reflective mirror (Fig. 4(a)). To avoid non-zero order evanescent diffraction waves entering the cavity layer, we include a 3/4λ air-gap between the SWG and the bottom structure. The λ/2 cavity layer and bottom DBR are identical to the conventional DBR cavity as in Fig. 2(a). We obtain Emax = 2.23 × 10−6V/pLxLy and Ω = 5.37 meV, both 19% larger compared to the conventional DBR cavity and comparable to the single-sided air-DBR cavity. The advantage of the SWG- DBR cavity is its high cavity quality and simplicity in fabrication. The structure is first 7 (a) 10-6 2 1.5 1 0.5 (b) 1 e c n a t c e l f e R 0.8 0.6 0.4 0.2 0 1000 1200 1400 z (nm) 1600 1800 0 1.52 QW dispersion turned off QW dispersion turned on 1.53 1.54 1.55 1.56 1.57 1.58 Energy (eV) FIG. 3. Vacuum field profile (a) and Rabi splitting (b) for Tamm-plasmon type cavity. The DBR structure is the same with previous DBR cavities. The excitation wave is launched from the substrate to top (left to right). The gold colored region represents a 45nm gold layer. Its refractive index is taken from Olmon, et al35 for evaporated gold at 800nm. The cavity Q is around 60 inferred from the linewidth, mainly due to the loss in the metal. The polariton splitting is 6.51meV, 43% larger than conventional DBR cavity. fabricated as a high-quality, monolithic crystal by epitaxial growth, as with the conventional DBR cavities. The SWG is then created by lithography and etching. Because the grating is the only air-suspended layer created, SWG-DBR hybrid cavity is easier to make and more robust mechanically than an air-DBR. Polariton lasing and excellent coherence properties have been demonstrated in SWG-DBR cavities19,36,37, but not yet in cavities using air-DBR or metal mirrors. Tighter cavity field confinement can be achieved when replacing the bottom DBR also by an SWG. As shown in Fig. 4 (b), a λ/2 AlAs cavity layer sandwiched by two Al0.15Ga0.85As λ/4 layers is air-suspended between two identical SWGs. Because of the evanescent diffrac- tion waves from the SWGs, the field at the QW is slightly non-uniform. We record an average vacuum strength at the QW to be 2.92 × 10−6V/pLxLy. The vacuum Rabi split- ting is Ω = 7.58 meV, 67% larger than that of the conventional DBR cavity. The fabrication of such a double-SWG cavity is more challenging than the SWG-DBR hybrid cavity, but within reach of established fabrication techniques. The top-SWG and the rest of the structure can be fabricated separately at first, then they can be bonded together by cold welding38 or stacked together using the micromanipulation technique applied to making 3D photonic crystals39. Alternatively, one could make the SWGs from Si/SiO2 wafer, then do a wafer bonding with the III-V active layer40. Additional practical challenges include heat dissipation and proper design of current injection path for electrical-pumped polariton devices. Finally we note that the SWG-based cavity may be particularly useful 8 (a) -200 0 200 400 600 800 ) m n ( z 1000 1200 0 10-6 (b) 2 1.5 1 0.5 -200 0 200 400 ) m n ( z 600 800 200 400 x (nm) 0 200 400 x (nm) 10-6 3.5 3 2.5 2 1.5 1 0.5 (c) 1 e c n a t c e l f e R 0.98 0.96 0.94 0.92 QW dispersion turned on QW dispersion turned off 1.545 1.55 1.555 (d) 1 e c n a t c e l f e R 0.98 0.96 0.94 0.92 QW dispersion turned on QW dispersion turned off 1.545 1.55 1.555 Energy (eV) Energy (eV) FIG. 4. Vacuum field profile and Rabi splitting for SWG-DBR hybrid cavity (a-b) and double SWG cavity (c-d). The design parameters for the SWG are, thickness of grating 95.6nm, period 503nm, filling factor 0.3, using Al0.15GaAs(n = 3.58), with air gap 511nm (left) and 144nm (right) between SWG and the cavity layer. Bottom DBR in the hybrid cavity is the same with conventional DBR cavity used above. The middle resonance in the reflection spectrum (d) is due to the exciton absorption of the first-order grating diffraction mode (evanescent in air but propagating in the cavity layer). for coupling to 2D materials, where similar procedures of integration are already necessary. Additionally, it is possible to hold 2D material with lower index materials, further enhancing the vacuum field. In summary, we have investigated the performance of strong coupling for several types of planar cavities. We use maximum vacuum field strength as the main figure of merit, in place of the effective cavity length or mode volume. The Tamm-DBR cavity can pro- vide ∼ 43% improvement in Rabi splitting with only one mirror replaced by a gold thin film. Researchers using active media with a large oscillator strength may find this struc- ture easy to fabricate and yet still sufficient for reaching strong-coupling. However, its low quality factor due to metal loss limits the coherence of the exciton-photon coupling and the resulting polariton modes. The air-DBR type cavities are shown to provide 21% (55%) improvement over conventional DBR cavities if one mirror (both mirror) is replaced with an air-DBR. SWG based cavities provide 19% (67%) improvement when one (two) SWG is used, comparable to (better than) that of air-DBR cavities. In practice, among all al- ternative structures to conventional DBR-DBR cavities, SWG-DBR cavities are so far the only ones where high cavity quality factor and polariton lasing have been demonstrated experimentally. Our findings here may guide the design of cavity systems for the research 9 and applications of semiconductor exciton-polaritons, especially for increasing their operat- ing temperature, lowering the density threshold for quantum phase transition and polariton lasing, and incorporating newer materials. ZW, HD acknowledge the support by the National Science Foundation (NSF) under Awards DMR 1150593 and the Air Force Office of Scientific Research under Awards FA9550- 15-1-0240. We thank Pavel Kwiecien for his open-source RCWA code used for calculations in this work. REFERENCES 1C. Weisbuch, M. Nishioka, a. Ishikawa, and Y. Arakawa, Physical Review Letters 69, 3314 (1992). 2G. Khitrova and H. M. Gibbs, Reviews of Modern Physics 71 (1999). 3H. Deng and Y. Yamamoto, Reviews of Modern Physics 82, 14891537 (2010). 4D. Sanvitto and S. Kna-Cohen, Nature Materials 15, 1061 (2016). 5S. Reitzenstein, C. Hofmann, A. Gorbunov, M. Strau, S. H. Kwon, C. Schneider, A. Lf- fler, S. Hfling, M. Kamp, and A. Forchel, Applied Physics Letters 90, 251109 (2007), http://dx.doi.org/10.1063/1.2749862. 6B. Nelsen, G. Liu, M. Steger, D. W. Snoke, R. Balili, K. West, and L. Pfeiffer, Phys. Rev. X 3, 041015 (2013). 7G. Christmann, R. Butte, E. Feltin, J.-F. Carlin, and N. Grandjean, Applied Physics Letters 93, 051102 (2008), 5. 8A. Das, J. Heo, M. Jankowski, W. Guo, L. Zhang, H. Deng, and P. Bhattacharya, Phys. Rev. Lett. 107, 066405 (2011). 9S. Kena-Cohen and S. R. Forrest, Nat Photon 4, 371 (2010). 10X. Liu, T. Galfsky, Z. Sun, F. Xia, and E.-c. Lin, Nature Photonics 9, 30 (2015), arXiv:1406.4826. 11Y. Yamamoto, S. MacHida, and G. Bjork, Physical Review A 44, 657 (1991). 12T. Norris, J. Rhee, C. Sung, Y. Arakawa, M. Nishioka, and C. Weisbuch, Physical review. B, Condensed matter 50, 14663 (1994). 13C. Grossmann, C. Coulson, G. Christmann, I. Farrer, H. E. Beere, D. a. Ritchie, and J. J. Baumberg, Applied Physics Letters 98, 231105 (2011). 10 14T. R. Nelson, J. P. Prineas, G. Khitrova, H. M. Gibbs, J. D. Berger, E. K. Lindmark, J.-H. Shin, H.-E. Shin, Y.-H. Lee, P. Tayebati, and L. Javniskis, Applied Physics Letters 69, 3031 (1996). 15L. A. Graham, Q. Deng, D. G. Deppe, and D. L. Huffaker, Applied Physics Letters 70, 814 (1997). 16a. R. Pratt, T. Takamori, and T. Kamijoh, Applied Physics Letters 74, 1869 (1999). 17J. Gessler, T. Steinl, A. Mika, J. Fischer, G. S¸ek, J. Misiewicz, S. Hofling, C. Schneider, and M. Kamp, Applied Physics Letters 104, 081113 (2014). 18M. C. Huang, Y. Zhou, and C. J. Chang-Hasnain, Nature Photonics 1, 119122 (2007). 19B. Zhang, Z. Wang, S. Brodbeck, C. Schneider, M. Kamp, S. Hofling, and H. Deng, Light: Science & Applications 3, e135 (2014). 20Z. Wang, B. Zhang, and H. Deng, Physical Review Letters 114, 1 (2015). 21B. Zhang, S. Brodbeck, Z. Wang, M. Kamp, C. Schneider, S. Hofling, and H. Deng, Applied Physics Letters 106 (2015), 10.1063/1.4907606. 22J. S. Foresi, P. R. Villeneuve, J. Ferrera, E. R. Thoen, G. Steinmeyer, S. Fan, J. D. Joannopoulos, L. C. Kimerling, H. I. Smith, and E. P. Ippen, Nature 390, 143 (1997). 23P. T. Kristensen, C. Van Vlack, and S. Hughes, Optics Letters 37, 1649 (2012). 24Y. Zhu, D. J. Gauthier, S. E. Morin, Q. Wu, H. J. Carmichael, and T. W. Mossberg, Physical Review Letters 64, 2499 (1990). 25R. Houdr´e, R. P. Stanley, U. Oesterle, M. Ilegems, and C. Weisbuch, Physical Review B 49, 16761 (1994). 26In all cavities, multiple QWs can be used to enhance the polariton splitting. 27L. C. Andreani and A. Pasquarello, Physical Review B 42, 8928 (1990). 28E. Yablonovitch, T. Gmitter, J. P. Harbison, and R. Bhat, Applied Physics Letters 51, 2222 (1987). 29M. Kaliteevski, I. Iorsh, S. Brand, R. A. Abram, J. M. Chamberlain, A. V. Kavokin, and I. A. Shelykh, Physical Review B - Condensed Matter and Materials Physics 76, 1 (2007). 30For strong dispersive media like metal, the energy density in field normalization equation 1 is modified by Eq.6.126 in Jackson, Classical Electrodynamics, 3rd edition, 2001. 31D. G. Lidzey, D. D. C. Bradley, M. S. Skolnick, T. Virgili, S. Walker, and D. M. Whittaker, Nature 395, 53 (1998). 32D. G. Lidzey, D. D. C. Bradley, T. Virgili, A. Armitage, M. S. Skolnick, and S. Walker, 11 Phys. Rev. Lett. 82, 3316 (1999). 33We have not been able to reproduce the field profile or reflection spectrum shown in Ref.13 in our simulation and instead found the 45 nm thick gold layer as the optimal for a Tamm- plasmon mirror. 34V. Savona, L. C. Andreani, P. Schwendimann, and A. Quattropani, Solid State Commu- nications 93, 733 (1995). 35R. L. Olmon, B. Slovick, T. W. Johnson, D. Shelton, S.-h. Oh, G. D. Boreman, and M. B. Raschke, Phys. Rev. B 86, 1 (2012). 36J. Fischer, S. Brodbeck, B. Zhang, Z. Wang, L. Worschech, H. Deng, M. Kamp, C. Schnei- der, and S. Hfling, Applied Physics Letters 104 (2014), 10.1063/1.4866776. 37S. Kim, B. Zhang, Z. Wang, J. Fischer, S. Brodbeck, M. Kamp, C. Schneider, S. Hfling, and H. Deng, Physical Review X 6, 011026 (2016). 38S. K´ena-Cohen, M. Davan¸co, and S. Forrest, Physical Review Letters 101, 1 (2008). 39K. Aoki, D. Guimard, M. Nishioka, M. Nomura, S. Iwamoto, and Y. Arakawa, Nature Photonics 2, 688 (2008). 40C. Sciancalepore, B. Ben Bakir, C. Seassal, X. Letartre, J. Harduin, N. Olivier, J. Fedeli, and P. Viktorovitch, IEEE Photonics Journal 4, 399 (2012). 12 

1808.09157

1

1808

Theory of charge-spin conversion at oxide interfaces: The inverse spin-galvanic effect

We evaluate the non-equilibrium spin polarization induced by an applied electric field for a tight-binding model of electron states at oxides interfaces in LAO/STO heterostructures. By a combination of analytic and numerical approaches we investigate how the spin texture of the electron eigenstates due to the interplay of spin-orbit coupling and inversion asymmetry determines the sign of the induced spin polarization as a function of the chemical potential or band filling, both in the absence and presence of local disorder. With the latter, we find that the induced spin polarization evolves from a non monotonous behavior at zero temperature to a monotonous one at higher temperature. Our results may provide a sound framework for the interpretation of recent experiments.

cond-mat.mes-hall

cond-mat

a Theory of charge-spin conversion at oxide interfaces: The inverse spin-galvanic effect Gotz Seibolda, Sergio Caprarab, and Roberto Raimondic aInstitut fur Theoretische Physik, BTU Cottbus-Senftenberg, PBox 101344, 03013 Cottbus, bDipartimento di Fisica Universit`a di Roma Sapienza, piazzale Aldo Moro 5, I-00185 Roma, Germany cDipartimento di Matematica e Fisica, Universit`a Roma Tre, Via della Vasca Navale 84, 00146 Italy Rome, Italy ABSTRACT We evaluate the non-equilibrium spin polarization induced by an applied electric field for a tight-binding model of electron states at oxides interfaces in LAO/STO heterostructures. By a combination of analytic and numerical approaches we investigate how the spin texture of the electron eigenstates due to the interplay of spin-orbit coupling and inversion asymmetry determines the sign of the induced spin polarization as a function of the chemical potential or band filling, both in the absence and presence of local disorder. With the latter, we find that the induced spin polarization evolves from a non monotonous behavior at zero temperature to a monotonous one at higher temperature. Our results may provide a sound framework for the interpretation of recent experiments. Keywords: Spin-orbit coupling, spin-charge conversion, oxides interfaces 1. INTRODUCTION It is well known that the breaking of the inversion symmetry leads to the so-called Rashba spin-orbit coupling (SOC),1 -- 3 where polar and axial vectors transform similarly.4 Basically this allows for two major possibilities of charge to spin conversion: The spin Hall (SH)5 and the inverse spin galvanic (ISG) effect,6, 7 as well as for their Onsager reciprocal effects. While the SH effect converts an electrical current into a spin imbalance at the sample edges via an induced perpendicular spin current, the ISG effect creates a bulk non-equilibrium spin polarization by a flowing electrical current.6, 8 -- 11 The inverse SG effect corresponds then to the production of electrical current via the pumping of spin polarization.12, 13 Both the SG12, 13 and the ISG14 -- 20 effects have been observed in semiconductors. In the first case an electrical current is measured after pumping spin polarized light (SG) whereas in the second case Faraday and Kerr spectroscopies measure the spin polarization induced by the applied current. The SG effect has also been very effectively measured by spin pumping from an adjacent ferromagnet into a metallic interface,21 into a topological insulator surface23, 24 and more recently into the two-dimensional electron gas (2DEG) in oxide LAO/STO heterostructures.25 -- 28 These latter materials have emerged29 -- 33 as very promising materials for the SG and ISG effect, due to the large values of the Rashba SOC parameter α as experimentally observed34 -- 37 and also theoretically calculated,38 -- 40 even though it is likely that, due to their complex band structure, the available theory41 of the SG/ISG effect developed for the 2DEG in semiconductors may not be able to capture a number of specific features. A first step in this direction has been made recently by a combination of analytical diagrammatic and numerical approaches.42, 43 The layout of the paper is the following. In the next section we introduce a model for the electron states relevant for describing transport at oxide LAO/STO interfaces. In section 3 we provide the necessary formalism of linear response theory for the SG and ISG effects. In section 4 we introduce an approximate effective model for electron states close to band minima. In section 5 we evaluate analytically the SG response for the effective Further author information: (Send correspondence to Roberto Raimondi.) Roberto Raimondi: E-mail: [email protected] H0 = εxy k 0 0 0 εxz k 0 0 0 εyz k  (1) model, whereas in section 6 we introduce disorder and the necessary formalism to handle it. Finally in section 7 we present a fully numerical approach which includes both cases without and with disorder. We conclude in section 8. A number of technical details are provided in the appendices. 2. THE MODEL The electronic structure of the 2DEG at LAO/STO interfaces, perpendicular to the (001) crystal direction, is usually described39, 44, 45 within a tight-binding Hamiltonian H0 for the Ti t2g orbitals, dxy, dxz,dyz, supplemented by local atomic spin-orbit interactions with Hamiltonian Haso and an interorbital hopping with Hamiltonian HI which is induced by the interface asymmetry. The hopping between the d orbitals of two neighbouring cubic cells is mediated via intermediate jumps to p orbitals. For instance, the hopping between two dxy orbitals along the x axis occurs via two successive hopping dxy → py and py → dxy. In the first hop, the overlap, which is of order ∼ tpd, yields a positive sign, whereas the sign is negative ∼ −tpd in the second one. Hence the effective dxy − dxy hopping goes like −t2 pd/∆E, with ∆E being the energy difference between d and p orbitals. As a result, in the basis xy(cid:105),xz(cid:105),yz(cid:105) the hopping between similar orbitals reads as with, setting to unity the lattice spacing, εxy k = −2t1[cos(kx) + cos(ky) − 2] − 4t3[cos(kx) cos(ky) − 1] εxz k = −2(t1 + t3)[cos(kx) − 1] − 2t2[cos(ky) − 1] + ∆ εyz k = −2(t1 + t3)[cos(ky) − 1] − 2t2[cos(kx) − 1] + ∆ where the energy difference ∆ between the xy(cid:105) and xz(cid:105),yz(cid:105) states is due to the confinement of the 2DEG in the xy-plane.46 The atomic SOC is given by with τ i denoting the Pauli matrices. Haso = ∆aso 0 iτ x −iτ y −iτ x 0 iτ z iτ y −iτ z 0  (2) (cid:48) Figure 1. At the interface an orbital polarization and (or) orbital displacement results in hopping processes ∼ t pd as between px- and zx-orbitals along the y-direction. The asymmetry is visualized by a small shift of px-orbitals along the z-direction. Hopping between different d orbitals may occur if inversion symmetry is broken, see Fig. 1. Consider, for instance, the two hops along the y direction, dxy → px and px → dzx. While the first hop ∼ tpd is as the first hop of the effective hopping between two dxy orbitals discussed above, the second hop ∼ t(cid:48) pd will be forbidden in the presence of inversion symmetry. To see this consider that tpd = (cid:104)px, R + a 2 yHdxy, R(cid:105), where R is the lattice site of the dxy orbital. In the same way t(cid:48) pd = (cid:104)dzx, R + ayHpx, R + a 2 y(cid:105). In both cases, H is the full Hamiltonian. If H is invariant with respect to the inversion z → −z, then necessarily t(cid:48) pd = 0, because px is even, while dzx is odd. Clearly if H has terms which are not invariant for z → −z, then t(cid:48) pd (cid:54)= 0. As a result the interface asymmetry hopping reads44 (3) HI = γ 0 2i sin(ky) 2i sin(kx) −2i sin(ky) −2i sin(kx) 0 0 0 0  . In the following we use the parameters, t1 = 0.277 eV, t2 = 0.031 eV, t3 = 0.076 eV, ∆ = 0.4 eV, ∆aso = 0.010 eV, γ = 0.02 eV, which have been derived in Ref.39 from projecting DFT on the t2g Wannier states. Note that for the splitting ∆ we take a value intermediate between the theoretical (∆ = 0.19 eV) and the experimental one (∆ = 0.6 eV). The left panel of Fig. 2 shows the band dispersions along the x axis for these values of the parameters. The bands come naturally in three pairs, which are split by the combined effect of the spin-orbit coupling and the inversion symmetry breaking. For our analysis we have selected three different values of the chemical potential for corresponding filling regimes. For µ = 0.3 eV, only the lowest pair of bands (1,2) is occupied. The chemical potential µ = 0.425 eV is close to the Lifshitz point, where the spin-orbit splitting is large and the pairs of bands (3,4) and (5,6) start to be filled. Finally, the chemical potential µ = 0.7 eV is in the regime, where all pairs of bands (1,2), (3,4) and (5,6) are occupied. We now analyze the chirality for each eigenstate band p = 1, . . . , 6 by computing the spin at each momentum point of the Fermi surface (FS) according to Sα(p, kF ) = (cid:88)n,σ,σ(cid:48) Φ∗n,σ(p, kF )τ α σ,σ(cid:48)Φn,σ(cid:48)(p, kF ) where Φn,σ(cid:48)(p, kF ) are the eigenfunctions of the system at momentum kF . The indices n = xy, xz, yz and σ label the orbital and its spin. Then the chirality of the p-th band can be obtained from kFS(p, kF (cid:19) . α(p) = arcsin(cid:18) kF × S(p, kF ) · ez Fig. 2 shows the chiralities for each pair of bands at selected chemical potentials and the corresponding FSs. For the lowest pair of bands (1,2) the momentum dependent spin pattern displays a vortex-type structure with the core centered at Γ = (0, 0). Thus, even when the FS changes from electron- to hole-like between µ = 0.5 and µ = 0.6, the corresponding chiralities are always confined to α(p) ≈ ±π/2 without any sign change in α. For the middle pair of bands (3,4) the spin structure is composed of two vortex patterns (with the same vorticity) centered at (π, 0) and (0, π). As a consequence, the spin texture vanishes along the diagonals and a Rashba-type description along this direction fails. In section 4 we will come back to this point. However, for small µ and all other momenta the chirality also starts at α ≈ ±π/2 but then on average becomes smaller with increasing chemical potential and eventually changes sign for µ ≈ 1.5. An analogous situation occurs for the uppermost pair of bands where the 'spin-vortex core' is centered at (π, π). In this case the chiralities also change sign upon increasing the chemical potential while at small µ one again recovers α ≈ ±π/2. 3. LINEAR RESPONSE THEORY In this paper we aim at evaluating the spin polarization induced by an externally applied electric field. To be definite we take the electric field along the x axis and the spin polarization along the y axis. To linear order in the applied field we write the spin polarization as sy(ω) = σISG(ω) Ex(ω), (4) Figure 2. Left panel: Structure of the t2g interface bands. The inset enlarges the region around the 'Lifshitz' point where the spin-orbit splitting is large. The horizontal dashed lines in the main panel refer to three values of chemical potential: µ = 0.3 eV (blue line), µ = 0.425 eV (red line) and µ = 0.7 eV (green line). The right panel displays the spin texture for the three pairs of bands together with their Fermi surfaces. For the lower pair of bands (1,2) the spins point in the opposite direction. where σISG, the "conductivity" for the ISG effect, can be obtained by the zero-momentum limit of the Fourier transform Ryx(ω) of the response function (henceforth the symbols in capital letters indicate the operators for spin density and charge current) defined as where the brackets stand for the quantum-statistical average and θ (t) is the Heaviside step function. The frequency-dependent ISG conductivity reads Ryx(t, r) = −ıθ(t)(cid:104)[Sy(t, r), Jx](cid:105) , (5) σISG(ω) = lim η→0+ (cid:60)(cid:20) Ryx(ω) ı(ω + ıη)(cid:21) = −πδ(ω)R(cid:48) yx(0) + P R(cid:48)(cid:48) yx(ω) ω ≡ DISGδ (ω) + P R(cid:48)(cid:48) yx(ω) ω , (6) where the first term will be referred to as the Drude singular term and the second as the regular term, in analogy with the terminology used in the case of the optical conductivity. Because under time reversal both the charge current and the spin polarization are odd, according to the Onsager relation, the SG and ISG conductivities are equal.41 For this reason we will use the term SG conductivity (SGC) for both direct and inverse effects. The calligraphic symbol P stands for the principal part. The real R(cid:48) yx(ω) parts of the yx(ω) and the imaginary R(cid:48)(cid:48) response function are related by the Kramers-Kronig relation (KKR) ∞ −∞ d ω(cid:48) R(cid:48)(cid:48) yx(ω(cid:48)) ω(cid:48) − ω . ∞ d ω σISG(ω) = 0 (7) (8) By integration over the frequency, thanks to the KKR, the SGC satisfies the following sum rule R(cid:48) yx(ω) = 1 π due to the fact that for the SGC there is no 'diamagnetic' contribution as opposed to the optical conductivity. In the following we are going to apply the above formulae to the model introduced in section 2. To this end, it is instructive to consider first the case of the Rashba SOC for a 2DEG with quadratic dispersion relation in −∞ -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx kxkydkx~ dSyEx-π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx -π-π/20π/2π-π-π/20π/2πky kx a)b)c)Band 1Band 2Band 3Band 4Band 5Band 65FIG.5:Leftpanel:Structureofthet2ginterfacebands.Theinsetenlargestheregionaroundthe'Lifshitz'pointwherethespin-orbitsplittingislarge.TheverticallinesinthemainpanelrefertothechemicalpotentialsforwhichtheFermisurfaceisshowninFig.1.TherightpanelshowdisplaysthechiralspinstructurefortheupperofthethreepairofbandstogetherwiththeFS.Forthelowerbandsthespinspointintheoppositedirection.PERTURBATIVECALCULATIONAroundthepointthenon-interactingpartofthehamiltonianreads"xyk=(t1+2t3)k2(12)"xzk=(t1+t3)k2x+t2k2y+(13)"yzk=(t1+t3)k2y+t2k2x+.(14)Theatomicspin-orbitcoupling(⇠⌧z,cf.Eq.(2))liftsthedegeneracybetweenxzandyzbutleavesthespindegeneracy,cf.Fig.7.OneobtainstheneweigenvaluesE±=+12(t1+t2+t3)k2±12q(t1+t3t2)2(k2xk2y)2+42asoandeigenfunctionxz,"i=ia+,"i+ib,"i(15)yz,"i=b+,"i+a,"i(16)xz,#i=ia+,#i+ib,#i(17)yz,#i=b+,#ia,#i(18)witha=1p2r1+"xz"yz("xz"yz)2+42aso(19)⇡1p21+14aso(t1+t3t2)(k2xk2y)b=1p2r1"xz"yz("xz"yz)2+42aso(20)⇡1p2114aso(t1+t3t2)(k2xk2y).InthisnewbasistheasymmetryhoppingisgivenbyHI=p2(ky+ikx)xy,"ih+,"+h.c.+p2(kyikx)xy,#ih+,#+h.c.+p2(kyikx)xy,"ih,"+h.c.+p2(ky+ikx)xy,#ih,#+h.c.andtheresidualcouplingofHasowiththexy-levelreadsHaso=p2asoxy,"ih+,#+h.c.+p2asoxy,#ih+,"+h.c.12p2(t1+t3t2)(k2xk2y)xy,"ih,#+h.c.12p2(t1+t3t2)(k2xk2y)xy,#ih,"+h.c..Inthefollowingwerestrictontheregionclosetothe-pointwhereaso>tk2andneglectthereforethetwoEk[eV](kx,0)the effective mass approximation. The insight gained in this simpler case will guide us also in the analysis of the model with a complex band structure. We consider then the Rashba-Bychkov Hamiltonian3 H = p2 2m + α(τ xpy − τ ypx), (9) where m is the effective mass and α the SOC. The 2DEG is confined to the xy plane and px and py are the momentum operators along the two coordinate axes. Clearly there are two eigenvalues E±(p) = p2/2m± αp with the corresponding eigenstates of (9) being plane waves whose spin quantization axis is fixed by the momentum direction p, s(cid:105) = 1 √2(cid:18) s ıe−ıθ 1 (cid:19) , s = ±1 where tan(θ) = py/px. The ISG response function at finite frequency and momentum reads Ryx(ω, q) = (cid:88)p,s1,s2 (cid:104)p, s1Syp, s2(cid:105)(cid:104)p, s2Jxp, s1(cid:105) f (Es1 (p) − µ) − f (Es2(p + q) − µ) ω + ıη + Es1(p) − Es2 (p + q) , (10) (11) where f (E) is the Fermi distribution function at temperature T . Depending on the values of the spin indices, one has intraband (s1 = s2 = ±1) and interband (s1 = −s2 = ±1) contributions. In the dynamic limit, when the momentum goes to zero at finite frequency, the intraband contribution vanishes. For the model of Eq. (9) the interband matrix elements for spin density Sy = τ y/2 and charge current Jx = (−e)(px/m − ατ y) read (cid:104)p, sSyp,−s(cid:105) = (cid:104)p,−sJxp, s(cid:105) = (−e)α(ıs) sin(θ), (−ıs) sin(θ), 1 2 and the zero-momentum response function becomes Ryx(ω) = 1 2 (−e)α(cid:88)ps sin2(θ) f (Es(p) − µ) − f (E−s(p) − µ) ω + ıη + Es(p) − E−s(p) . (12) At zero temperature, there are two FSs corresponding to the two spin helicity bands with Fermi momenta p± = (cid:112)2mµ + (mα)2 ∓ mα. The evaluation of the imaginary part of the zero-momentum response function leads to (η → 0+) R(cid:48)(cid:48) yx (ω) = e 32α ω [θ (ω − 2αp−) − θ (ω − 2αp−)] , (13) showing an antisymmetric behavior with respect to the frequency ω. The spectral weight, at positive frequency, is confined in the range 2αp+ < ω < 2αp−. The two frequencies delimiting the interval are nothing but the spin- orbit splitting at the two Fermi surfaces. We note, and this will turn out useful when discussing the numerical calculations, that at finite η, the imaginary part remains finite and acquires a linear-in-frequency behavior around the origin, whose slope vanishes as η. The Drude weight, according to Eq. (6) can be easily obtained by the KKR relation (7) to read DISG = − π 2 eN0α, (14) where N0 = m/(2π) is the single-particle density of states of the 2DEG. For the sake of simplicity we have chosen units such  = 1. There are two features worth noticing. The first is that the Drude weight is controlled by the sign of the SOC. The second is that the Drude weight arises from the interband transitions between the spin-orbit split bands. This must be compared with the case of optical conductivity for the electron gas, where the Drude weight arises from the diamagnetic contribution to the current. In the present case, due to the sum rule (8), the Drude low-frequency peak yields information about the spectral weight of interband transitions at finite frequency. To the best of our knowledge this feature has not been noticed before. In the following of the paper we will consider the effect of disorder, but it is instructive to make here an heuristic discussion. In the presence of spin-independent disorder, due to the form (10) of the eigenstates, the electron spin acquires a finite relaxation rate τ−1 . This mechanism, which is known as the Dyakonov-Perel relaxation, arises because, at each scattering event, the change in momentum also affects the spin eigenstate. As a result, in the diffusive approximation, ωτ (cid:28) 1, the spin density obeys a Bloch equation51 s dsy dt = − 1 τs (sy − s0), (15) where s0 = −eαN0τ E represents the steady-state nonequilibrium spin polarization8 induced by an applied electric field E along the x axis and τ is the momentum relaxation scattering time (not to be confused with the Pauli matrices τ i). According to Ref.51 the Dyakonov-Perel relaxation rate reads By Fourier transforming (15) to frequency ω, one obtains the SGC in the form 1 τs = 1 2τ 4α2p2 F τ 2 F τ 2 . 1 + 4α2p2 σISG(ω) = −eαN0 τ τs τ−1 s ω2 + τ−2 s , (16) (17) which has a Lorentzian lineshape and evolves to a singular contribution in the weak scattering limit τ → ∞. More precisely by integrating over frequency one obtains ∞ −∞ dωσISG(ω) = −eαN0π τ τs π 2 = − eαN0, (18) which reproduces the Drude weight of Eq. (14). Notice that in the last step we made use of the fact that the spin relaxation time becomes twice the momentum relaxation time in the weak scattering limit according to Eq. (16). Eq. (18) seems to violate the sum rule (8), but this is not the case. The form (17) for the SGC has been derived in the diffusive approximation, which is valid for frequencies ω (cid:28) τ−1 well below the region of the interband spectral weight. Hence, the form (17) captures only the low frequency spectral weight, which evolves in the singular Drude weight in the limit of vanishing disorder. The effect of disorder is then to eliminate the Drude singular contribution and to yield a finite SGC at zero frequency, which is the result of a finite slope of the imaginary part of the response function. The microscopic approach in the presence of disorder is discussed in section 6 and details about the frequency dependence are developed in the appendix B. Around the Γ point the non-interacting part of the Hamiltonian (1) reads 4. EFFECTIVE MODELS εxy k = (t1 + 2t3)k2 k = (t1 + t3)k2 εxz εyz k = (t1 + t3)k2 x + t2k2 y + t2k2 y + ∆ x + ∆ , (19) (20) (21) x + k2 y. The atomic SOC [∼ τ z, cf. Eq. (2)] lifts the degeneracy between xz and yz but leaves where k2 = k2 the spin degeneracy, cf. Fig. 3. One obtains the new −(+), corresponding to the pairs of bands (3,4) and (5,6) respectively, E±σ = ∆ + (t1 + t2 + t3)k2 1 2 ± 1 2(cid:113)(t1 + t3 − t2)2(k2 x − k2 y)2 + 4∆2 aso and eigenfunctions with 1 √2(cid:114)1 + a ≈ xz,↑(cid:105) = ıa+,↑(cid:105) + ıb−,↑(cid:105) yz,↑(cid:105) = −b+,↑(cid:105) + a−,↑(cid:105) xz,↓(cid:105) = ıa+,↓(cid:105) + ıb−,↓(cid:105) yz,↓(cid:105) = b+,↓(cid:105) − a−,↓(cid:105) (22) (23) (24) (25) εxz − εyz (εxz − εyz)2 + 4∆2 1 √2(cid:18)1 + 1 4∆aso aso ≈ (t1 + t3 − t2)(k2 x − k2 y)(cid:19) , b =(cid:112)1 − a2 In the basis +,↑(cid:105),+,↓(cid:105),−,↑(cid:105),−,↓(cid:105) the asymmetry hopping is given by HI = √2γ(ky + ıkx)xy,↑(cid:105)(cid:104)+,↑ + h.c. + √2γ(ky − ıkx)xy,↓(cid:105)(cid:104)+,↓ + h.c. + √2γ(ky − ıkx)xy,↑(cid:105)(cid:104)−,↑ + h.c. + √2γ(ky + ıkx)xy,↓(cid:105)(cid:104)−,↓ + h.c. and the residual coupling of Haso with the xy-level reads Haso = √2∆asoxy,↑(cid:105)(cid:104)+,↓ + h.c. + √2∆asoxy,↓(cid:105)(cid:104)+,↑ + h.c. x − k2 − x − k2 (t1 + t3 − t2)(k2 (t1 + t3 − t2)(k2 1 2√2 1 2√2 − y)xy,↑(cid:105)(cid:104)−,↓ + h.c. y)xy,↓(cid:105)(cid:104)−,↑ + h.c. . In the following we restrict to the region close to the Γ point, where ∆aso > tk2, and neglect therefore the two latter terms in Haso resulting in the effective coupling structure depicted in panel (b) of Fig. 3. We can now calculate the effective interactions between levels α,β in 2nd order perturbation theory (cid:104)αH (2)β(cid:105) = − 1 2(cid:88)n (cid:18) 1 En − Eα + 1 En − Eβ(cid:19) Hα,nHn,β (26) and α,β either corresponds to the xy or to the ± levels. For the xy states one finds (cid:104)xy,↑ H (2)xy,↓(cid:105) ≈ −(cid:104)xy,↑ HI+,↑(cid:105)(cid:104)+,↑ Hasoxy,↓(cid:105) − (cid:104)xy,↑ Haso+,↓(cid:105)(cid:104)+,↓ HIxy,↓(cid:105) ∆ ∆ and similarly for (cid:104)xy,↓ H 2xy,↑(cid:105). Inserting the matrix elements yields an effective Rashba SOC ∼ kyτ x − kxτ y H SOC xy = −4 γ∆aso ∆ (cid:18) 0 ky + ıkx ky − ıkx 0 (cid:19) = −αxy(τ xky − τ ykx) (27) with a negative coupling constant with αxy = 4γ∆aso/∆. From Fig. 3 one can see that the same matrix elements also mediate the 2nd order interaction between the +, σ and +,−σ states. Since in this case the denominator in Eq. (26) is negative we obtain a positive coupling α+ = 4γ∆aso/∆ for the E+ states H SOC E+ = 4 γ∆aso ∆ (cid:18) 0 ky + ıkx ky − ıkx 0 (cid:19) = α+(τ xky + τ ykx) . (28) Figure 3. Level structure and interactions of the three-band hamiltonian. Interactions of Haso ∼ ∆aso Eq. (2) are shown in red and those of the asymmetry hopping HI Eq. (3) are indicated in blue. The red dashed lines correspond to a atomic SO interaction which around the Γ-point (i.e. ∼ tk2 < ∆aso) is much smaller than the matrix elements represented by the solid lines. Panel (a) corresponds to the original Hamiltonian whereas in (b) the atomic SOC between xz and yz has been diagonalized. Moreover the off-diagonal matrix elements in Eq. (28) are c.c. to those of Eq. (27) which means that the +, σ and +,−σ states are interacting via a Dresselhaus coupling ∼ kyτ x + kxτ y. The effective interactions between −, σ and −,−σ can be again obtained from 2nd order perturbation theory in the limit tk2 < ∆aso which now involves the matrix elements represented by the dashed lines in Fig. 3. The resulting effective coupling reads γ(t1 + t3 − t2)(k2 x − k2 y) 0 ky + ıkx − H SOC x − k2 and therefore corresponds to a linear Rashba SOC but with a coupling constant ∼ (k2 ky − ıkx = − y)(τ xky − τ ykx) x − k2 y). ∆ 0 (cid:19) = −β(k2 (cid:18) 5. THE CLEAN LIMIT In this section we evaluate the Drude weight for the effective models discussed in section 4. 5.1 xy bands In this case the eigenvalues and eigenvectors corresponding to the Hamiltonian (27) read (29) (30) Exy ± = k2 2m ± αxyk, ±(cid:105) = 1 √2(cid:18) ∓ıe−ıθ 1 (cid:19) with px = cos(θ) and py = sin(θ). As for the Rashba model (9), the spin operator is simply the Pauli matrix Sy = τ y/2 and the charge current is similar to the 2DEG case Jx = (−e)(kx/mτ 0 + αxyτ y). The interband matrix elements read (cid:104)k, sSyk,−s(cid:105) = 1 2 is sin(θ), (cid:104)k,−sJxk, s(cid:105) = −is sin(θ)(−e)αxy and the response function (in the zero-temperature limit) gives R(cid:48) yx(ω → 0) = 1 2 (−e)αxy(cid:88)ps sin2(θ) θ(Exy s (p) − µ) − θ(Exy s (p) − Exy Exy −s(p) −s(p) − µ) = 1 2 (−e)αxyN0, which leads to the Drude weight with an opposite sign as compared to the 2DEG case of Eq. (14). DISG xy = π 2 eαxyN0 (31) a)xyxyxyxyxzxzyzyz+−−+b)5.2 E− bands In this case the eigenvalues and eigenvectors corresponding to the Hamiltonian (29) read E− ± = k2 2m ± βζ k3, ±(cid:105) = 1 √2(cid:18) ∓ı ζ e−ıθ ζ 1 (cid:19) (32) where ζ = k2 x − y)τ y/2, while the charge current has a more complicated structure as compared to the 2DEG case Jx = k2 In this case the spin operator reads Sy = −γ(k2 y with kx = cos(θ) and ky = sin(θ). x − k2 (−e)(cid:0) kx m τ 0 − 2βkxkyτ x + β(3k2 x − k2 sin(θ) y)τ y(cid:1). The interband matrix elements read ζ (cid:104)k, sτ yk,−s(cid:105) = ıs ζ ζ (cid:104)k,−sτ yk, s(cid:105) = −ıs ζ ζ (cid:104)k, sτ xk,−s(cid:105) = ıs cos(θ) ζ ζ (cid:104)k,−sτ xk, s(cid:105) = −ıs ζ cos(θ). sin(θ) In the response function R(cid:48) xy(0) = −(−e)(cid:88)k,,s = (−e) γ 2 0 2π 2π 1 2 (−γk2)(k2 dθ 2π (k2 x − k2 y)2 γ = (−e) 2 = (−e)γk2 0 F βk2 dθ 2π x − k2 (k2 y)2 F N0(cid:18)− 4(cid:19) 1 y)k2(cid:104)−2βk2 x k2 k+ dk 2π k− + − k3 k3 − 6π x − k2 k2 y ζ k2 y ζ k2 y + β(3k2 x k2 y − k4 y)(cid:105) θ(µ − E−s (k)) − θ(µ − E− 2sβζk3 −s(k)) the factors ζ disappear and the Drude weight reads DISG − = (−e)(cid:16) π 4 γp2 F βp2 F N0(cid:17) . 5.3 E+ bands In this case the eigenvalues and eigenvectors corresponding to the Hamiltonian (28) read (33) (34) with kx = cos(θ) and ky = sin(θ). In this case the spin operator reads Sy = γ(k2 x − k2 y)τ y/2, while the charge current is Jx = (−e)(cid:0) px 1 Exy ± = k2 2m ± α+k, ±(cid:105) = 1 (cid:19) √2(cid:18) ∓ıeıθ m τ 0 + α+τ y(cid:1). The interband matrix elements are (cid:104)k, sτ yk,−s(cid:105) = −ıs sin(θ), (cid:104)k,−sτ yk, s(cid:105) = ıs sin(θ). The response function is R(cid:48) xy(0) = −(−e)γ(cid:88)k,s k2(k2 x − k2 y)k2 y = −(−e) = (−e)γk2 2π dθ 2π γ 2 x − k2 (k2 4(cid:19) 1 0 F α+N0(cid:18)− s (k)) − θ(µ − E+ −s(k)) 2sα+k α+ 2 θ(µ − E+ k+ k2dk 2π y)k2 y k− and the Drude weight is DISG + = (−e)(cid:16) π 4 γp2 F α+N0(cid:17) . (35) 6. THE DISORDERED LIMIT It is well known that in the presence of disorder, the Drude weight in the formula for the optical conductivity is suppressed and the spectral weight goes into the regular part. In the Drude model, the regular part, as function of the frequency, has a Lorentzian shape whose width is controlled by the scattering rate τ−1 (not to be confused with the Pauli matrices). Such a transfer of spectral weight from the singular to the regular part occurs also in the case of the SGC. To this end we need to introduce disorder in our model. This will be done in the numerical computation of the next section, whereas in this section we introduce disorder within the effective models derived in section 4 by using the standard diagrammatic impurity technique. This technique has been applied to the Rashba model for the evaluation of the ISG effect,8 anisotropy magnetoresistance47, 50 and spin Hall effect.48 We review here the basic aspects by focusing on the case of the xy-bands, which is equivalent to the Bychkov-Rashba model in the 2DEG. By following the standard procedure, disorder is introduced as a random potential V (r), with zero average (cid:104)V (r)(cid:105) = 0 and white-noise correlations (cid:104)V (r)V (r)(cid:105) = niu2δ(r − r(cid:48)), with ni being the impurity concentration. By Fermi golden rule, one associates a scattering rate τ−1 = 2πniu2N0, where N0 is the single-particle density of state previously introduced in Eq. (14). We will consider the weak-disorder limit which is controlled by the small parameter (EF τ )−1, with EF the Fermi energy. In the diagrammatic impurity technique, the first step is the introduction of the irreducible self-energy in the self-consistent Born approximation for the electron Green function. 6.1 The case of the Exy bands The Green function, due to the SOC of the lowest pair of bands H SOC matrices as G = G0τ 0 + G1τ x + G2τ y and explictly reads xy of Eq. (27), can be expanded in Pauli G(
, k) = G+(
, k) + G−(
, k) 2 τ 0 − (τ xky − τ y kx) G+(
, k) − G−(
, k) 2 where and the self-energy has the form G±(
, k) =(cid:2)
− Exy ± (k) − Σ(
)(cid:3)−1 , Σ(
) = ∓ ı 2τ τ 0, (36) (37) (38) the minus and plus signs applying to the retarded (R) and advanced (A) sectors, respectively. The scattering time τ entering Eq. (38) is exactly the one required by the Fermi golden rule. It is worth noticing that the self-energy is proportional to the identity matrix in the spin space.50 Once the Green function is known, we may compute the SGC by means of the Kubo formula Figure 4. Ladder diagrams for the determination of the dressed vertex. The gray-filled triangle represents the infinite sum of diagrams, which results from repeated scattering. The solid lines with arrows are Green function propagators for electrons, whereas the dashed lines represent the operation of impurity average. σISG = 1 2π(cid:10)Tr(cid:2)SyGRJxGA(cid:3)(cid:11)dis av (39) which can be obtained from the expression (5), after averaging over the disorder configurations, represented as (cid:104). . .(cid:105)dis av. In the above the Tr . . . symbol involves all degrees of freedom, i.e. spin and space coordinates. The disorder average in Eq. (39) enters in two ways. The first is to use the disorder-averaged Green function given in Eq. (36). The second is the introduction, to lowest order in the expansion parameter (EF τ )−1, of the so-called ladder diagrams, which lead to vertex corrections. The vertex corrections procedure can be performed either for the spin or charge vertex of Eq. (39). Here we consider the vertex correction for the charge current vertex. The dressed vertex Jx obeys the Bethe -- Salpeter equation Jx = Jx + niu2(cid:88)k GR(
, k)JxGA(
, k), (40) which results from the infinite summation of ladder diagrams, as shown in Fig. 4. In terms of the dressed vertex the SGC reads σISG = 1 2π(cid:88)k tr(cid:104)SyGR(
, k)JxGA(
, k)(cid:105) , (41) where now the lower case trace symbol involves the spin degrees of freedom only. The problem is then reduced to the solution of the Bethe -- Salpeter equation (40) and to the evaluation of the bubble (41). In general the Bethe -- Salpeter equation is an integral equation. However, in the present case of white-noise disorder, the Bethe -- Salpeter equation becomes an algebraic one, even though still having a spin structure. In the appendix A we provide the details of the solution of Eq. (40), which leads to which shows that the vertex corrections exactly cancel the interband matrix elements of the charge current vertex. As a result, the evaluation of Eq. (41) leads to Jx = (−e) kx m , (42) σISG xy = eN0αxyτ, (43) which must be compared with the Drude weight evaluated in Eq. (31). 6.2 The case of the E− bands According to the analysis of appendix A, the dressed charge current vertex reads Jx = (−e)(cid:20) kx m The evaluation then of Eq. (41) leads to τ 0 − 2βkxkyτ x + β(3k2 x − k2 y)τ y + 1 4 βp2 F τ y(cid:21) . Γ(4)≡L=(c)=+(a)=+++...(b)=+++...p+q,αp,γp′+q,δp′,βϵn+ωνϵnσISG − = − 5 8 eN0γβp4 F τ, (44) which has a sign opposite to that of the Exy bands. In Eq. (44) the combination βp2 SOC, whereas γp2 F is the spin dressing factor accounting for the interactions in the original model. F plays the role of an effective 6.3 The case of the E+ bands According to the analysis of appendix A, the dressed charge current vertex reads The evaluation then of Eq. (41) leads to Jx = (−e) kx m τ 0. which shows again a change of sign with respect to that of the E− bands. Also here the combination γp2 spin dressing factor accounting for the interactions in the original model. F is the σISG + = (γp2 F )eN0α+τ, (45) 7. THE NUMERICAL APPROACH In this section we present our numerical results. The starting point is the response function defined in Eq. (5), which may be expressed as follows Ryx(ω) = 1 N (cid:88)k,p (fp − fk) (cid:104)pSyk(cid:105)(cid:104)kJxp(cid:105) ω + iη + Ep − Ek , (46) where k and p are quantum numbers labelling the eigenstates of the Hamiltonian. For instance, in the absence of disorder, the index k includes the crystal momentum, the orbital and spin degrees of freedom. The symbol fk stands for the Fermi function evaluated at the energy of the eigenstate k. In Eq. (46) N is the number of lattice sites. The numerical evaluation is performed on a finite system and then it is convenient to separate from the outset the Drude singular weight from the regular part as follows DISG = − π N (cid:88)k,p σISG reg = lim ω→0P R(cid:48)(cid:48) yx (ω) ω = − fp − fk Ep − Ek (cid:60)(cid:104)pSyk(cid:105)(cid:104)kJxp(cid:105), N (cid:88)k,p Ep − Ek (cid:61)(cid:104)pSyk(cid:105)(cid:104)kJxp(cid:105) fp − fk iη + Ep − Ek 1 (47) (48) , where (cid:60) and (cid:61) indicate the real and imaginary parts. Fig. 5 shows the behavior of the spin-orbit split gap at the Fermi surface. Inspection of Fig. 5 reveals that for µ = 0.3 eV the gap has extrema at energies ∆ ≈ 0.005 and 0.009 eV which are expected to dominate the response due to the "saddle-point" character of the corresponding states as discussed below. For µ = 0.7 , as shown in Fig. 2, all bands are occupied and all the gaps will appear in the response function. In particular the pair of bands (1,2) contributes to the gap at energies ∆ ≈ 0.015 eV, the pair of bands (3,4) at energies ∆ ≈ 0.005−0.015 eV, and the pair of bands (5,6) at energies ∆ ≈ 0. Fig. 6 shows the frequency dependence of the real and imaginary parts of the response function Ryx(ω) for the three chemical potentials µ = 0.3 eV, µ = 0.425 eV and µ = 0.7 eV. The underlying ground state is for a homogeneous system but we investigate the influence of the particle-hole lifetime parameter η. As compared with the approach discussed in Sec. 6 this mimics the inclusion of momentum relaxation without considering vertex corrections. For µ = 0.3 eV, when only the lowest pair of bands is occupied, one may interpret the observed behavior in terms of the Rashba model of Eq. (9). The low energy structure is determined by transitions across Figure 5. Size of the spin-orbit split gap around the Fermi surface for µ = 0.3 eV (a) and µ = 0.7 eV (b,c). This is obtained by determining the cut kF of each band with µ and then calculating the energy difference to the 'other' SO split band at the same kF . If the SO interaction has some significant momentum dependence around kF the gap determined for each of the two bands from a pair may slightly differ as in panel (a). The angle Θ is defined with respect to the kx-axis. the spin-orbit split gap of the same t2g band. In fact, it is exactly in this energy range that the imaginary part of Ryx(ω) develops a peak structure as discussed in Eq.(13). In the clean system, i.e. lifetime parameter η → 0, the imaginary part vanishes for energies below the minimum gap excitations and therefore the slope of R(cid:48)(cid:48) yx(ω), which determines the regular ISG response σISG yx,reg, is zero, whereas the limiting value of the real part fixes the Drude weight of the singular contribution. As shown in Fig. 6 (panel (a)) a finite η (or, similarly, a finite temperature) broadens the excitations and therefore induces a finite slope of R(cid:48)(cid:48) yx(ω) at ω = 0 leading thus to a finite ISG response. This is evidenced by making η larger in Fig. 6: the red curve is for η = 10−5, while the blue one for η = 10−3. The fact, that the ISG response vanishes for a clean system is consistent with the analysis in Ref.41 and with the discussion at the end of section 3. According to the KKR (7), the sign of the imaginary part of the ISG response function is determined by the sign of the effective Rashba SOC. The negative sign shown by the numerical evaluation of Fig. 6 agrees with the sign found for the coupling −αxy in the effective model for the lowest pair of xy bands in Eq. (31). We remind that the Drude weight is determined by the zero-frequency value of the real part, which is obtained from the imaginary part via the KKR (7). For the chemical potential µ = 0.425 eV, close to the Lifshitz point, all the bands are occupied and the imaginary part of the response function gets contributions from the interband transitions across the spin-orbit split gaps of all the pairs of bands as well as from the interband transitions involving different pairs of bands simultaneously. In panel (b) of Fig. 6 this is evidenced by showing, together with the full imaginary part (red line) also the contribution of the individual pairs of bands: (1,2) (green line), (3,4) (blue dashed line) and (5,6) (yellow line). From this we conclude that the large spectral weight at energy ω = 0.02 is due to interband transition between different pairs of bands. The green curve shows that the contribution due exclusively to the lowest pair of bands (1,2) is still around the same energy as in panel (a) and hence the behavior of this pair of bands is still 2bandscuttheFermilevelforwhichthegapsareshowninFig.2b,c.Thepairofbandscorrespondingtothelowesteigenvalues(1,2)hasanopenFermisurfaceatµ=0.7(cf.Fig.1)andtheSOgapvanishesattheantinodalpoints.ThemiddleanduppermostpairofbandshaveafiniteSOgapallaroundtheFermisurface.Fortheuppermostpairitismaximalalongthediagonaldirec-tionwhereasforthemiddlepairitisminimalalongthediagonalandattheantinodalpoints.-π/20π/2-π/20π/2-π/20π/2-π/20π/2µ=0.7 eVµ=0.425 eVµ=0.3 eVFIG.1:FermisurfacesforthechemicalpotentialsindicatedbygreylinesinFig.??.WeproceedbycalculatingtheinverseEdelsteine↵ect(IEE)whichisdefinedviatheresponsefunction[8]IEE(!)=hhJx;Syii!i(!+i⌘)(4)00,0050,01∆ [eV]band 1band 200,0050,010,0150,02∆ [eV]band 3band 3-1-0,500,51Θ/π00,0050,010,0150,02∆ [eV]a) ∆(µ=0.3 eV)b) ∆(µ=0.7 eV)c) ∆(µ=0.7 eV)bands 1,2bands 3,4bands 5,6band 1band 2FIG.2:Sizeofthespin-orbitsplitgaparoundtheFermisur-faceforµ=0.3eV(a)andµ=0.7eV(b,c).ThisisobtainedbydeterminingthecutkFofeachbandwithµandthencal-culatingtheenergydi↵erencetothe'other'SOsplitbandatthesamekF.IftheSOinteractionhassomesignificantmo-mentumdependencearoundkFthegapdeterminedforeachofthetwobandsfromapairmayslightlydi↵erasinpanel(a).Theangle⇥isdefinedwithrespecttothekx-axis.withhhJx;Syii!=1NXk,p(fpfk)hpSykihkJxpi!+i⌘+EpEk(5)whereJxisthetotalchargecurrent(obtainedfromtheusualPeierlssubstitution)andSydenotesthetotaly-polarizedspin.OnefindsJx=Xk[2t1sin(kx)+4t3cos(ky)sin(kx)]c†k,xyck,xyXk2(t1+t3)sin(kx)c†k,xzck,xzXk2t2sin(kx)c†k,yzck,yz+2iXkcos(kx)[c†k,xyck,yzc†k,yzck,xy]andforthespinSy=12Xk,,0↵=xy,xz,yzc†k,↵⌧y0ck0,↵.(6)TherealpartiscomposedofaIEE'Drude'andareg-ularpart<IEE(!)=DIEE(!)+=hhJx;Syii!!(7)withDIEE=⇡<hhJx;Syii!whichforourcleansys-temisfinite(cf.below)butisexpectedtovanishinthepresenceofdisorder.well described by the effective Rashba model of Eq.(27). On the other hand, the inset around zero energy in panel (b) shows how the low energy contribution is dominated by both the pairs of bands (3,4) and (5,6), which at this chemical potential have a small Fermi surface and a small spin-orbit split gap. Notice that the sign of the imaginary part is opposite to that of bands (1,2), indicating an opposite sign for the Drude spectral weight in the limit of vanishing lifetime parameter η in agreement with Eqs.(33) and (35). Furthermore one may notice that for the small but finite value used for the lifetime parameter η both pairs of bands yield a finite positive slope at zero frequency. Whereas the regular part at zero frequency has the sign due to the pair or pairs of bands with the lowest gap, the zero-frequency value of the real part, which is associated to the Drude spectral weight, is obtained from the integrated spectral weight of all the interband transitions. As a consequence, also interband transitions at high energy may contribute provided they have a strong spectral weight, which must compensate the big frequency denominator of the KKR relation (7). As it is apparent from panel (b), close to the Lifshitz point, the very small value of the gap of the pairs of bands (3,4) and (5,6) is sufficient to determine a positive value of the zero-frequency real part. For the chemical potential µ = 0.7 eV as well, the energy response is determined by the gap structure of all the bands (1,2), (3,4) and (5,6), as is evident from Fig. 2, even though now we are far away from the Lifshitz point. As shown in panel (c) of Fig. 6 and the inset at zero frequency, the peak coming from the smallest gap excitations at ω ≈ 0.0001 eV belongs to the pair of bands (5,6) with the smaller kF . The next higher excitation comes from the pair of bands (3,4). As a result at a finite value of the lifetime parameter, the regular part of the ISG response is finite and positive. However, in this case, in contrast to what happens close to the Lifshitz point, the opposite-in-sign spectral weight of the interband transitions at higher energies is sufficiently strong to drive the sign of the real part to a negative value. −5 eV . In panel (a), the additional blue line is for η = 10 Figure 6. Frequency dependent real (black line) and imaginary part (red line) of the spin-current correlation function Ryx(ω) evaluated for chemical potentials µ = 0.3 eV (a), µ = 0.425 eV (b) and µ = 0.7 eV (c) and lifetime parameter −3 eV and the slope of the imaginary part at ω = 0 η = 5 · 10 defines the SGC σISG reg . In panel (b), the individual contribution of the pair of bands (1,2) (green line), (3,4) (blue dashed line) and (5,6) (yellow line) is also shown. The inset details the behavior around ω = 0, dominated by the pairs of bands (3,4) and (5,6). In panel (c) the individual contribution of the different pairs of bands is shown as in panel (b). The inset around ω = 0 evidences the contribution from the pair (5,6) at two different values of the lifetime parameter −5(yellow line), whereas the inset around ω = 0.075 shows the contribution from the η = 10 pair of bands (3,4) at η = 10 −4(brown line) and η = 10 −4 (blue dashed line) and η = 10 −5(magenta line). The analysis carried out in Fig. 6 can be extended to all values of the chemical potential and the result is reported in Fig. 7. Panel (a) of Fig. 7 shows the full Drude spectral weight together with the contribution of the individual pairs of bands at T = 10 K as a function of the chemical potential. The Drude part, which is associated to the integrated imaginary part of the response, does not depend significantly on the temperature and on the lifetime parameter η (Fig. 7 is for η = 1 · 10−6 ). Close to the Γ points of all the bands one finds a negative Drude coefficient for the pair of bands (1,2) and a positive coefficient for the pairs of bands (3,4) and (5,6). This is in agreement with results of the effective model discussed in section 4. Panel (b) of Fig. 7 reports on the other hand the regular part of the ISG response as function of the chemical potential. For small lifetime parameter η = 1 · 10−6 (inset) the response is only significant around the energies where the DOS displays a van-Hove singularity. In particular, the response at low chemical potentials is sup- pressed because there the spin-orbit split gap is large and η = 1· 10−6 is not sufficient to broaden the excitations up to ω = 0. On the other hand, for η = 1 · 10−4 (main panel) one now observes a ISG response at all energies -0,00200,002<Sy, Jx>Re Im -0,02-0,0100,010,02ω [eV]-0,0200,02<Sy , Jx>Im 1,2Im 3,4-0,02-0,0100,010,02<Sy, Jx>0Im 5,6a) µ=0.3 eVc) µ=0.7 eVb) µ=0.425 eVand also the sign change upon crossing around the Lifhitz point as discussed for panel (b) of Fig. 6. Such a sign change has been also found in the experiment of Ref.26 It can also be seen that the total ISG regular response is given by the sum of the three contributions coming from the interband transitions between each of the three pairs of the spin-orbit split bands. In fact, we have seen that a finite σISG requires a broadening of the same order than the energy of the contributing low energy excitation. Therefore interband transitions between different pairs of bands cannot contribute due to their high excitation energies. Since we investigate a clean system, we also obtain a finite value for the Drude part DISG which we checked not to depend on the system size but is a robust result. In the presence of (real) disorder we expect DISG = 0 which then guarantees the stationarity of the solution. Figure 7. Top panel: Drude coefficient at T = 10 K (black) of the ISG response and the contribution of the individual bands. Middle panel: Regular part of the inverse ISG response as a function of chemical potential for lifetime parameter −4 and the contribution of the individual bands. In both panels the DOS is shown in grey for comparison. The η = 1 · 10 −3 −6 and T = 10. Lowest panel: Regular part for η = 1· 10 inset to the middle panel reports the regular part for η = 1· 10 and different temperatures. The inset to the lowest panel resolves the region around the Lifshitz point with a significant temperature dependence. Calculations have been done for a lattice with 6354 × 6354 k points. To implement the effect of disorder scattering we perform the calculation of the SGC on finite lattices. In order to reduce the finite size effects we average over twisted boundary conditions, i.e. for a Lx × Ly lattice we set with Φx,y ∈ [0, 2π] and we typically average over 50 randomly chosen (Φx, Φy). The inset to Fig. 8 demon- strates that the averaged finite lattice computation reproduces the doping dependent SGC of the 'infinite' lattice calculation. Ψ(Ri)(cid:105) = eıΦx,yΨ(Ri + Lx,y)(cid:105) 400,511,522,5-8-4048DOSDOS00,511,522,5chemical potential-0,01-0,00500,0050,01Drude coefficient bands 1,2bands 1,2bands 5,6full responsea)01200,51σIEE00,511,522,5chemical potential04812DOSDOS00,511,522,5chemical potential00,2σIEET=10 KT=300 K123456b)η=1e-6η=1e-400,511,522,5chemical potential-0,4-0,3-0,2-0,100,1σieekT=10kT=100kT=3000,500,1kT=10kT=100kT=300FIG.4:Toppanel:DrudecoecientatT=10K(black)oftheEdelsteinresponseandthecontributionoftheindividualbands.Middlepanel:RegularpartoftheinverseEdelsteinresponseasafunctionofchemicalpotentialforlifetimepa-rameter⌘=1e4eVandthecontributionoftheindividualbands.InbothpanelstheDOSisshowningreyforcompar-ison.Theinsettothemiddlepanelreportstheregularpartfor⌘=1e6eVandT=10K.Lowestpanel:Regularpartfor⌘=1e3eVanddi↵erenttemperatures.TheinsettothelowestpanelresolvestheregionaroundtheLifshitzpointwithasignificanttemperaturedependence.Calculationshavebeendoneforalatticewith6354⇥6354k-points.putingthespinateachmomentumpointoftheFSasS↵(p,kF)=Xn,,0⇤n,(p,kF)⌧↵,0n,0(p,kF)(10)wheren,0(p,kF)aretheeigenfunctionsofthesystematmomentumkFandtheindex'p'labelsthe6eigenstates.Theindicesn=xy,xz,yzandlabeltheorbitalanditsspin.Thenthechiralityofthep-thbandcanbeobtainedfrom↵(p)=asin[kF⇥S(p,kF)]ezkFS(p,kF.(11)Fig.5showsthechiralitiesforeachpairofbandsatselectedchemicalpotentialsandthecorrespondingFermisurfaces.ForthelowestpairofbandstheFermisurfacechangesfromelectrontohole-likebetweenµ=0.5andµ=0.6,however,thecorrespondingchiralitiesareal-waysconfinedto↵⇡±⇡/2withoutanysignchangein↵.Forthemiddlepairofbandsthechiralityalsostartsat↵⇡±⇡/2forlowµbutthenonaveragebecomessmallerwithincreasingchemicalpotentialandeventu-allychangessignforµ⇡1.5.Alsothechiralitiesfortheuppermostpairofbandschangessignuponincreasingthechemicalpotential.Hereweobservethataroundthediagonaldirectionthechiralitiesofbothbandsdonotjustdi↵erinsignsothatalongthediagonalsaRashba-typedescriptionfortheuppermostbandsfails.InanycasethereisnodirectionrelationbetweenthedopingdependenceofthechiralitiesandthecontributionoftheindividualpairofbandstotheEdelsteincoecientasshowninFig.4.Foreachpairofbandswecancalculatethek-spaceresolvedcontributiontotheEdelsteincoecientwhichisshowninFig.6.Forexample,theuppermostrowshowsthecontri-butionfrombands1,2atchemicalpotentialsµ=0.1,0.5,0.6.Atthelowestchemicalpotentialµ=0.1thiscontributionisnegativeandlargest(inmagnitude)aroundtheantinodalregionswithkx=0,ky=±pi.However,whentheFSbecomesholelikethisnegativecontributionisovercompensatedbyaverylargeposi-tivecontributionwhichisconfinedtotheantinodalre-gionsat(kx=±⇡,kysmall)(indicatedbythearrows)wheretheSOgapvanishes.Inthisregionthechargecurrentoperatorhasonlycontributionsfromtheassy-metrichopping⇠whereasattheantinodalregionswithkx=small,ky=±⇡)alsotheconventionalhop-pingcontributestothechargecurrentoperator.SimilarlytheotherpanelsindicatethepositiveandnegativecontributionsforthemiddleanduppermostpairofbandswhichuponintegrationyieldtheEdelsteincon-tributionfortheindividualpairofbandsasshowninFig.4.TheproblemisthatthissignchangeisnotdirectlyrelatedtoneitherachangeinchiralitynortheshapeoftheFermisurface.00,511,522,5-8-4048DOSDOS00,511,522,5chemical potential-0,01-0,00500,0050,01Drude coefficient bands 1,2bands 3,4bands 5,6full responsea)400,511,522,5-8-4048DOSDOS00,511,522,5chemical potential-0,01-0,00500,0050,01Drude coefficient bands 1,2bands 1,2bands 5,6full responsea)01200,51σIEE00,511,522,5chemical potential04812DOSDOS00,511,522,5chemical potential00,2σIEET=10 KT=300 K123456b)η=1e-6η=1e-400,511,522,5chemical potential-0,4-0,3-0,2-0,100,1σieekT=10kT=100kT=3000,500,1kT=10kT=100kT=300FIG.4:Toppanel:DrudecoecientatT=10K(black)oftheEdelsteinresponseandthecontributionoftheindividualbands.Middlepanel:RegularpartoftheinverseEdelsteinresponseasafunctionofchemicalpotentialforlifetimepa-rameter⌘=1e4eVandthecontributionoftheindividualbands.InbothpanelstheDOSisshowningreyforcompar-ison.Theinsettothemiddlepanelreportstheregularpartfor⌘=1e6eVandT=10K.Lowestpanel:Regularpartfor⌘=1e3eVanddi↵erenttemperatures.TheinsettothelowestpanelresolvestheregionaroundtheLifshitzpointwithasignificanttemperaturedependence.Calculationshavebeendoneforalatticewith6354⇥6354k-points.putingthespinateachmomentumpointoftheFSasS↵(p,kF)=Xn,,0⇤n,(p,kF)⌧↵,0n,0(p,kF)(10)wheren,0(p,kF)aretheeigenfunctionsofthesystematmomentumkFandtheindex'p'labelsthe6eigenstates.Theindicesn=xy,xz,yzandlabeltheorbitalanditsspin.Thenthechiralityofthep-thbandcanbeobtainedfrom↵(p)=asin[kF⇥S(p,kF)]ezkFS(p,kF.(11)Fig.5showsthechiralitiesforeachpairofbandsatselectedchemicalpotentialsandthecorrespondingFermisurfaces.ForthelowestpairofbandstheFermisurfacechangesfromelectrontohole-likebetweenµ=0.5andµ=0.6,however,thecorrespondingchiralitiesareal-waysconfinedto↵⇡±⇡/2withoutanysignchangein↵.Forthemiddlepairofbandsthechiralityalsostartsat↵⇡±⇡/2forlowµbutthenonaveragebecomessmallerwithincreasingchemicalpotentialandeventu-allychangessignforµ⇡1.5.Alsothechiralitiesfortheuppermostpairofbandschangessignuponincreasingthechemicalpotential.Hereweobservethataroundthediagonaldirectionthechiralitiesofbothbandsdonotjustdi↵erinsignsothatalongthediagonalsaRashba-typedescriptionfortheuppermostbandsfails.InanycasethereisnodirectionrelationbetweenthedopingdependenceofthechiralitiesandthecontributionoftheindividualpairofbandstotheEdelsteincoecientasshowninFig.4.Foreachpairofbandswecancalculatethek-spaceresolvedcontributiontotheEdelsteincoecientwhichisshowninFig.6.Forexample,theuppermostrowshowsthecontri-butionfrombands1,2atchemicalpotentialsµ=0.1,0.5,0.6.Atthelowestchemicalpotentialµ=0.1thiscontributionisnegativeandlargest(inmagnitude)aroundtheantinodalregionswithkx=0,ky=±pi.However,whentheFSbecomesholelikethisnegativecontributionisovercompensatedbyaverylargeposi-tivecontributionwhichisconfinedtotheantinodalre-gionsat(kx=±⇡,kysmall)(indicatedbythearrows)wheretheSOgapvanishes.Inthisregionthechargecurrentoperatorhasonlycontributionsfromtheassy-metrichopping⇠whereasattheantinodalregionswithkx=small,ky=±⇡)alsotheconventionalhop-pingcontributestothechargecurrentoperator.SimilarlytheotherpanelsindicatethepositiveandnegativecontributionsforthemiddleanduppermostpairofbandswhichuponintegrationyieldtheEdelsteincon-tributionfortheindividualpairofbandsasshowninFig.4.TheproblemisthatthissignchangeisnotdirectlyrelatedtoneitherachangeinchiralitynortheshapeoftheFermisurface.Figure 8. Inset: Comparison of the ISG response for a homogeneous (V0 = 0) 24 × 24 lattice (red symbols) with the calculation for a 3762 × 3762 lattice (black line). Main panel: Averaged orbital occupation as a function of chemical potential for a 24 × 24 lattice and disorder potential V0 = 0.1 eV. Further parameters: temperature T = 100 K, η = 1 · 10 −3 eV. Disorder is introduced by a random local potential V =(cid:88)i,σ Vi (xyi,σ(cid:105)(cid:104)xyi,σ + xzi,σ(cid:105)(cid:104)xzi,σ + yzi,σ(cid:105)(cid:104)yzi,σ) with Vi randomly chosen on each site in the interval [−V0, +V0]. We then compute the SGC at some specified values of the chemical potential and average over phases Φx,y and the disorder configurations. The main panel of Fig. 8 demonstrates that for V0 = 0.1 eV the averaged orbital occupations are still well defined for a given value of the chemical potential. According to the analytical results of section 6, in the presence of disorder, the SGC is positive for the pair of bands (1,2) due to the dxy orbitals (cf. Eq.(43)), is negative for the pair of bands (3,4) (cf. Eq.(44)), associated to the effective model of bands E−, finally is positive again for the pair of bands (5,6) (cf. Eq.(45)), associated with bands E+. One then would expect a double change of sign as the chemical potential enters the bottom of the different pairs of bands. The numerical analysis of the clean limit with inclusion of the effect of all the bands has shown a more complex behavior. Close to the Γ point, the behavior of the regular SG response at zero frequency of the individual bands is well described by the effective model. Instead, the Drude weight, which also includes all interband transitions, cannot be simply interpreted in terms of the individual contributions of the different pairs of bands. Fig. 9 shows the SGC of the disordered system for four different temperatures, obtained by averaging over 50 disorder configurations and over 100 phase pairs (Φx, Φy) for each disorder realization. To estimate the effective strength of the disorder, we have evaluated the frequency-dependent electrical longitudinal conductivity, whose Lorentzian lineshape allows to extract the eleastic scattering time τ , used in the analytical theory of section 6. For two chemical potentials µ = 0.2 eV and µ = 0.6 eV, below and above the Lifshitz point, the estimated scattering time is of the order of 10−2 ps, which corresponds to a level broadening of the order of 10−5 eV. In the presence of the SOC a crucial parameter is the ratio between the spin-orbit split gap and the disorder-induced broadening. Keeping in mind the typical size of the spin-orbit split gap shown in Fig. 5, one may conclude that the condition of weak scattering limit is satisfied. At zero temperature, the black line in Fig. 9 shows that the SGC changes sign twice. One sees that the two sign changes occur in a very restricted range of chemical potentials, when first the pair of bands (3,4) starts to be occupied and then also the pair of bands (5,6) becomes occupied as well. One then is tempted to associate the positive sign with the initial filling of bands (3,4) and the negative sign with the filling of bands (5,6) in agreement with the analytical results of Eqs.(44) and (45). The effect of the temperature reduces the value of the SGC. This happens when the energy scale associated with the temperature becomes larger than disorder broadening, which is the situation already at 100 K. At finite 00,10,20,30,40,5chemical potential00,10,20,30,40,5orbital densityxy-statesxz-statesyz-states00,5chem. potential-0,2-0,100,10,2σieeV0=0.1 eVV0=0 eVFigure 9. Regular part of the SGC response as a function of chemical potential. Disorder potential is V0 = 0.1 eV and calculations are performed on 24 × 24 lattices. temperature the SGC is likely to be an effective average over its value at different chemical potentials, and hence over the values associated to the different pairs of bands. As a result, at the highest temperature 300 K, there is only one sign change before the Lifshitz point. Previously42 it has been noticed that the behavior at T = 300 K is compatible with the experimental behavior of Ref.,28 whereas the sign change upon voltage reversal of the experiment of Ref.,26 performed at T = 7 K, can be interpreted as the second sign change of our T = 0 K curve. 8. CONCLUSIONS In this paper we have presented a detailed theoretical investigation of the spin galvanic effect in a multi-band model describing the electron states at a LAO/STO metallic interface. Starting from a tight-binding description, we have derived a low-energy continuum model, which well describes the original model close to the Γ point. The resulting effective Rashba-like models correspond to a linear-in-momentum SOC for the lowest and highest pair of bands while it is cubic for the middle pair of bands. For these effective models we have performed analytical calculations both in the absence and in the presence of disorder. In particular, we have used the standard diagrammatic approach of impurity technique valid in the metallic regime. We have also performed exact numerical calculations, which are in agreement with the analytical ones close to the Γ point. The main results can be summarized as follows. 1) In the absence of disorder, the SGC as a function of frequency of the driving electric field has a singular delta-like behavior reminiscent of the Drude peak in the standard optical electrical conductivity. The spectral strength associated to the delta function gets contributions from all the interband transitions and, in general, cannot simply attributed to a single pair of spin-orbit split bands. 2) The frequency-dependent SGC has also a regular contribution, which in the absence of disorder vanishes exactly at zero frequency. This regular part has a number of spectral features, whose associated frequencies correspond to the possible interband transitions. 3) A generic level-broadening mechanism leads to a finite regular part at low frequency, whose behavior is then dominated by the smallest energy interband transition. The latter then can be directly linked to a specific pair of spin-orbit split bands. A numerical calculation inevitably requires a finite level broadening and we have shown the effect of varying the size of the broadening. 4) The presence of disorder guarantees a stationary solution and introduces an intrinsic level broadening, whose effective strength we have estimated by looking at the Lorentzian lineshape of the electrical conductivity as function of frequency. 00,20,40,60,8chemical potential-3-2,5-2-1,5-1-0,500,51SGE response [-e]0 K100 K200 K300 KNote that in contrast to the SGC, the spin Hall effect for a Rashba model with linear coupling (as for the lowest xy-type bands) would vanish48 under stationary conditions and can only be sustained under special conditions, as e.g. a periodic modulation of the chemical potential.49 5) The behavior of the SGC as a function of the chemical potential shows a non monotonous behavior at zero temperature, which evolves to a monotonous one when the temperature becomes larger than the level broadening. 6) Our theoretical results are compatible with recent experiments and call for a systematic study of the voltage dependence as a function of the temperature. G. S. acknowledges support from the Deutsche Forschungsgemeinschaft under SE806/19-1. S. C. acknowledge financial support from the University of Rome Sapienza Research Project No. RM116154AA0AB1F5. ACKNOWLEDGMENTS APPENDIX A. THE BETHE -- SALPETER EQUATION FOR THE CHARGE CURRENT VERTEX In this appendix we provide a few details on the solution of the Bethe -- Salpeter equation for the vertex. We follow closely the discussion developed for the case of the Rashba 2DEG model.50 A.1 The case of the Exy bands We begin with the case of the lowest pair of bands due to the dxy orbitals. This case is practically equivalent to the standard Rashba 2DEG model. Since vertex corrections do not modify the momentum dependence of the vertex, it us useful to write the full vertex as where all the momentum dependence is limited to the bare vertex (−e)kx/m. The spin-dependent part of the vertex Γx satisfies then a new Bethe -- Salpeter equation Jx = (−e) kx m τ 0 + Γx, (49) where the effective bare vertex is defined by Γx = γx + niu2(cid:88)k GRΓxGA, γx = (−e)αxyτ y + niu2(cid:88)k GR(−e) kx m GA. (50) (51) In the above we have omitted for the sake of simplicity the explicit frequency and momentum dependence of the Green functions. To evaluate the integral over the momentum, one must use the Pauli matrix expansion of the Green function shown in Eq. (36). Because of the factor kx in the integral, only the combination GR 2 and its complex conjugate appear. As a result the integral in the right hand side of Eq. (51) is proportional to τ y and exactly cancels the first term so that the vertex γx vanishes (see Ref.48 for details) and the full vertex reduces to the standard current vertex as shown in Eq. (42). 0 GA A.2 The case of the E− bands We follow the same strategy as in the previous case. The Green function has now the form (we omit the frequency and momentum dependence for brevity) where ζ was introduced in Eq. (32). In this case the effective bare vertex reads G = G+ + G− 2 − (τ xky − τ y kx) G+ − G− 2 , ζ ζ γx = (−e)niu2(cid:88)k GR(cid:18) kx m τ 0 − 2βkxkyτ x + β(3k2 x − k2 y)τ y(cid:19) GA = (−e) (52) (53) 1 4 βp2 F τ y, which must be inserted in Eq. (50) with the form of the Green functions given by Eq. (52). In the above pF is the Fermi momentum in the absence of SOC. Given the form (53), we look for a solution of the form Γx = Γy xτ y. With this ansatz, one easily sees that the integral over the momentum in Eq. (50) yields a term proportional to τ y. As a result one has the closed equation where Γy x = γy x + IΓy x = γy x 1 − I , (54) (55) I = niu2(cid:88)k 1 2 Tr(cid:2)τ yGRτ yGA(cid:3) = 1 − 1 2(cid:28) 4β2p6 1 + 4β2p6 F ζ 2τ 2(cid:29) , F ζ 2τ 2 where (cid:104). . .(cid:105) stands for the angle average over the direction of momentum. In the weak disorder limit, τ → ∞, I = 1/2. As a result Γy x = (−e)βp2 F /2. A.3 The case of the E+ bands In this case the Green function reads G = G+ + G− 2 + (τ xky + τ y kx) G+ − G− 2 . (56) The evaluation of the effective bare vertex is similar to the case of the Exy bands with the replacement αxy → α+. As a result γx = 0 and the dressed vertex coincides with the momentum depedent part of the bare vertex. APPENDIX B. THE BETHE -- SALPETER EQUATION AT FINITE FREQUENCY For the Rashba 2DEG model (9), the Bethe -- Salpeter equation at finite frequency reads 1 2 Γx = γx + niu2(cid:88)k γx = eατ y + niu2(cid:88)k Tr(cid:110)τ yGR Tr(cid:110)τ yGR 1 2 k (ω/2)τ yΓxGA k (ω/2)(−e) k (−ω/2)(cid:111) kx m GA k (−ω/2)(cid:111). (57) (58) which has the solution The dressed vertex reads then Γx = −e γx τ /τs − iωτ , γx = (−e)α iω −iω + 1/τ τ y. When the full dressed vertex (59) is used in the Kubo formula (39) one obtains Eq. (18). Jx = (−e) kx m τ 0 + −e 1 iω α τ /τs − iωτ −iω + 1/τ τ y. (59) REFERENCES [1] E. I. Rashba, Fiz. Tverd. Tela 2, 1224 (1960) [Sov. Phys. Solid State 2, 1109 (1960)] 1 [2] [1] Yu. A. Bychkov and E. I. Rashba, Sov. Phys. - JETP Lett. 39, 78 (1984). 1 [3] Yu. A. Bychkov and E. I. Rashba, J Phys C: Solid State Phys. 17, 6039 (1984). 1, 3 [4] S. D. Ganichev, M. Trushin, and J. Schliemann, Spin orien- tation by electric current, in Handbook of Spin Transport and Magnetism, edited by E. Y. Tsymbal and I. Zutic (Chapman and Hall, Boca Raton, FL, 2016), second edition, extended. 1 [5] M. I. Dyakonov and V. I. Perel, Phys. Lett. A 35, 459 (1971). 1 [6] E. L. Ivchenko and G. E. Pikus, JETP Lett. 27, 604 (1978). 1 [7] L. E. Vorob'ev, E. L. Ivchenko, G. E. Pikus, I. I. Farbshten, V. A. Shalygin, and A. V. Shturbin, JETP Lett. 29, 441 (1979). 1 [8] V. M. Edelstein, Solid State Communications 73, 233 (1990). 1, 3, 6 [9] E. L. Ivchenko, Y. B. Lyanda-Geller, and G. E. Pikus, JETP Lett. 50, 175 (1989). 1 [10] A. G. Aronov and Y. B. Lyanda-Geller, JETP Lett. 50, 431 (1989). 1 [11] L. Levitov, Y. V. Nazarov, and G. Eliashberg, Sov. Phys. JETP 61, 133 (1985). 1 [12] S. D. Ganichev, E. L. Ivchenko, S. N. Danilov, J. Eroms, W. Wegscheider, D. Weiss, and W. Prettl, Phys. Rev. Lett. 86, 4358 (2001). 1 [13] S. D. Ganichev, E. L. Ivchenko, V. V. Belkov, S. A. Tarasenko, M. Sollinger, D. Weiss, W. Wegscheider, and W. Prettl, Nature (London) 417, 153 (2002) 1 [14] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Phys. Rev. Lett. 93, 176601 (2004). 1 [15] Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Science 306, 1910 (2004). 1 [16] V. Sih, R. C. Myers, Y. K. Kato, W. H. Lau, A. C. Gossard, and D. D. Awschalom, Nat. Phys. 1, 31 (2005). 1 [17] C. L. Yang, H. T. He, L. Ding, L. J. Cui, Y. P. Zeng, J. N. Wang, and W. K. Ge, Phys. Rev. Lett. 96, 186605 (2006). 1 [18] H. J. Chang, T. W. Chen, J. W. Chen, W. C. Hong, W. C. Tsai, Y. F. Chen, and G. Y. Guo, Phys. Rev. Lett. 98, 136403 (2007). 1 [19] B. M. Norman, C. J. Trowbridge, D. D. Awschalom, and V. Sih, Phys. Rev. Lett. 112, 056601 (2014). 1 [20] M. Luengo-Kovac, S. Huang, D. Del Gaudio, J. Occena, R. S. Goldman, R. Raimondi, V. Sih Phys. Rev. B 96, 195206 (2017). 1 [21] J. C. R. S´anchez, L. Vila, G. Desfonds, S. Gambarelli, J. P. Attan´e, J. M. D. Teresa, C. Mag´en, and A. Fert, Nat. Commun. 4, 2944 (2013). 1 [22] L. Chen, M. Decker, M. Kronseder, R. Islinger, M. Gmitra, D. Schuh, D. Bougeard, J. Fabian, D. Weiss, and C. H. Back, Nat. Commun. 7, 13802 (2016). [23] A. R. Mellnik, J. S. Lee, A. Richardella, J. L. Grab, P. J. Mintun, M. H. Fischer, A. Vaezi, A. Manchon, E.-A. Kim, N. Samarth et al., Nature (London) 511, 449 (2014). 1 [24] Y. Shiomi, K. Nomura, Y. Kajiwara, K. Eto, M. Novak, K. Segawa, Y. Ando, and E. Saitoh, Phys. Rev. Lett. 113, 196601 (2014). 1 [25] J.-Y. Chauleau, M. Boselli, S. Gariglio, R. Weil, G. de Loubens, J.-M. Triscone, and M. Viret, Europhys. Lett. 116, 17006 (2016). 1 [26] E. Lesne, S. O. Y. Fu, J. C. Rojas-S´anchez, D. C. Vaz, H. Naganuma, G. Sicoli, J.-P. Attan´e, M. Jamet, E. Jacquet, J.-M. George et al., Nat. Mater. 15, 1261 (2016). 1, 7, 7 [27] Y. Wang, R. Ramaswamy, M. Motapothula, K. Narayanapillai, D. Zhu, J. Yu, T. Venkatesan, and H. Yang, Nano Lett. 17, 7659 (2017). 1 [28] Q. Song, H. Zhang, T. Su, W. Yuan, Y. Chen, W. Xing, J. Shi, J. Sun, and W. Han, Sci. Adv. 3, e1602312 (2017). 1, 7 [29] S. Caprara, Nat. Materials 15, 1124 (2016). 1 [30] Y. Ando and M. Shiraishi, J. Phys. Soc. Jpn. 86, 011001 (2017). 1 [31] A. Soumyanarayanan, N. Reyren, A. Fert, and C. Panagopoulos, Nature (London) 539, 509 (2016). 1 [32] J. Varignon, L. Vila, A. Barth´el´emy and M. Bibes, Nat. Physics 14, 322 (2018). 1 [33] Wei Han, Y. Otani and S. Maekawa, Quantum Materials 3, 27 (2018). 1 [34] A. D. Caviglia, M. Gabay, S. Gariglio, N. Reyren, C. Cancellieri, and J.-M. Triscone Phys. Rev. Lett. 104, 126803 (2010). 1 [35] S. Hurand, A. Jouan, C. Feuillet-Palma, G. Singh, J. Biscaras, E. Lesne, N. Reyren, A. Barth´el´emy, M. Bibes, J. E. Villegas, C. Ulysse, X. Lafosse, M. Pannetier-Lecoeur, S. Caprara, M. Grilli, J. Lesueur and N. Bergeal, Sci. Rep. 5, 12751 (2015). 1 [36] K. Gopinadhan, A. Annadi, Y. Kim, A. Srivastava, B. Kumar J. Chen, J. M. D. Coey, Araindo, T. Venkatesan, Adv. Mater. 3, 1500114 (2015). 1 [37] Haixing Liang, Long Cheng, Laiming Wei, Zhenlin Luo, Guolin Yu, Changgan Zeng, and Zhenyu Zhang, Phys. Rev. B 92, 075309 (2015). 1 [38] Shanavas, K. V., Popovi´c, Z. S. and Satpathy, S. Phys. Rev. B 90, 165108 (2014). 1 [39] Z. Zhong, A. T´oth, and K. Held, Phys. Rev. B 87, 161102 (2013). 1, 2, 2 [40] Bucheli, D., Grilli, M., Peronaci, F., Seibold, G. & Caprara, S. Phys. Rev. B 89, 195448 (2014). 1 [41] K. Shen, G. Vignale, and R. Raimondi, Phys. Rev. Lett. 112, 096601 (2014). 1, 3, 7 [42] G. Seibold, S. Caprara, M. Grilli, R. Raimondi, Phys. Rev. Lett. 119, 256801 (2017). 1, 7 [43] C. S¸ahin, G. Vignale, and M. E. Flatt´e, arXiv:1804.00061. 1 [44] G. Khalsa, B. Lee, and A. H. MacDonald, Phys. Rev. B 88, 041302 (2013). 2, 2 [45] Y. Kim, R. M. Lutchyn, and C. Nayak, Phys. Rev. B 87, 245121 (2013). 2 [46] N. Scopigno, D. Bucheli, S. Caprara, J. Biscaras, N. Bergeal, J. Lesueur, and M. Grilli, Phys. Rev. Lett. 116, 026804 (2016). 2 [47] R. Raimondi, M. Leadbeater, P. Schwab, E. Caroti and C. Castellani, Phys. Rev. B 64, 235110 (2001). 6 [48] R. Raimondi and P. Schwab, Phys. Rev. B 71, 033311 (2005). 6, 8, A.1 [49] G. Seibold, S. Caprara, M. Grilli, R. Raimondi, EPL 112, 1286 (2015). 8 [50] P. Schwab and R. Raimondi, European Physical Journal B 25, 483-495 (2002). 6, 6.1, A [51] R. Raimondi, C. Gorini, P. Schwab, and M. Dzierzawa, Phys. Rev. B 74, 035340 (2006). 3, 3 [52] R. Raimondi, P. Schwab, C. Gorini, and G. Vignale, Ann. Phys. (Berlin) 524, 153 (2012). 

1103.3249

2

1103

Electron spin synchronization induced by optical nuclear magnetic resonance feedback

We predict a new physical mechanism explaining the electron spin precession frequency focusing effect observed recently in singly charged quantum dots exposed to a periodic train of resonant circularly polarized short optical pulses [A. Greilich et al, Science 317, 1896 (2007), Ref. 1]. We show that electron spin precession in an external magnetic field and a field of nuclei creates a Knight field oscillating at the frequency of nuclear spin resonance. This field drives the projection of the nuclear spin onto magnetic field to the value that makes the electron spin precession frequency a multiple of the train cyclic repetition frequency, which is the condition at which the Knight field vanishes.

cond-mat.mes-hall

cond-mat

Electron spin synchronization induced by optical nuclear magnetic resonance feedback M. M. Glazov,1 I. A. Yugova,2 and Al. L. Efros3 1Ioffe Physical-Technical Institute RAS, 194021 St.-Petersburg, Russia 2Institute of Physics, St. Petersburg State University, 198504 St.-Petersburg, Russia 3Naval Research Laboratory, Washington DC 20375, USA (Dated: October 4, 2018, Glazov-nuc-ae22.tex, printing time = 0 : 38) We predict a new physical mechanism to explain the electron spin precession frequency focusing effect recently observed in singly charged quantum dots exposed to a periodic train of resonant circularly polarized short optical pulses [A. Greilich et al, Science 317, 1896 (2007), Ref. [1]]. We show that electron spin precession in an external magnetic field and a field of nuclei creates a Knight field oscillating at the frequency of the nuclear spin resonance. This field drives the projection of the nuclear spin onto the magnetic field to the value that makes the electron spin precession frequency a multiple of the train cyclic repetition frequency, the condition at which the Knight field vanishes. PACS numbers: 78.67.Hc, 72.25.Fe, 74.25.nj An electron spin localized in a single quantum dot (QD) is a natural qubit candidate for solid state quantum information processing [2 -- 4]. However, various optical or electrical control operations on an electron spin in a QD affect the nuclear spin polarization (NSP), which was ob- served as the Overhauser shift of the electron spin pre- cession frequency in a magnetic field using various pump- probe techniques [5 -- 8]. The NSP is changed by electron- nuclear spin flip-flop processes resulting from Fermi con- tact hyperfine interactions [9, 10]. Such processes, how- ever, are suppressed in a strong magnetic field because of an approximately three orders of magnitude mismatch in energy between the electron and nuclear Zeeman split- tings. The NSP could be preserved on the time scale of hours [11], unless special energy-conserving conditions for electron spin-flip are reached [12]. Consequently, various manipulations with an electron spin by optical or electri- cal means become a main source of nuclear spin pumping, because during the action of these time dependent per- turbations spin-flip processes can occur without energy conservation [1, 8, 13 -- 17]. One of the most remarkable demonstrations of such a phenomenon is the nuclear induced frequency focus- ing (NIFF) effect that was discovered in an ensemble of singly charged QDs under excitation by a periodic train of short resonant pulses of circularly polarized light [1]. This experiment showed that the nuclei change their po- larization to values that allowed precession frequencies of all electrons in the ensemble to satisfy the phase syn- chronization conditions (PSC). These are the frequencies at which the Larmor precession period of electron spin is equal to an integer fraction of the pump pulse repetition period [18]. Why does the NSP, which changes randomly under light excitation, reach the value allowing electrons to satisfy the PSC? The authors of Ref. [1] suggested a connection with suppression of nuclear spin dynamics in such dots. Indeed, the train synchronizes the spin preces- sion of electrons satisfying PSC and makes them optically passive at the moment of pulse arrival. This significantly slows down the light-stimulated random dynamics of the NSP in these QDs, leading to the accumulation of elec- tron spins satisfying the PSC [1]. In this Letter we demonstrate that the NIFF could be the result of the Knight field feedback-stimulated nu- clear magnetic resonance (NMR). Our calculations treat the electron spin and NSP as classical vectors precess- ing around each other and an external magnetic field, and show that the NSP increases its projection onto the magnetic field monotonically with time therefore modi- fying the electron spin precession frequency. When the electron spin precession frequency has reached the PSC, the time-averaged electron spin polarization generated by the train and the corresponding Knight field causing the NSP modifications vanish. The suggested mechanism should result in much faster frequency focusing than that connected with random fluctuations of the NSP [1, 13, 17] We consider a singly negatively charged QD exposed to the train of circularly polarized pump pulses propa- gating along the structure growth axis z , arriving at the QD with the repetition period TR , and also to a trans- verse magnetic field B k ex , where ex is the unit vector along x -axis (see inset in Fig. 1a). It is assumed that the optical transition involves the excitation of a singlet X − trion with the hole spin projection on the growth axis being ±3/2 for σ+ and σ− pump pulses, respectively. The pulse duration τp is short as compared with the spin precession period in the external magnetic field and as compared with the photocreated trion lifetime. The optical selection rules are therefore the same as in the absence of a magnetic field. In the interval between the optical pulses, the electron spin interacts with the NSP, m = Pi Ii where Ii are the nuclear spins and the sum is over a mesoscopic number ( N ∼ 105 ) of nuclear spins. At equilibrium, in the studied magnetic fields, nuclei are practically unpolarized: they are randomly oriented and the NSP magnitude is controlled by random fluctuations of nuclear spin directions m ∼ √N ∼ 3 × 102 . To de- scribe this electron-nuclei interaction we treat the elec- tron spin polarization, S , and m as classical vectors [19] and adopt the box model [20 -- 22] in which the in- teraction between electrons and nuclear spins does not depend on their positions. These approximations lead to the following equations for S and m in the interval be- tween the optical pulses, (n− 1)TR 6 t < nTR , where n is the pulse number [23]: dS dt dm dt = [(Ω + αm(t)) × S(t)], = [(ω + αS(t)) × m(t)]. (1a) (1b) Here Ω = Ωex and ω = ωex are the electron and nu- clear spin precession frequencies in an external field, and α is the hyperfine coupling constant between the electron and nuclear spins in the QD. The difference of electron and nuclear magnetic moments gives ω/Ω ∼ 10−3 . The electron and nuclear spins in Eqs. (1) are coupled via an Overhauser field of NSP fluctuation acting on the elec- tron, αm , and a Knight field of the electron spin acting on the nuclei, αS . We neglect completely a slow nuclear spin relaxation connected with dipole-dipole interactions between nuclei in Eq. (1b). The dynamics of the electron and nuclear spins in the QD have several very different time-scales. Under exper- imental conditions [1] following inequalities hold: 2π Ω ≪ 2π αm . TR ≪ 2π ω ≪ 2π α . These inequalities mean (i) that electron spin dynamics in the interval between pulses occurs in the permanent field of the frozen fluctuation of NSP; and (ii) that the dynamics of NSP is controlled only by the electron spin polarization averaged over the pulse repetition period: S0 = 1 TR Z nTR (n−1)TR S(t)dt . (2) Straightforward calculation shows that 2 not affect the nuclei if the PSC is fulfilled. If the PSC is not satisfied, however, the weak Knight field, αS0 , modifies the NSP and drives its projection, mx(t) , to the value that allows Ωeff to satisfy the PSC. B (a) (b) (c) Figure 1: Time dependence of z component of the electron spin polarization Sz calculated after the train initiation (a), and after ∼ 4000 repetition periods of the train (b). Panel (c) shows electron spin precession frequency calculated numeri- cally (magenta) and analytically from Eq. (8) (black solid) and Eq. (11) (black dashed) curves. Inset to panel (a) illus- trates geometry of a single QD excitation and shows a pump pulse and an electron (red) and nuclear (black) spins. Inset to panel (c) shows the absolute value of electron spin as a function of time. Calculations were conducted for α = 0.4 , m = 23.5 , which corresponds to approximately 2200 nu- clei with spin I = 1/2 , Θ = 2π/3 , ΩTR/(2π) = 8.5 , and ω = Ω/500 . To describe this effect we need to complement Eq. (1), which describes the electron-nuclear spin dynamics in the interval between pulses, by the relationship between the electron spin polarization before, S(b) , and after, S(a) , the pump pulse, which for resonant excitation read [24] S0 = n(n · S(a)) + ΩeffTR [S(a) − n(n · S(a))] × n + Ωeff TR S(a) − n(n · S(a)) sin (Ωeff TR) [1 − cos (Ωeff TR)], (3) where S(a) is the electron spin polarization right after the excitation pulse, n = (Ω+ αm)/Ωeff is a unit vector along the effective field and Ωeff = Ω+αm ≈ Ω+αmx . One can see from Eq. (3) that the average electron spin polarization S0 transverse to n vanishes when Ωeff sat- isfies the PSC: Ωeff TR = 2πK, with K = 1, 2, . . . The longitudinal component does not vanish at the PSC due to a small deviation of n from the magnetic field direc- tion caused by the nuclear field. As we show below, the transverse components of an electron spin are required for the NSP modification. As a result, the electron spin does Q2 + 1 Q2 − 1 4 2 + S(b) z , S(a) S(a) z = x , S(a) x = QS(b) y = QS(b) y , (4) where Q = cos Θ/2 and 1 − Q2 is the probability of trion creation by the short circularly polarized pulse with area Θ . Numerical integration of Eqs. (1), which uses Eq. (4), clearly demonstrates the NIFF effect as one can see in Fig. 1. Calculations were conducted for the elec- tron spin precession frequency, which does not satisfy the PSC: ΩTR/(2π) = 8.5 , and an initial condition for the NSP, which was selected as m k ez . We exaggerated the value of α and reduced the number of nuclei from a typical value in a QD N ∼ 105 down to N ∼ 2× 103 to conduct numerically accurate calculations within reason- able computational time. Otherwise the difference in the characteristic times of electron and nuclei spin dynamics requires carrying out calculations on a timescale covering nine orders of magnitude. Figure 1(a) shows the temporal dynamics of the elec- tron spin z -component for the 6th and 7th repetition periods where the electron spin dynamics is already sta- tionary [18] but the nuclear effects have not come into play. Panel (b) shows those dynamics for the 3998th and 3999th periods when the nuclear spin precession had already taken place. One can see that the slow nuclear spin dynamics changes qualitatively the character of elec- tron spin precession in this time interval: the amplitude of electron spin polarization is strongly enhanced and reaches its maximum value 1/2, see inset in Fig. 1(c). The effect is connected with the temporal dynamics of the electron effective spin precession frequency shown in Fig. 1(c). Apart from the oscillations of frequency ω re- lated to the NSP precession, Ωeff initially grows linearly in time and then saturates at the multiple of 2π/TR . The periodic train of short pulses synchronizes the elec- tron spin precession in the QD where Ωeff satisfies the PSC, leading to complete polarization of electron as seen in Fig. 1(b). To understand physical mechanism responsible for NIFF demonstrated in Fig. 1, let us first consider the effect of the nuclear spin precession on the electron spin dynamics. Since nuclear spin precession is slow as com- pared with electron spin precession and pump pulse rep- etition periods, one can treat the electron spin dynam- ics in the interval between pulses as a precession in the static field Ω + αm (see Fig. 2a). Using procedure from Ref. [24] and Eqs. (3), (4) we derive a steady state (on the timescale of TR ) value of Sx exposed to the train of optical pulses: Sx ≡ Sx,0(t) = αmz(t)Cx/Ω, (5) Cx = − 2Q sin2 (ΩeffTR/2) + (Q − 1)2/2 (Q − 1)2 + 2(Q + 1) sin2 (ΩeffTR/2) , (a) (b) Figure 2: (a) Schematic illustration of electron spin precession in quasi-static field Ω + αm(t) (top) and temporal depen- dence of Sx (bottom). (b) Geometry of NMR induced by steady state αSy,0 , and alternating, αSx(t) , Knight fields. Bottom panel shows static and oscillating fields in the (xy) plane. 3 One can see from Eq. (5) that Sx(t) oscillates slowly with the NSP precession frequency ω . This occurs be- cause the electron spin precession axis (see Fig. 2a) is tilted from the x axis in the (xz) plane by the small angle αmz(t)/Ω . The precession leads to a non-zero Sx -projection of the electron spin, which value is pro- portional to the tilt angle oscillating at the frequency ω . The same geometrical arguments show that S0,y and S0,z are the sum of the time independent terms S0,y and S0,z and small terms oscillating at 2ω that can be neglected. As a result, the NSP, which precesses around the static field ω + αS0 slightly tilted from the x axis experiences the alternating Knight field αSx(t) (see Fig. 2b). Since Sx(t) oscillates with ω it drives the NMR and leads to slow modification of mx , as shown below. To describe the time dependence of mx(t) we need to take into account that the NMR driving field, αSx(t) , is almost parallel to the static field. At first, it creates a time-independent shift of the NSP, ¯mz , along the z axis. Indeed, in the first approximation on α : mz(t) = m⊥ cos(cid:20)ωt +Z t 0 αSx(t′)dt′(cid:21) (6) where m⊥ = pm2 − m2 x is the perpendicular compo- nent of the NSP. In the same approximation on α , Eq. (6) can be rewritten as: mz(t) ≈ m⊥ cos ωt + ¯mz , where ¯mz = −α2m2 ⊥Cx/(2ωΩ) . The analogous calcula- tion shows that ¯my = 0 . Secondly, the NMR is caused only by the component of the alternating field perpendicular to the NSP precession axis, which is equal to α(αS0/ω)Sx(t) . Averaging the x -component of Eq. (1b): dmx/dt = α(Symz − Szmy) over a sufficiently long temporal interval ∆T ≫ 1/ω ≫ 1/Ω we obtain the standard NMR expression: dmx dt = αS y,0 ¯mz = − α3Sy,0Cxm2 ⊥ 2ωΩ , (7) where the averaged my : ¯my = 0 . Generally, the right hand side of Eq. (7) depends on mx via Sy,0 and Cx dependence on Ωeff . One can neglect this dependence if Ωeff is not very close to the PSC. In this case we obtain for mx(t) : mx(t) m⊥(0) ≈ t τnf , 1 τnf = − α3m⊥(0) 2ωΩ Sy,0Cx, (8) where m⊥(0) is the initial value of the perpendicular component of the NSP. One can see from Eq. (8) that mx(t) , and consequently Ωeff , grow linearly with time. The analytical dependence Ωeff (t) shown by the solid line in Fig. 1(c) is in good agreement with results of the numerical calculations. In the case that Ωeff is close to the phase synchroniza- , one can tion condition, which is fulfilled if mx = mPSC x rewrite Eq. (7) using Eq. (3) as: (a) dmx dt = (cid:0)mx − mPSC x mτ ′ nf (cid:1)2 , where 1 τ ′ nf = α5mTR 16ωΩ2 1 + Q 1 − Q (cid:2)m2 − (mPSC x )2(cid:3) . The dynamics of mx(t) in this case is described by mx(t) = mPSC x − m τ ′ nf t − t0 , (9) (10) (11) where t0 is an arbitrary constant, chosen to merge the time dependencies given by Eqs. (8) and (11) at t ∼ τnf . Corresponding long-time asymptote of Ωeff is plotted in Fig. 1(c) by a dashed line. x x x < 0 . x Although dmx/dt = 0 at mx = mPSC only if its fluctuation δ = mx − mPSC these mx 's are only the saddle points in the mx time dependence. the points are stable if mx < mPSC For positive τ ′ nf and unstable otherwise. This means that mx returns to mPSC If δ > 0 , the fluctuation causes the deterministic growth of mx until the next PSC with larger mx is met. Including a weak nuclear spin relaxation in Eq. (1b) gives two mx = mPSC x mτnf ′/T1 , at which dmx/dt = 0 , where T1 ≫ τnf ′ is the nuclear spin relaxation time. One of these solutions, mx = mPSC x mτnf ′/T1 , is a stable point of mx(t) . A switch of the light polarization from σ+ to σ− does not change the direction of the mx growth, as can be seen from Eqs. (8) and (11). x −pmPSC x ±pmPSC Figure 3 shows the dependence of the NIFF time, τnf , defined Eq. (8) on the pump pulse area, Θ . One can see that τnf becomes extremely long for Θ ≪ 1 because electron spin orientation is inefficient and the averaged electron spin S0 is very small under these conditions. Growth of Θ increases S0 and consequently the Knight field, αS0 , which in turn shortens τnf . Further increase of τnf with Θ seen in Fig. 3 is connected with the pe- riodic dependence of S0 on Θ . The frequency focusing time τnf ∝ Ω2ω increases significantly with a magnetic field. This explains the rapid increase of τnf with Ω seen in Fig. 3. We have considered electron-nuclear spin dynamics in a single QD with a certain initial orientation of NSP. To describe a QD ensemble we average over different initial orientations of the NSP. The time dependence of the aver- age z -component of the electron spin Sz(t) is shown in Fig. 3(b),(c). The initial NSP orientations m(0) were chosen to be isotropically distributed, with the magni- tude m(0) = m = 23.5 used to describe the spin dy- namics of a single QD in Fig. 1. Figure 3(b) shows the electron spin dynamics in the absence of nuclear spin dy- namics, which is simulated using α = 0 and ω = 0 in Eq. (1b). One can see that Sz(t) partially decays be- tween the pump pulses and the phase of spin beats jumps 4 (b) (c) Figure 3: (a) Dependence of the NIFF time, τnf , on the pulse area, Θ . The three curves were calculated at the magnetic fields which give the following electron spin preces- sion frequencies: ΩTR/(2π) = 50.5 (black), 100.5 (red) and 150.5 (blue). The parameters used: α = 5 × 106 sec −1 , TR = 13 ns, ω/Ω = 10−3 and m = 126 , which could be created by 6 × 104 nuclei with spin 1/2 , were selected to keep the calculations relevant to Refs. [1, 18]. (b) and (c) Time dependence of the average z component of the elec- tron spin Sz(t) during 3997TR < t < 3999TR time inter- val. The averaging was conducted over 25 initial directions of NSP m . Panel (b) is calculated for the frozen nuclear fluc- tuation ( α = ω = 0 in Eq. (1b)), and panel (c) is calculated taking the nuclear spin precession into account ( α = 0.4 , ω = Ω/500 ). Other parameters are the same as in Fig. 1. at each repetition period. Comparison with Fig. 1(a) shows that the amplitude of Sz(t) after 4000 repetition periods is the same as at the initial precession stage. The nuclear spin precession in the external magnetic field and in the Knight field tunes up the electron spin precession frequency and leads to a very pronounced mode-locking of electron spin coherence seen in Fig. 3(c). One can see a clear rise of Sz(t) before pulse arrival. The NIFF effect also significantly increases the Sz(t) amplitude relative to those shown in Fig. 3(b). For the parameters used in this calculation αmTR/(2π) ≈ 1.5 the main contribu- tion to Sz(t) comes from the two Ωeff 's satisfying the PSC ΩeffTR/2π = 8 and 9 . We note that if the NIFF effect is a consequence of random fluctuations of the NSP as suggested in Refs. [1, 13], the rate of this process can be estimated as γn ∼ (1− Q)α2/(Ω2TR) . The current model leads to a much faster NIFF in the QD ensemble studied in Ref. [1] because 1/(γnτnf ) ∼ αm/ω ∼ 10 in these experiments. In summary, we have suggested a new physical mech- anism of the NIFF effect for the electron spin precession [1]. This mechanism leads to a monotonic shift of the electron spin precession frequency with time and allows this frequency to reach phase synchronization condition with the train repetition period much faster than in the case when the NIFF is a consequence of random fluctu- ations of the nuclear spins as was suggested earlier. Fur- ther experimental studies of the NIFF time dependence on a magnetic field and a pulse area should provide evi- dence for the suggested mechanism. Acknowledgements. Authors thank A. Braker, I.V. Ignatiev, E.L. Ivchenko, E.I. Rashba for valuable dis- cussions and RFBR, "Dynasty" Foundation -- ICFPM, Alexander von Humboldt Foundation and the Office of Naval Research for support. [1] A. Greilich et al., Science 317, 1896 (2007). [2] D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). [3] S. A. Wolf et al., Science 294, 1488 (2001). [4] D. D. Awschalom, D. Loss, N. Samarth, eds. Semicon- ductor Spintronics and Quantum Computation (Springer- Verlag, Heidlberg, 2002) [5] S. W. Brown et al., Phys. Rev. B 54 R17339 (1996). [6] J. M. Kikkawa and D. D. Awschalom, Science 287, 473 (2000). [7] F. H. L. Koppens et al., Nature, 442, 766 (2006). [8] T. D. Ladd et al., Phys. Rev. Lett. 105, 107401 (2010). [9] M. I. Dyakonov and V. I. Perel' , JETP 38, 177 (1974); in: Optical orientation, eds. F. Meier and B. P. Za- 5 kharchenja (North-Holland, Amsterdam, 1984). [10] A. Abragam, Principles of Nuclear Magnetism, Claren- don Press, Oxford, 1961. [11] D. Paget, Phys. Rev. B 25, 4444 (1982). [12] J. R. Petta et al., Phys. Rev. Lett. 100, 067601 (2008). [13] S. G. Carter et al., Phys. Rev. Lett. 102, 167403 (2009). [14] J. Danon and Y. V. Nazarov, Phys. Rev. Lett. 100, 056603 (2008). [15] M. S. Rudner and L. S. Levitov, Phys. Rev. Lett. 99, 246602 (2007). [16] E. A. Laird et al., Phys. Rev. Lett. 99 246601, (2007). [17] V. L. Korenev, preprint arXiv:1006.5144 (2010). [18] A. Greilich et al., Science 313, 341 (2006). [19] I. A. Merkulov, Al. L. Efros, and M. Rosen, Phys. Rev B 65, 205309 (2002). [20] S. M. Ryabchenko, Yu. G. Semenov, JETP 57, 825 (1983). [21] G. Kozlov, JETP 105, 803 (2007). [22] G. Chen, D. L. Bergman, and L. Balents, Phys. Rev. B 76, 045312 (2007). [23] I. A. Merkulov et al., Phys. Rev. B 81, 115107 (2010). [24] I. A. Yugova et al., Phys. Rev. B 80, 104436 (2009). 

1001.2705

2

1001

Analysis of the two dimensional Datta-Das Spin Field Effect Transistor

An analytical expression is derived for the conductance modulation of a ballistic two dimensional Datta-Das Spin Field Effect Transistor (SPINFET) as a function of gate voltage. Using this expression, we show that the recently observed conductance modulation in a two-dimensional SPINFET structure does not match the theoretically expected result very well. This calls into question the claimed demonstration of the SPINFET and underscores the need for further careful investigation.

cond-mat.mes-hall

cond-mat

Analysis of the Two Dimensional Datta-Das Spin Field Effect Transistor P. Agnihotri and S. Bandyopadhyay aDepartment of Electrical and Computer Engineering, Virginia Commonwealth University, Richmond, VA 23284, USA Abstract An analytical expression is derived for the conductance modulation of a ballistic two-dimensional Datta-Das Spin Field Effect Transistor (SPINFET) as a function of gate voltage. Using this expression, we show that the recently observed conductance modulation in a two-dimensional SPINFET structure does not match the theoreti- cally expected result very well. This calls into question the claimed demonstration of the SPINFET and underscores the need for further careful investigation. Key words: spintronics, spin field effect transistor, Ramsauer resonances PACS: 85.75.Hh, 72.25.Dc, 71.70.Ej Preprint submitted to Elsevier 1 November 2018 Even two decades after the original proposal of the Datta-Das Spin Field Effect Transistor (SPINFET) [1], the exact analytical expression for the chan- nel conductance of a two-dimensional device structure has remained some- what obscure [see the Note at the end]. Although Datta and Das proposed a two-dimensional transistor structure in their original work [1], the expres- sion they derived for the channel conductance as a function of gate voltage was based on the assumption that the carrier's wavevector component trans- verse to the direction of current flow is zero, which effectively corresponds to a one-dimensional structure. No expression was derived for the conductance modulation in a two-dimensional structure, possibly because Datta and Das realized that the conductance modulation will be severely suppressed in a two-dimensional system. Recently, a report has appeared in the literature claiming demonstration of the Datta-Das SPINFET for the first time. The claim is predicated on the fact that a conductance modulation was observed in a two-dimensional SPINFET structure as a function of gate voltage, which could be fitted exactly with the equation ∆G = Acos(cid:16)2m∗α [VG] L/2 + φ(cid:17) , (1) where α [VG] is the gate-controlled Rashba spin-orbit interaction strength in the two-dimensional channel, VG is the gate voltage, m∗ is the charge carrier's effective mass, L is the source-to-drain separation (channel length) and φ is an arbitrary phase shift. The authors of [2] measured the expected amplitude A and the quantity α [VG] in their structure independently, and then using φ as the only fitting parameter, they could fit the experimentally observed conductance modulation ∆G in their structure with Equation (1). This "fit" 2 (among others) was offered as proof that the Datta-Das transistor has been demonstrated. Ref. [2] took Equation (1) from ref. [1], not realizing that it applies only to a strictly one-dimensional channel since ref. [1] had derived it assuming that the wavevector component transverse to the direction of current flow is exactly zero 1 . Equation (1) does not hold for a two-dimensional channel since there the transverse wavevector component will not be zero. Recently, one of us pointed this out [3] and derived the correct equation for a two-dimensional channel (of finite width) assuming that only the electron energy is conserved in ballistic transport. Subsequently, it was pointed out [4] that if the width of the channel is semi-infinite so that periodic boundary conditions can be imposed along the width, then the transverse wavevector (perpendicular to the direction of current flow) is also a good quantum number and will be conserved in ballistic transport. This is reminiscent of two-dimensional coherent resonant tunneling devices of semi-infinite width, where the transverse wavevector is conserved during tunneling [5]. Conservation of the transverse wavevector greatly simplifies the equation de- rived in [3]. Additionally, low temperature and low bias conditions cause fur- ther simplification, resulting in the following simple equation for the conduc- tance modulation in a two-dimensional SPINFET: ∆G = B kF Z0 dkz"1 − k2 z k2 F# cos [Θ (kF , kz, α [VG]) L] , (2) 1 The expression derived in ref. [1] did not contain the phase shift φ, but it could arise if α [VG] 6= 0 when VG = 0. 3 where B is a constant and Θ (kF , kz, α [VG]) = − (2m∗α [VG] /2) kF + (m∗)2 α2 [VG] /4 qk2 F − k2 z . (3) Here, kz is the transverse wavevector component (along the width) and kF is the Fermi wavevector. Derivation of Equation (2) is given in Appendix I. Clearly, Equation (2) has no similarity with Equation (1). Therefore, the con- ductance modulations in the one- and the two-dimensional cases are very different. Particularly, Equation (1) predicts that the conductance modula- tion could reach 100%, whereas Equation (2) shows unambiguously that it will never reach 100% because ensmeble averaging represented by the integra- tion over the transverse wavevector component kz will dilute the modulation considerably. Only in a strictly one-dimensional channel where Equation (1) holds, the conductance modulation can be 100%, while in a two-dimensional channel, it will never be 100%. Ref. [6] has independently derived Equation (2) for a two-dimensional SPIN- FET and found that it can be approximated as ∆G ≈ B 2qπm∗α [VG] L cosh2m∗α [VG] L/2 + π/4i . (4) Equation (4) does not quite match Equation (1) either since the amplitude of the cosine function in Equation (4) is not constant, but gate-voltage de- pendent. Therefore, Equation (2) or Equation (4) cannot be reconciled with Equation (1). However, ref. [6] also found that for the particular experimental parameters of ref. [2], Equation (1) and Equation (2) yield similar curves for ∆G versus VG over the range of VG used in the experiment. This similarity is coincidental and will not be sustained over extended ranges of VG. More 4 importantly, we have found that if we use the values of m∗, α [VG], kF and L reported in [2], then the ∆G versus VG curve computed from the correct Equation (2) does not match the experimental ∆G versus VG curve reported in [2] very well. We show these two curves in Fig. 1. This disagreement between the correct theoretical result and the experimental observation casts doubt on the claimed demonstration of the Datta-Das SPINFET. We emphasize that the above disagreement however does not establish con- clusively that the Datta-Das SPINFET was not demonstrated in [2]. Instead, it casts doubt on the claimed demonstration and highlights the need for fur- ther investigation. Finally, the important question is if the observed voltage modulation was not due to the Datta-Das effect, what could it have been due to? Ref. [2] showed that the conductance modulation ∆G versus VG disap- peared if the source and drain contacts were magnetized in a direction such that they injected and detected spins parallel to the effective magnetic field caused by the Rashba interaction. The modulation reappeared when the di- rection of magnetization was rotated by 90◦ so that the injected spins became perpendicular to the effective magnetic field. This is consistent with the Datta- Das effect which relies on precession of the injected spins around the effective magnetic field caused by Rashba interaction. Since precession cannot occur if the spins are parallel to the effective magnetic field, the Datta-Das mod- ulation will disappear in that case. The precession will occur if the injected spins are perpendicular to the effective magnetic field, so that the Datta-Das effect is recovered when the contacts' magnetizations are rotated by 90◦. This observation is certainly supportive of the Datta-Das effect, but it could also be caused by other phenomena. One likely phenomenon is Ramsauer reso- nances in the channel [7] which can also give rise to a voltage modulation ∆G 5 versus VG. Ramsauer resonances are exacerbated by a magnetic field in the direction of current flow. In the experiment, when the contacts were magne- tized in the direction perpendicular to the effective magnetic field caused by the Rashba interaction, they caused a real magnetic field to appear in the channel in the direction of current flow. This could have induced Ramsauer resonances. When the direction of magnetization of the contacts was rotated by 90◦, the real magnetic field in the channel disappeared, which could have quenched or abated the Ramsauer resonances. Thus, the observed effect is also consistent with Ramsauer resonances. Consequently, further tests are required to identifythe origin of the observed conductance modulation unambiguously. The expected oscillation periods for Ramsauer resonances and the Datta-Das effect are of course very different, but since barely one oscillation period was observed in the experiment of ref. [2], it is difficult to discriminate between these two effects from the observed modulation. In summary, we have shown that the conductance modulation observed in ref. [2] cannot be fitted very well by the correct equation governing such a device, contrary to the claim of ref. [2]. Moreover, there can be alternate explanations for the origin of the observed conductance modulation of the device. Therefore, further careful study is required to resolve these controversies definitively. Note: After the submission and acceptance of this work, we became aware of a paper [M. G. Pala, M. Governale, J. Konig and U. Zulicke, Europhys. Lett., 65, 850 (2004)] which has derived an analytical expression for the channel conductance of a two-dimensional SPINFET as a function of different orienta- tions of the contacts' magnetization. That expression reduces to Equation (2) when the contacts are magnetized in the +x-direction. We thank Prof. Ulrich Zulicke for bringing this to our attention. 6 Appendix I In this Appendix, we derive the expression for the channel conductance of a two-dimensional Datta-Das SPINFET as a function of gate voltage. Consider the two-dimensional channel of a Spin Field Effect Transistor (SPIN- FET) in the x-z plane (shown in Fig. 2(a)), with current flowing in the x-direction. An electron's wavevector components in the channel are desig- nated as kx and kz, while the total wavevector is designated as kt. Note that k2 t = k2 x + k2 z as shown in Fig. 2(b). The gate terminal induces an electric field in the y-direction which causes Rashba interaction. The Hamiltonian operator describing an electron in the channel is H = x + p2 p2 2m∗ z [I] + α [VG] (σzpx − σxpz) , (5) where the p-s are the momentum operators, the σ-s are the Pauli spin matrices and [I] is the 2×2 identity matrix. Since this Hamiltonian is invariant in both x- and z-coordinates, the wavefunctions in the channel are plane wave states ei(kxx+kz z). Consequently, in the basis of these states, the Hamiltonian is H =   2k2 t 2m∗ + α [VG] kx −α [VG] kz −α [VG] kz 2k2 t 2m∗ − α [VG] kx .   (6) Diagonalization of this Hamiltonian yields the eigenenergies and the eigen- spinors in the two spin-split bands in the two-dimensional channel: 7 El = 2k2 t 2m∗ − α [VG] kt (lower band); Eu = 2k2 t 2m∗ + α [VG] kt (upper band). (7) [Ψ]l = sinθ cosθ     (lower band); [Ψ]u = −cosθ sinθ     (upper band). (8) where θ = (1/2)arctan (kz/kx). The energy dispersion relations in the two bands (one broken and the other solid) are plotted in Fig. 3. Note that an electron of energy E has two different wavevectors in the two bands given by k(1) t and k(2) t . We will assume that the source contact of the SPINFET is polarized in the +x-direction and injects +x-polarized spins into the channel under a source- to-drain bias. We also assume that the spin injection efficiency at the source is 100%, so that only +x-polarized spins are injected at the complete exclusion of −x-polarized spins. An injected spin will couple into the two spin eigenstates in the channel. It is as if the x-polarized beam splits into two beams, each corresponding to one of the channel eigenspinors. This will yield: 1 √2 1 1     = C1 sinθ cosθ     + C2 −cosθ sinθ   ,   (9) +x − polarized where the coupling coefficients C1 and C2 are found by solving Equation (9). The result is C1 = C1 (kx, kz) = sin (θ + π/4) C2 = C2 (kx, kz) = −cos (θ + π/4) 8 (10) Note that the coupling coefficients depend on kx and kz. At the drain end, the two beams recombine and interfere to yield the spinor of the electron impinging on the drain. Here, we are neglecting multiple reflection effects between the source and drain contacts in the spirit of ref. [1]. Since the two beams have the same energy E and transverse wavevector kz (these are good quantum numbers in ballistic transport), they must have different longitudinal wavevectors k(1) x and k(2) x since k(1) t 6= k(2) t . Therefore, these two beams have slightly different directions of propagation in the channel. In other words, the channel behaves like a "birefrigent" medium where waves with anti- parallel spin polarizations travel in slightly different directions. Hence, the spinor at the drain end will be: [Ψ]drain = C1   sinθ cosθ i(cid:16)k(1) x L+kzW(cid:17) + C2 e   −cosθ sinθ     i(cid:16)k(2) x L+kzW(cid:17) e = eikzW sin (θ + π/4) sinθeik (1) x L + cos (θ + π/4) cosθeik (2) x L sin (θ + π/4) cosθeik (1) x L − cos (θ + π/4) sinθeik (2) x L   ,(11)   where L is the channel length (distance between source and drain contacts) and W is the transverse displacement of the electron as it traverses the channel. Since the drain is polarized in the same orientation as the source, it transmits only +x-polarized spins, so that spin filtering at the drain will yield a trans- mission probability T2 where T is the projection of the impinging spinor on the eigenspinor of the drain. It is given by 9   T = 1 √2   1 1  sin (θ + π/4) sinθeik (1) x L + cos (θ + π/4) cosθeik (2) x L sin (θ + π/4) cosθeik (1) x L − cos (θ + π/4) sinθeik (2) x L = eikzW (cid:20)sin2 (θ + π/4) eik (1) x L + cos2 (θ + π/4) eik (2) x L(cid:21) . Here, we have assumed 100% spin filtering efficiency. Therefore, T2 = cos4 (θ + π/4)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = cos4 (θ + π/4) + sin4 (θ + π/4) + cos2(2θ)cos(ΘL), 1 + tan2 (θ + π/4) e ihk (1) x −k 2 (2) x iL(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 2   eikzW (12) (13) where Θ = k(1) x − k(2) x . From Equation (8), we get that k(1) t − k(2) t = −2m∗α [VG] /2. Expressing the wavevectors in terms of their x- and z-components, we get: rhk(1) x i2 + k2 z −rhk(2) x i2 + k2 z = −2m∗α [VG] /2, which yields Θ = k(1) x − k(2) x = −2m∗α [VG] /2k(2) hk(1) x + k(2) x i /2 t + 2 (m∗)2 α2 [VG] /4 From Equation (8), we also get that k(2) t = m∗α [VG] 2 ±vuut m∗α [VG] 2 !2 + k2 0 ≈ k0 + m∗α [VG] 2 3 2 where k0 = √2m∗E/. (14) (15) (16) . , Now, if α [VG] is small, then hk(1) results in Equation (15), we get x + k(2) x i /2 ≈ qk2 0 − k2 z. Substituting these 10 Θ = − (2m∗α [VG] /2) k0 − (m∗)2 α2 [VG] /4 = − (2m∗α [VG] /2)√2m∗E/ − (m∗)2 α2 [VG] /4 0 − k2 z qk2 q2m∗E/2 − k2 z . (17) The current density in the channel of the SPINFET (assuming ballistic trans- port) is given by the Tsu-Esaki formula: J = q Wy ∞ Z0 1 h dEZ dkz π T2 [f (E) − f (E + qVSD)] , (18) where q is the electronic charge, Wy is the thickness of the channel (in the y-direction), VSD is the source-to-drain bias voltage and f (η) is the electron occupation probability at energy η in the contacts. Since the contacts are at local thermodynamic equilibrium, these probabilities are given by the Fermi- Dirac factor. In the linear response regime when VSD → 0, the above expression reduces to J = q2VSD Wy ∞ Z0 1 h dEZ dkz π T2"− ∂f (E) ∂E # . This yields that the channel conductance G is G = ISD VSD = JWyWz VSD = q2Wz πh ∞ Z0 dEZ dkzT2"− ∂f (E) ∂E # , (19) (20) where ISD is the source-to-drain current and Wz is the channel width. Using Equations (13) and (17), we finally get that the channel conductance is G = G0 + = G0 + q2Wz 2πh q2Wz 2πh ∞ Z0 Z0 ∞ dEZ dkzcos2(2θ)cos(ΘL)"− cos(ΘL)"− dEZ dkz k2 x k2 x + k2 z ∂f (E) ∂E # ∂E # ∂f (E) 11 ≈ G0 + q2Wz 2πh ∞ Z0 dEZ dkz"1 − 2k2 z 2m∗E# cos(ΘL)"− ∂f (E) ∂E # , (21) where G0 is a constant independent of Θ and hence the gate voltage. It is easy to show that G0 = q2Wz 2πh R ∞ 0 dER dkzh1 + 2k2 2m∗Eih− ∂f (E) ∂E i. z If the temperature is low so that − ∂f (E) reduces to ∂E ≈ δ(E − EF ), then the last equation G = G0 + = G0 + q2Wz 2πh q2Wz 2πh ∞ Z0 Z0 kF dEZ dkz"1 − dkz"1 − k2 z k2 2k2 z 2m∗E# cos(ΘL)δ(E − EF ) F# cos [Θ (kF , kz, α [VG]) L] = q2Wz πh kF Z0 dkz"1 − k2 z k2 F# F (α [VG] , L, kF , kz) , (22) where F (α [VG] , L, kF , kz) = (cos2 [Θ (kF , kz, α [VG]) L] /2 + k2 z k2 F sin2 [Θ (kF , kz, α [VG]) L] /2) , (23) and kF is the Fermi wavevector. Therefore, ∆G = G − G0 = q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F# cos [Θ (kF , kz, α [VG]) L] . (24) The last equation is identical with Equation (2). 12 Appendix II In this appendix, we will derive an expression for ∆G assuming non-ideal spin injection and detection. Let us call the spin injection efficiency at the source contact ηS and the spin filtering efficiency in the drain contact ηD. If these efficiencies are less than 100%, then the probability of a +x-polarized spin being injected by the source is (1 + ηS) /2 and the probability of it being filtered at the drain is (1 + ηD) /2 when both contacts are magnetized in the +x-direction. Therefore the contribution to ∆G arising from +x-polarized injection and +x-polarized detection is [∆G]+x,+x = (1 + ηS) (1 + ηD) 4 q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F# cos [Θ (kF , kz, α [VG]) L] .(25) Next consider the situation when the source injects a +x-polarized spin, but it transmits into the -x-polarized band in the drain because spin filtering is imperfect. In this case, since the spin injected from the source is +x-polarized, Equation (11) is still valid for the spinor of the electron impinging on the drain. However, we now have to re-calculate the projection of the impinging spinor on the -x- polarized state in the drain, which will give sin (θ + π/4) sinθeik(1) x L + cos (θ + π/4) cosθeik(2) x L sin (θ + π/4) cosθeik(1) x L − cos (θ + π/4) sinθeik(2) x L  T+x,−x = = 1   1 −1 √2   eikzW cos(2θ)(cid:20)eik(2) 1 2 x L − eik(1) x L(cid:21) . 13   eikzW (26) This yields that T+x,−x2 = cos2(2θ) 2 [1 − cos(ΘL)] = 1 2 "1 − k2 z k2 0#{1 − cos [Θ (k0, kz, α [VG]) L]} . (27) Since the probability of injecting a +x-polarized spin at the source contact is (1 + ηS) /2 and the probability of its transmitting into the -x-polarized band at the drain contact is (1 − ηD) /2, the corresponding contribution to ∆G will be [∆G]+x,−x = (1 + ηS) (1 − ηD) 4 q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F#{1 − cos [Θ (kF , kz, α [VG]) L]}. (28) Now, consider the situation when the source injects a -x-polarized spin, but it transmits into the +x-polarized band in the drain. In this case, the spin injected from the source contact is -x-polarized and we will have to recalculate the spinor of the electron impinging on the drain. Equation (9) will now be replaced by 1 √2   1 −1   = C ′ 1 sinθ cosθ     + C ′ 2 −cosθ sinθ   ,   which yields C ′ 1 = sin (θ − π/4) C ′ 2 =−cos (θ − π/4) . Therefore, the spinor of the electron impinging on the drain is 14 (29) (30) [Ψ]drain = C ′ 1   sinθ cosθ i(cid:16)k(1) x L+kzW(cid:17) + C ′ 2 e   −cosθ sinθ     i(cid:16)k(2) x L+kzW(cid:17) e = eikzW sin (θ − π/4) sinθeik(1) x L + cos (θ − π/4) cosθeik(2) x L sin (θ − π/4) cosθeik(1) x L − cos (θ − π/4) sinθeik(2) x L   ,   (31) The projection of this spinor on the +x-polarized state in the drain gives  sin (θ − π/4) sinθeik(1) x L + cos (θ − π/4) cosθeik(2) x L x L − cos (θ − π/4) sinθeik(2) x L T−x,+x = = 1   1 1 √2   eikzW cos(2θ)(cid:20)eik(2) 1 2 sin (θ − π/4) cosθeik(1) x L(cid:21) . x L − eik(1)   eikzW (32) Therefore, once again, T+x,−x2 = = cos2(2θ) 2 1 2 "1 − [1 − cos(ΘL)] k2 0# {1 − cos [Θ (k0, kz, α [VG]) L]} . z k2 (33) Since the probability of injecting a -x-polarized spin at the source is (1 − ηS) /2 and the probability of its transmitting into the +x-polarized band at the drain is (1 + ηD) /2, the corresponding contribution to ∆G is [∆G]−x,+x = (1 − ηS) (1 + ηD) 4 q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F#{1 − cos [Θ (kF , kz, α [VG]) L]}. (34) 15 Finally, consider the situation when a -x-polarized electron is injected at the source and transmits into the -x-polarized band of the drain contact. In this case, Equation (31) will describe the spinor of the electron impinging on the drain and the projection of this spinor on the -x-polarized state in the drain will give T−x,−x = 1  1 −1 √2     sin (θ − π/4) sinθeik(1) x L + cos (θ − π/4) cosθeik(2) x L sin (θ − π/4) cosθeik(1) x L − cos (θ − π/4) sinθeik(2) x L = eikzW (cid:26)sin2 (θ − π/4) eik(1) x L + cos2 (θ − π/4) eik(2) x L(cid:27) .   eikzW (35) Consequently, the contribution to ∆G will be [∆G]−x,−x = (1 − ηS) (1 − ηD) 4 q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F# cos [Θ (kF , kz, α [VG]) L] .(36) Using the Principle of Superposition, the total gate voltage dependent con- ductance modulation will be ∆G =" (1 + ηS) (1 + ηD) dkz"1 − q2Wz 2πh 4 kF × Z0 (1 − ηS) (1 + ηD) 4 4 (1 + ηS) (1 − ηD) − k2 F# cos [Θ (kF , kz, α [VG]) L] z k2 − + (1 − ηS) (1 − ηD) 4 # = ηSηD q2Wz 2πh kF Z0 dkz"1 − k2 z k2 F# cos [Θ (kF , kz, α [VG]) L] . (37) Therefore, non-ideal spin injection and filtering reduces the amplitude of any non-local voltage modulation by the factor ηSηD. 16 References [1] S. Datta and B. Das, Appl. Phys. Lett., 56, 665 (1990). [2] H. C. Koo, J. H. Kwon, J. Eom, J. Chang, S. H. Han and M. Johnson, Science, 325, 1515 (2009). [3] S. Bandyopadhyay, arXiv:cond-mat/0911.0210. [4] S. Datta, private communication. [5] F. Capasso, S. Sen, F. Beltram and A. Y. Cho, in Physics of Quantum Electron Devices, Springer Series in Electronics and Photonics, Vol. 28, Ed. F. Capasso, (Springer-Verlag, Berlin-Heidelberg, 1990), Chapter 7. [6] A. N. M. Zainuddin, S. Hong, L. Siddiqui and S. Datta, arXiv:cond- mat/1001:1523. [7] M. Cahay and S. Bandyopadhyay, Phys. Rev. B., 68, 115316 (2003). 17 ) s t i n u . b r a ( G ) s t i n u . b r a ( G Fig. 1. Plots of channel conductance modulation ∆G of a two-dimensional SPIN- FET versus gate voltage VG. The solid lines are theoretical results calculated from Equation (2) where we have used the values of m∗, kF and α [VG] reported in ref- erence [2] and the points are (approximated) experimental results reported in [2]. The amplitudes of the theoretical plots are adjusted to match the experimental results as closely as possible. Note that neither the periods, nor the phases of the experimental plots agree very well with the theoretical plots. The plots are for two different channel lengths of L = 1.65 µm and 1.25 µm used in the experiments of ref. [2]. The experiments were carried out at low temperatures and biasesThe spin splitting energy in the dot specified in Fig. 2 as a function of Rashba interaction strength. 18 D D Source Drain (a) z x y kz kt (b) kx Fig. 2. (a) A two-dimensional SPINFET channel, and (b) the wavevector compo- nents in the plane of the channel. 19 Energy E (1) kt (2) k t Wavevector kt Fig. 3. Schematic representation of the dispersion relations in the two spin split bands, under the influence of the gate voltage inducing Rashba interaction in the channel. 20 

1303.1570

3

1303

Localized Many-Particle Majorana Modes with Vanishing Time-Reversal Symmetry Breaking in Double Quantum Dots

We introduce the concept of spinful many-particle Majorana modes with local odd operator products, thereby preserving their local statistics. We consider a superconductor-double-quantum-dot system where these modes can arise with negligible Zeeman splitting when Coulomb interactions are present. We find a reverse Mott-insulator transition, where the even- and odd-parity bands become degenerate. Above this transition, Majorana operators move the system between the odd-parity ground state, associated with elastic cotunneling, and the even-parity ground state, associated with crossed Andreev reflection. These Majorana modes are described in terms of one, three, and five operator products. Parity conservation results in a 4% periodic supercurrent in the even state and no supercurrent in the odd state.

cond-mat.mes-hall

cond-mat

Localized many particle Majorana fermions with vanishing time reversal symmetry breaking in double quantum dots Anthony R. Wright∗ School of Mathematics and Physics, University of Queensland, Brisbane, 4072 Queensland, Australia Menno Veldhorst† ARC Centre of Excellence for Quantum Computation and Communication Technology, School of Electrical Engineering & Telecommunications, The University of New South Wales, Sydney 2052, Australia (Dated: March 18, 2013) We introduce the concept of spinful many-particle Majorana fermions. These emergent particles are nonlocal in even operator products, but local in odd operator product s, thereby preserving their local statistics. We consider a superconductor - double quantum dot system where these modes can arise in the presence of vanishing Zeeman splitting. We find a reverse Mott-insulator transiti on, where the even parity band crosses the odd parity band. Above this transition, Majorana operators move the system between the odd parity ground state, associated with elastic co-tunneling, and the even parity ground state, associated with crossed Andreev reflection. These Majorana modes are described in terms of one, three and five op erator products. Parity conservation results in a 4π periodic supercurrent in the even state and no supercurrent in the odd state. The necessity of only small magnetic fields and the broad parameter space where Majorana fermions appear simpli fies the experimental realization greatly, paving the way towards the detection of the Majorana fermion. The prediction for the existence of Majorana fermions [1] has attracted enormous attention in condensed matter physics in recent years [2–7]. Allured by the possibility of con- structing topological qubits for quantum computation [2, 4], a plethora of schemes promising the positive identification of Majorana fermions has emerged [8–12]. Superconductors provide intrinsic electron-hole coupling resulting in cha rge- less quasiparticles, and most proposals to realize Majorana fermions are based on this principle. Standard s-wave super- conductors do not support a zero energy Majorana mode. In the presence of strong Rashba spin-orbit coupling, however, an effective spinless p-wave superconductor can arise. This idea was first recognized by Fu and Kane, who considered the interface of a superconductor and a topological insulator [5]. A topological state harnessing Majorana modes can also arise using semiconducting nanowires in the presence of Zee - man and Rashba fields [6, 7]. Experimentally, supercurrents [13], Fraunhofer patterns and Shapiro steps [14], SQUIDs [15], and zero bias conductance peaks [16] have been ob- served in topological insulator systems, which together wi th the zero bias conductance peaks in nanowire systems [17] pro - vide prospects for the observation of the Majorana fermion, but to date, no conclusive evidence has been observed. Quantum dots can also be used to realize localized Ma- jorana modes [11, 18, 19]. Quantum dot proposals include the use of Rashba coupling [11], but the presence of an anisotropic magnetic field can also mimic this effect [19]. Although these proposals are already experimentally quite promising, the strong Rashba or anisotropic magnetic fields required makes them very challenging and limits material flexibility. For example, anisotropic magnetic fields only r e- sult in spinless localized Majorana modes in fields EZ ≫ ∆, with ∆ the induced superconducting gap. Nonetheless, here we will show that in the presence of small anisotropic mag- FIG. 1: (a) Schematic representation of the device considered. Two superconductors (with phase difference φ− ) are connected via a dou- ble quantum dot with on-site Coulomb repulsion U and in the pres- ence of a magnetic field EZ,⊥ ∼ kB T . A nanomagnet introduces a localized EZ,k , which rotates the field near one of the two dots, and we define θ to be the relative angle between the local fields. (b) Tun- ing to the degeneracy of the ground state of the double quantum dot system. The color scale represents the energy gap between the lowest even and odd energy eigenstates, Eg = (ǫ1 − ǫ2 )/∆. The parameters are U = 3∆ = −3t. The degenerate ground states support many- particle Majorana fermions with local statistics, for all θ 6= 0. The system is symmetric around φ = π , but not around θ = π/2, since the presence of regular Andreev reflection distinquishes th e quantum dots having parallel and antiparallel spin axes. netic fields, B ∼ kB T , a new arena emerges: the concept of spinful many-particle Majorana fermions. These Majorana fermions are still localized in their odd operator products , and thereby preserve the local statistics of spinless proposal s [21]. We consider an s-wave superconductor - double quantum dot system as depicted in Fig.[1], where crossed Andreev re- flection (CAR) dominates elastic co-tunneling (EC), a regim e that is readily achieved [22–26]. Double electron occupanc y in a quantum dot, as we will show, inhibits, but does not pre- vent the appearance of Majorana fermions. On-site Coulomb repulsion U can be used to tune the system from the dou- ble occupancy regime, which we de fine as U = 0, towards the single electron regime, U = ∞. Upon increasing the Coulomb repulsion, we find a clear phase transition where it becomes possible to tune the even and odd parity states to be- come degenerate, see Fig.[2]. This is reminiscent of a Mott metal-insulator transition, however in reverse, where the on- site Coulomb repulsion drives the system toward having a de- generate ground state. Above this transition, we can always tune towards the degeneracy point using a second supercon- ductor with a phase difference φ− . Many-particle Majorana fermions appear right at the crossing and are described by on e, three and five operator products. The even operator products are nonlocal, while the odd operator products are local for a ll angles θ, the angle between the local fields on the dots. The requirements of only small fields and the use of quan- tum dots opens large material flexibillity. For example, the system could be based on III-V heterostructures [27], but also on materials such as silicon and diamond with large fraction s of non-magnetic nuclei which increase spin coherence times [28]. In order to elucidate the novel many-particle physics that emerges, we will take all results in the limit EZ → 0, valid as T → 0, meaning we have spinful, spin rotation symmetric, ground states. The superconductor - double quantum dot system with on- site Coulomb repulsion in the presence of an anisotropic mag - netic field, Fig.[1], is described by the Hamiltionian H = HS + HD + HT + HZ . (1) The superconducting part, HS , is HS = k,i,σ c† k,i,σ ck,i + ∆eiφi c† ǫk,i,σ c† −k,i, ¯σ . 2 Xi=1 Xk,σ Here, ¯σ denotes the spin that is not σ . The double quantum dot is described by, HD , (2) HD = 2 Xj=1 Xσ j,σ cj,σ + Xj ǫj c† with U the onsite Coulomb repulsion, and ǫj the onsite energy of the j th dot. The third term introduces the tunneling and is described by, HT , U nj↑ nj,↓ , (3) Γij c† i,σ cj,σ . HT = Xi,j,σ The overlap integral Γij is between the ith quasiparticle in the superconducting lead, and the j th dot. Finally, (4) 2 +∆ei of Ref.[29], to obtain the effective Hamiltonian of the double dot-superconductor system. Before presenting the final effective theory, we first con- sider the anisotropic magnetic field, which de fines the spin axes. The angle θ, the angle between the two local magnetic fields, modify EC and CAR as follows: tc† 1,σ c2,σ + ¯σ t sin(θ/2)c† 1,σ c2,σ → t cos(θ/2)c† 1,σ c2, ¯σ ∆c1,σ c2, ¯σ → σ∆ sin(θ/2)c1,σ c2,σ + ∆ cos(θ/2)c1,σ c2, ¯σ , (6) where σ, ¯σ = ±, and t, ∆ are the effective EC hopping and CAR Cooper pairing amplitudes [29]. The final effective Hamiltonian of the dot-superconductor system is nj,↑nj,↓ − EZ Xj (cid:0)nj,↑ − nj,↓ (cid:1) j,σ cj,σ + U Xj Hef f = Xj,σ ǫj c† + Xσ (cid:16)t cos(θ/2)c† 1,σ c2,σ + ¯σ t sin(θ/2)c† 1,σ c2, ¯σ φ− φ+ 2 (cid:1)(cid:0)σ sin(θ/2)c† 2,σ + cos(θ/2)c† 1,σ c† 1,σ c† +∆ei 2, ¯σ (cid:1) 2 cos(cid:0) φ− φ+ j, ¯σ + h.c.(cid:17), 2 (cid:1) Xj c† j,σ c† 2 cos(cid:0) (7) where φ± = φ1 ± φ2 is the sum (difference) between the phases of the two superconductors. From hereon we will as- sume that the onsite energy of the two dots has been tuned to the chemical potential of the superconductors, which we de- fine as our zero of energy ( ǫ1 = ǫ2 = 0). A discussion of the effects of the onsite energies deviating from this ‘sweet spot’ has been presented elsewhere [19]. The Hamiltonian Eq. [7] cannot be decomposed into to- tal spin sectors, as the anisotropic magnetic field mixes the se. However, the fermion parity is a conserved quantity. The to- tal Hilbert space has dimension (2nσ )nj /2 = 8 for each par- ity sector (i.e. even and odd), which makes exact diagonal- ization particularly straight-forward. In the occupation rep- resentation, we can de fine an occupation basis 1, 2i, where the numbers correspond to the two dots, and we will use ar- rows to denote the spin, and then we can construct a basis of the sixteen possible con figurations. In the U → ∞ limit, the total Fock space is restricted to nine possible states, 5 for even parity, and 4 for the odd, which can be easily solved to give the eigenstates that determine the ground state, in the even and odd sectors. We solve for finite Zeeman splitting, and obtain in the odd parity sectors that the eigenstates are ǫodd = ±t ± EZ . We take the limit EZ → 0 in the low- est energy odd parity state to obtain the limit of vanishingly small magnetic fields of the ground state even and odd parity wavefunctions HZ = −EZ Xj (cid:0)c† j,↑ cj,↑ − c† j,↓ cj,↓ (cid:1). The Hamiltonian is quadratic in the leads, and so we can inte- grate out the superconducting leads, following the procedu re (5) Ψeven = √2e 1 2 h± iφ+ 2 0, 0i + cos(θ/2)(cid:0) ↑, ↓i + ↓, ↑i(cid:1) + sin(θ/2)(cid:0) ↑, ↑i − ↓, ↓i(cid:1)i(8) 3 U=∞ U=20∆ U=0 1 0.5 0 J / J 0 -0.5 -1 0 0.5 Γ 1 φ- /π 1.5 2 FIG. 3: Josephson supercurrent of the even parity sector at T = 0.01∆ for the limiting cases U = 0 and ∞ and intermediate U = 20∆, with ∆ = t and θ = π/2. When U = 0, reg- ular Andreev reflection dominates and the Josephson current is 2π periodic and no Majorana fermions exist at all. When U → ∞, reg- ular Andreev reflection is completely forbidden, there zero energy states are local odd operator Majorana fermions, and crosse d An- dreev reflection results in a 4π periodic Josephson effect. For finite U , hybridization between the even parity eigenstates leads to a sharp transition at φ = π . For large U , Zener tunneling Γ can restore the 4π periodicity, but Majorana modes are always present. J0 = e∆/ is the maximum supercurrent at U = 0. The odd parity supercurrent is always 2π periodic, and small, vanishing completely in the limit U → ∞. and ni,σ = c† (θ) has a similar form with site i,σ ci,σ . γ U→∞ 2 indices 1, 2 interchanged. Eq. [11] has the form of a usual Majorana operator [21], except that the phase dependence is the total phase of the two superconductors divided by four, or equivalently half the average phase of the two superconduc- tors. The form of three-operator Majorana fermion in Eq. [10] is nonlocal in general. However, it is interesting to note that the relative phase of the Majorana components, which is re- sponsible in general for their non-Abelian braiding statis tics, is still localized, as the nonlocality of the Majoranas as de fined above is restricted to number operators only. It is assumed, in the above Majorana expressions, that the system is projected into the singly occupied Fock space, since U → ∞. When U is finite, a phase transition occurs at a critical value of the on-site repulsion. For Coulomb repulsions be- low this critical point, the ground state is non-degenerate , with the lowest energy even parity state always having lower en- ergy than the lowest energy odd parity state. In Fig.[2] we have plotted the excitation energy of the first excited state Eg = (ǫ2−ǫ1)/∆ as a function of onsite Coulomb repulsion, and relative spin angle θ. The phase transition corresponds to the vanishing of the excitation gap at a critical value of on-site Coulomb repulsion. This critical value varies as a function of angle, but for all angles it is quite small, being of the same order as the effective pairing ∆. For finite on-site Coulomb repulsion, the structure of the ground state, and Majorana states are much more complicated than in the in finite U case, though qualitatively do not differ. The odd parity ground state develops a finite weighting on the FIG. 2: Phase diagram showing the onset of a degenerate ground state energy gap as a function of on-site Coulomb repulsion, for ∆ = t. Upon increasing the on-site Coulomb repulsion U , a re- verse Mott-insulator transition occurs where the lowest energy state changes from even to odd parity. At the crossing point, Majorana operators move the system between the degenerate ground sta tes. with corresponding eigenvalue ǫeven = ∓√2∆ cos(φ− /2), and Ψodd = 1 √2 h sin(θ/4)(cid:0) ↑, 0i + 0, ↑i(cid:1) − cos(θ/4)(cid:0) ↓, 0i − 0, ↓i(cid:1)i, with eigenvalue ǫodd = −t (taking t > 0). A degenerate ground state is obtained when ǫeven = ǫodd . Crucially, for any t < √2∆, there is always a φ− which can be chosen, depending on the value of θ, such that a degenerate ground state can be obtained. We emphasize that as we are using the occupation number basis, the ground state is given by the lowest energy eigen- state, and not the zero energy eigenstate as in the more famil- iar Bogolubov de-Gennes theory. The two degenerate ground states have even and odd parity, respectively, and so are protected from hybridizing when th e total system conserves particle number parity. When there a re two degenerate ground states, a pair of Majorana operators (γ1 , γ2 ) can be constructed which transform the two ground states into each other, such that γ1Ψodd = Ψeven , for exam- ple. These Majorana operators are given by √2 "x(γ1↑ + γ1↓ ) + (cid:16) cos(θ/2) 1 cos(θ/4) − x(cid:17)n2↓γ1↑ (θ) = + (cid:16) sin(θ/2) sin(θ/4) − x(cid:17)n2↑γ1↑ + (cid:16) cos(θ/2) sin(θ/4) − x(cid:17)n2↑γ1↓ + x(cid:17)n2↓γ1↓ (cid:21) − (cid:16) sin(θ/2) cos(θ/4) where x = (sin(θ/4) − cos(θ/4))−1 , and γ φ+ 1,↑ = (eiφ+ /4 c† 1,↑ + e−iφ+ /4 c1,↑ ), γ U→∞ 1 (9) (10) (11) terms of the form ↑↓, ↑i, together with the three equivalent combinations of this, whilst the even parity ground state de- velops a finite weighting on doubly occupied dots ( ↑↓, 0i and 0, ↑↓i), together with the four fermion, double occupied quantum dot pair ↑↓, ↑↓i. The Majorana fermions, how- ever, retain the same phase locality condition. The Majo- rana fermions in this case acquire five fermion operator prod - ucts, together with single and three fermion operator produ cts. Generically, the Majorana has the form γ1 = γ φ+ 1,↑ (cid:0)a↑ + Xσj σj nσj + Xσσ′ jj ′ b↑ c↑ σσ′ jj ′ nσj nσ′ j ′ (cid:1)+ ↑↔↓, (12) where bσ σ′ ,σ,j,1 = 0, and when φ = π/2, σ,1 = cσ σ,σ′ ,1,j ′ = cσ all operators where j, j ′ = 2 have coefficient zero. The ap- pearance of three and five operator products are a clear gen- eralization of the Majorana fermion concept, which is usually based on single operator products [20, 21]. The regime where Majorana fermions appear can be found by analyzing the Josepshon supercurrent through the double dot system. The Josephson current is calculated from the derivative of the free energy with respect to the supercondu ct- ing phase difference φ− [30]. In contrast to topological su- perconductor systems, the odd parity ground states dispers e only very weakly with superconducting phase difference, and are strictly 2π periodic. At U → ∞, they are completely flat, as CAR cannot possibly excite the odd parity ground state. The absence of supercurrent in the odd parity state will be a strong signature of parity conservation. In Fig.[3] we have plotted the Josephson current for the even parity system as a function of superconducting phase difference for the U → ∞ case, where Majorana fermions are present (i.e. CAR splits the Cooper pairs and a gapless super-current flows between the superconductors) versus the case of regular Andreev re- flection where the final term in Eq. [7] dominates, as in the U = 0 case. The periodicity of the two cases is clearly visi- ble. We have also plotted an intermediate case, U = 20∆, which is 2π periodic in the adiabatic limit. We note that while in a p-wave superconductor the boundstates are pro- tected by fermion parity, the dispersive bound states in the double quantum dot system have both even parity, and scat- tering between the branches is possible even when the total parity is conserved. Interestingly, the matrix element coupling the branches is a two Majorana operator product. The anoma- lous current phase relationship, however, can still be obse rved in non-equilibrium measurements [31] or using dc SQUIDs [32, 33]. At finite U , the bonding and anti-bonding even parity states hybridize as φ− drives them toward a degeneracy point, nat- urally resulting in a signal where the bonding state tunnels into the anti-bonding, and one where it does not, as shown in Fig.[4]. A superposition of a 2π periodic signal and a 4π pe- riodic signal is expected in this case, since by means of Zener tunneling Γ the quasiparticle can overcome the hybridization gap [34, 35]. This transition from 4π to 2π periodicity is not 4 FIG. 4: Energy eigenstates of the double dot system for U =20∆, ∆ = t and θ = π/2. The +, − signs indicate the even par- ity ground state bonding and antibonding state, determined by the relative sign between the empty and two-fermion occupations, for example Ψ± = a0000i ± b1010i + .... At infinite U , the two states are orthogonal, but at finite U the two develop an anticross- ing. The arrow indicates the evolution of the ground state that gives a 4π periodic Josephson effect, which is still possible for small hy- bridization gaps by means of Zener tunneling Γ. As the bonding and anti-bonding states become increasingly hybridized, the 2π periodic Josephson effect becomes dominant. We emphasize that as we are using the occupation number basis, the ground state is given by the lowest energy eigenstate, and not the zero energy eigenstate as in the more familiar Bogolubov de-Gennes theory. a transition where Majorana modes disappear, but a transition where the five operator products vanish. Also, finite U results in a weakly dispersing odd parity ground state, clearly visi - ble in Fig.[4], and introduces a supercurrent in the odd pari ty state. In conclusion, we have presented an experimentally fea- sible scenario whereby many particle Majorana fermions in double quantum dots may be realized. We have shown that a large Zeeman splitting is not necessary to realize nonloca l Majorana fermions in double quantum dots, due to the onsite Coulomb repulsion, and in fact a small Zeeman field is suffi- cient. Furthermore, we have shown that by connecting two out of phase superconductors to both quantum dots, the parame- ter space can be tuned so as to obtain a degenerate ground state protected by parity, with Majorana operators connect - ing the ground state. The Majorana operators are local in the odd operators, which determine their statistics, for any an gle θ. The effect of a finite, rather than in finite, on-site interac - tion does not affect the locality conditions of the Majoranas, nor their parity. We expect that the proposed device will ex- pand Majorana physics even further, not in the least due to its spinful ground state, and the fact that the corresponding Majorana fermions are constructed from one, three, and five fermion creation/annihilation operator products, togeth er with their Hermitian conjugates. We would like to thank Ross H. McKenzie, Jacopo Saba- tini and Andrew Dzurak for enlightening discussions. ARW is financially supported by a University of Queensland Post- doctoral Reasearch Fellowship. MV is financially supported by the Australian Research Council Centre of Excellence for Quantum Computation and Communication Technology (project number CE11E0096), the US Army Research Office (W911NF-13-1-0024) and the Netherlands Organization for Scientific Research (NWO) by a Rubicon grant. ∗ Electronic address: [email protected] † Electronic address: [email protected] [1] E. Majorana, Nuovo Cimento 14, 171-184 (1937). [2] G. Moore, and N. Read, Nucl. Phys. B360, 362 (1991). [3] A. Kitaev, Ann. Phys. (N.Y.) 303, 2 (2003). [4] C. Nayak , S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, Rev. Mod. Phys. 80, 1083 (2008). [5] L. Fu and C.L. Kane, Phys. Rev. Lett. 100, 096407 (2008). [6] J.D. Sau, R.M. Lutchyn, S. Tewari, S.D. Sarma, Phys. Rev. Lett. 104, 040502 (2010). [7] J. Alicea, Phys. Rev. B 81, 125318 (2010). [8] S. Das Sarma, S. M. Freedman, and C. Nayak, Phys. Rev. Lett. 94, 166802 (2005). [9] A. R. Akhmerov, J. Nilsson, and C. W. J. Beenakker, Phys. Rev. Lett. 102, 216404 (2009). [10] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). [11] J. D. Sau and D. Das Sarma, Nature Comm. 3, 964 (2012). [12] A. R. Wright and B. Rosenow, Phys. Rev. B 86, 115329 (2012). [13] B. Sac´ep´e, J.B. Oostinga, J.L. Li, A. Ubaldini, N.J.G. Couto, E. Giannini, and A.F. Morpurgo, Nature Comm. 2, 575 (2011). [14] M. Veldhorst, M. Snelder, M. Hoek, T. Gang, X.L. Wang, V.K. Guduru, U. Zeitler, W.G. v.d.Wiel, A.A. Golubov, H. Hilgenkamp, and A. Brinkman, Nature Materials 11, 417 (2011). 5 [15] M. Veldhorst, C.G. Molenaar, X.L. Wang, H. Hilgenkamp, and A. Brinkman, Appl. Phys. Lett. 100, 072602 (2012). [16] S. Sasaki, M. Kriener, K. Segawa, K. Yada, Y. Tanaka, M. Sato, and Y. Ando, Phys. Rev. Lett. 107, 217001 (2011). [17] V. Mourik, K. Zuo, S.M. Frolov, S.R. Plissard, E.P.A.M. Bakkers, and L.P. Kouwenhoven, Science 336, 1003 (2012). [18] I.C. Fulga, A. Haim, A.R. Akhmerov, and Y. Oreg, arXiv:1212.1355 (2012). [19] M. Leijnse, and K. Flensberg, Phys. Rev. B 86, 134528 (2012). [20] C. Nayak and F. Wilczek, Nucl. Phys. B479, 529 (1996). [21] D. A. Ivanov, Phys. Rev. Lett. 86, 268 (2001). [22] J.M. Byers and M.E. Flatt ´e, Phys. Rev. Lett. 74, 306 (1995). [23] G. Deutscher and D. Feinberg, Appl. Phys. Lett. 76, 487 (2000). [24] L. Hofstetter et al., Nature 461, 960 (2009). [25] L. G. Herrmann et al., Phys. Rev. Lett. 104, 026801 (2010). [26] M. Veldhorst and A. Brinkman, Phys. Rev. Lett. 105, 107002 (2010). [27] R. Hanson, L.P. Kouwenhoven, J.R. Petta, S. Tarucha, and L.M.K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). [28] F. Zwanenburg, A.S. Dzurak, A. Morello, M.Y. Simmons, L.C.L. Hollenberg, G. KLimeck, S. Rogge, S.N. Coppersmith, and M.A. Eriksson arXiv arXiv:1206.5202v1. [29] A. V. Rozhkov and D. P. Arovas, Phys. Rev. Lett. 82, 2788 (1999). [30] S. Droste, S. Andergassen, and J. Splettstoesser, J. Phys.: Con- dens. Matter 24, 415301 (2012). [31] D.M. Badiane, M. Houzet, and J.S. Meyer, Phys. Rev. Lett. 107, 177002 (2011). [32] M. Veldhorst, C.G. Molenaar, C.J.M. Verwijs, H. Hilgenkamp, and A. Brinkman, Phys. Rev. B 86, 024509 (2012). [33] M. Veldhorst, M. Snelder, M. Hoek, C.G. Molenaar, D.P. Leusink, A. A. Golubov, H. Hilgenkamp, and A. Brinkman, Phys. Stat. Sol. RRL 7, 26 (2013). [34] A. Jacobs, R. Kummel, Phys. Rev. B 71, 184504 (2005). [35] H. Kroemer, Superlattices Microstruct. 25, 877 (1999). 

1507.01410

1

1507

Helicity sensitive terahertz radiation detection by dual-grating-gate high electron mobility transistors

We report on the observation of a radiation helicity sensitive photocurrent excited by terahertz (THz) radiation in dual-grating-gate (DGG) InAlAs/InGaAs/InAlAs/InP high electron mobility transistors (HEMT). For a circular polarization the current measured between source and drain contacts changes its sign with the inversion of the radiation helicity. For elliptically polarized radiation the total current is described by superposition of the Stokes parameters with different weights. Moreover, by variation of gate voltages applied to individual gratings the photocurrent can be defined either by the Stokes parameter defining the radiation helicity or those for linear polarization. We show that artificial non-centrosymmetric microperiodic structures with a two-dimensional electron system excited by THz radiation exhibit a dc photocurrent caused by the combined action of a spatially periodic in-plane potential and spatially modulated light. The results provide a proof of principle for the application of DGG HEMT for all-electric detection of the radiation's polarization state.

cond-mat.mes-hall

cond-mat

Helicity sensitive terahertz radiation detection by dual-grating-gate high electron mobility transistors P. Faltermeier,1 P. Olbrich,1 W. Probst,1 L. Schell,1 T. Watanabe2, S. A. Boubanga-Tombet2, T. Otsuji2, and S. D. Ganichev1 1 Terahertz Center, University of Regensburg, 93040 Regensburg, Germany and 2 Research Institute of Electrical Communication, Tohoku University, 980-8577 Sendai, Japan We report on the observation of a radiation helicity sensitive photocurrent excited by terahertz (THz) radiation in dual-grating-gate (DGG) InAlAs/InGaAs/InAlAs/InP high electron mobility transistors (HEMT). For a circular polarization the current measured between source and drain contacts changes its sign with the inversion of the radiation helicity. For elliptically polarized radiation the total current is described by superposition of the Stokes parameters with different weights. Moreover, by variation of gate voltages applied to individual gratings the photocurrent can be defined either by the Stokes parameter defining the radiation helicity or those for linear polarization. We show that artificial non-centrosymmetric microperiodic structures with a two- dimensional electron system excited by THz radiation exhibit a dc photocurrent caused by the combined action of a spatially periodic in-plane potential and spatially modulated light. The results provide a proof of principle for the application of DGG HEMT for all-electric detection of the radiation's polarization state. PACS numbers: 78.67.De,07.57.Kp,85.30.Tv,85.35.-p I. INTRODUCTION (FETs) have Field-effect-transistors emerged as promising devices for sensitive and fast room tempera- ture detection of terahertz (THz) radiation [1, 2]. They are considered as a good candidate for real-time THz imaging and spectroscopic analysis [3, 4] as well as future THz wireless communications [5]. Devices employing plasmonic effects in FETs have already been applied for room temperature detection of radiation with frequencies from tens of GHz up to several THz and enable the com- bination of individual detectors in a matrix. They are characterized by high responsivity (up to a few kV/W), low noise equivalent power (down to 10 pW/√Hz), fast response time (tens of picoseconds) and large dynamic range (linear power response up to 10 kW/cm2), see e.g. Ref.[2, 6 -- 11]. The operation principle of FET THz de- tectors used so far is based on the nonlinear properties of the two-dimensional (2D) plasma in the transistor chan- nel. The standard Dyakonov-Shur model [12] assumes that radiation is coupled to the transistor by an effective antenna, which generates an ac voltage predominantly on one side of the transistor. Both resonant [13] and non-resonant [14] regimes of THz detection have been studied. While research aimed to development of THz FET detectors is focused on single gate structures re- cently several groups have shown that higher sensitivi- ties are expected for structures with periodic symmetric and asymmetric metal stripes or gates [9, 15 -- 23]. In par- ticular, dual-grating-gate FET are considered as a good candidate for sensitive THz detection. The first data obtained on dual-gated-structures demonstrated a sub- stantial enhancement of the photoelectric response and an ability to control detector parameters by variation of individual gate bias voltage [9]. At the same time, THz electric field applied to FETs with asymmetric periodic dual gate structure is expected to give rise to electronic ratchet effects [23 -- 27] (for review see [25]) and plasmonic ratchet effects [28]. Besides improving the figure of merits of FET detectors, ratchet effects may also result in new functionalities. In particularly, they may induce pho- tocurrents driven solely by the radiation helicity. Here, we report on the observation of a radiation he- licity sensitive photocurrent excited by THz radiation in dual-grating-gate InAlAs/InGaAs/InAlAs/InP high electron mobility transistors (HEMT). We show that arti- ficial non-centrosymmetric microperiodic structures with a two-dimensional electron system excited by THz ra- diation exhibit a dc photocurrent caused by the lateral asymmetry of the applied static potential and terahertz electric field. We demonstrate that depending on gate voltages applied to the individual gratings of the dual- grating-gate the response can be proportional to either the Stokes parameters [29] defining the radiation helicity or those for linear polarization. As an important result, for a wide range of gate voltages we observed a pho- tocurrent jC being proportional to the radiation helicity Pcirc = (Iσ+ − Iσ− )/(Iσ+ + Iσ− ), where Iσ+ and Iσ− are intensities of right- and left-handed circularly polarized light. For the circular photocurrent jC measured between source and drain contacts changes its sign with the inver- sion of the radiation helicity. This observation is of par- ticular importance for a basic understanding of plasmon- photogalvanic and quantum ratchet effects. It also has a large potential for the development of an all-electric de- tector of the radiation's polarization state, which was so far realized applying less sensitive photogalvanic effects only [30 -- 32]. The observed phenomena is discussed in the framework of electronic ratchet [22, 23, 25 -- 27] and plasmonic ratchet effects excited in a 2D electron system with a spatially periodic dc in-plane potential [9, 22, 28]. II. EXPERIMENTAL TECHNIQUE 2 The based on device is high-electron an In- structure mobility AlAs/InGaAs/InAlAs/InP transistor (HEMT) and incorporates doubly inter- digitated grating gates (DGG) G1 and G2. A sketch and a photograph of the gates are shown in Fig. 1(a) and inset in Fig. 1(b). The 2D electron channel is formed in a quantum well (QW) at the heterointerface between a 16 nm-thick undoped InGaAs composite channel layer and a 23 nm-thick, Si-doped InGaAs carrier-supplying layer. The electron density of the 2DEG is about 3 × 1012 cm2, electron effective mass normalized on free- electron mass m0 and room temperature mobility are m/m0 = 0.04 and µ0 = 11000 cm2/(Vs), respectively. The DGG gate is formed with 65 nm-thick Ti/Au/Ti by a standard lift-off process. The footprint of the narrower gate fingers G1 was defined by an E-beam lithography, whereas that of the wider gate fingers G2 was defined by a photolithography. In all studied structures, the metal fingers of the grating gates G1 and G2 have the same length, being dG1 = 200 nm and dG2 = 800 nm. The spacing between narrow and wide DGG fingers is asymmetric with aG1 = 200 nm and aG2 = 400 nm, see Fig. 1. The size of the active area, covered with the grating is about 20 µm×20 µm. Ohmic contacts, forming source and drain of HEMTs, were fabricated by highly doped 15 nm thick InAlAs and InGaAs layers. The axis along the gate's fingers is denoted as x and that along source and drain as y. The characteristic source/drain current - gate voltage dependence obtained by transport measurement is shown for sample #A in Fig. 1(b). All experiments are performed at room temperature. The HEMT structures were illuminated with polarized THz and microwave (MW) radiation at normal inci- dence. For optical excitation we used low power cw optically pumped CH3OH THz laser [33, 34] and Gunn diodes providing monochromatic radiation with frequen- cies f = 2.54 THz and 95.5 GHz, respectively. The radia- tion peak power P , being of the order of several milliwats at the sample's position, has been controlled by pyro- electric detectors and focused onto samples by parabolic mirrors (THz laser) or horn antenna (Gunn diode). The spatial beam distribution of THz radiation had an al- most Gaussian profile, checked with a pyroelectric cam- era [35, 36]. THz laser radiation peak intensity, I, for laser spot being of about 1.2 mm diameter on the sam- ple, was I ≈ 8 W/cm2. The profile of the microwave radiation and, in particular, the efficiency of the radia- tion coupling to the sample couldn't be determined with satisfactory accuracy. Thus, all microwave data are given FIG. 1: (a) Sketch of the dual-grating-gate HEMT. Cross- section of the structure shows the layer sequence and indicates the width of the fingers (d1/2) and the fingers spacings (a1/2). THz radiation at 2.54 THz is applied at normal incidence. (b) Drain-to-source current as a function of the gate voltage UG1 measured at UG2 = 0 V. Inset shows the photograph of the structure. Here G1/G2, S and D denote first/second gate, source and drain, respectively. Part of G1/G2 structure is highlighted by yellow lines for visualization. in arbitrary units. The polarization state of THz radia- tion has been varied applying crystal quartz λ/4- or λ/2- plates [37]. To obtain circular and elliptically polarized light the quarter-wave plate was rotated by the angle, ϕ, between the initial polarization plane and the optical axis of the plate. The radiation polarization states for several angles ϕ are illustrated on top of Fig. 2. Orienta- tion of the linearly polarized radiation is defined by the azimuth angle α, with α = ϕ = 0 chosen in such a way that the electric field of incident linearly polarized light is directed along x-direction. Different orientation of lin- early polarized MW radiation were obtained by rotation of a metal wire grid polarizer. The photocurrent excited between source and drain is measured across a 50 Ω load resistor applying the standard lock-in technique. III. PHOTOCURRENT EXPERIMENT Illuminating the structure with elliptically (circular) polarized radiation of terahertz laser operating at fre- quency f = 2.54 THz we observed a dc current strongly depending on the radiation polarization. Figure 2(a) shows the photocurrent as a function of the phase an- gle ϕ defining the radiation polarization state. The data are obtained for zero gate voltage at the gate 2, UG2 = 0 3 FIG. 3: (a) THz radiation induced normalized photocurrent jy/I as a function of the angle ϕ defining the radiation helic- ity. The current is measured for comparable voltages applied to the first (UG2 = −1.1 V) and the second (UG2 = −0.92 V) gates. Full line shows fit to the total current calculated af- ter Eq. (1). The ellipses on top illustrate the polarization states for various ϕ. Right inset shows amplitudes of pho- tocurrent contributions jC/I, driven by the light helicity, and j1/I (j2/I), induced by linear polarization, as a function of the gate voltage UG1 or UG2. Upper inset schematically shows corresponding gate potentials. Dashed lines are guide for the eye indicating the potential asymmetry in y-direction. (b) shows amplitudes of the photocurrent contributions jC/I, driven by the light helicity, as a function of the gate voltage UG1 (UG2 = 0) measured for three different structures #A, #B, and #C. The inset shows photovoltage measured in sam- ple #D across 50 Ω load resistance (RL ≪ Rs) and directly from the sample over the lock-in amplifiers input resistance being much larger than the sample resistance Rs. Note that the former signal is multiplied by factor 25. rameters change after cos 4ϕ + 1 s0 ≡ Ex2 + Ey2, s1 ≡ Ex2 − Ey2 = xEy = s2 ≡ ExE ∗ xEy) = −Pcirc = − sin 2ϕ , s3 ≡ i(ExE ∗ y + E ∗ y − E ∗ 2 , sin 4ϕ , 2 (2) (3) (4) (5) Here s0 determines the radiation intensity, s1 and s2 de- FIG. 2: THz radiation induced normalized photocurrent jy/I as a function of the angle ϕ defining the radiation helicity. The current is measured for different voltages applied to the first and second gates. (a) shows the data for UG1 = −1.06 V at gate 1 and zero gate voltage at gate 2. (b) shows the photocurrent measured for zero gate voltage at gate 1 and UG2 = −0.9 V. Full lines show fits to the total current calcu- lated after Eq. (1). The ellipses on top illustrate the polariza- tion states for various ϕ. Insets show amplitudes of photocur- rent contributions jC/I, driven by the light helicity, and j1/I (j2/I), induced by linear polarization, as a function of the gate voltages UG1 or UG2. Second set of the insets schemat- ically show corresponding gate potentials. Dashed lines are guide for the eye indicating the potential asymmetry in y- direction. Note that presence of the metal gates results in a nonzero potential even for UG = 0. and UG1 = -1.06 V. The principal observation is that for right- (σ+) and left-handed (σ−) polarizations, i.e., for ϕ = 45◦ and 135◦, the signs of the photocurrent jy are opposite. The overall dependence jy(ϕ) is well described by jy(ϕ) = j0s0 + j1s1(ϕ) + j2s2(ϕ) + jCs3(ϕ) , (1) and corresponds to the superposition of the Stokes pa- rameters with different weights given by the coefficients j0, j1, j2, and jC, which in the experimental geometry applying rotation of quarter-wave plate the Stokes pa- 4 circuit configuration the measured photovoltage increases at larger negative bias voltages and achieves maximum at the threshold voltage, Uth = −1.3 V. Corresponding data will be presented and discussed below. While the non- monotonic behavior of the signal for gate voltage varia- tion is well known for FET detectors [1, 2, 39] the signal sign inversion upon a change of the radiation polariza- tion, see Fig. 2(a), is generally not expected for stan- dard Dyakonov-Shur FET detectors indicating crucial role of the lateral superlattice in the photocurrent gen- eration. To demonstrate that the observed effect indeed stems from the lateral asymmetry of the periodic poten- tial we interchanged the voltages applied to the gates. Figure 2(b) shows the results obtained for zero gate volt- age at the first gate and UG2 = −0.9 V at the second one. The figure reveals that changing the sign of the lat- eral potential asymmetry, see insets of Fig. 2(a) and (b), results in the sign inversion of all contributions besides the polarization independent offset. The situation holds for almost all values of UG2, see the insets in Fig. 2(a) and (b). Significantly, the proper choice of the relation between amplitudes of the individual gate potentials al- lows one to suppress completely one or the other pho- tocurrent contribution. Figure 3(a) demonstrates that for close values of gate voltages the circular photocurrent vanishes (corresponding potential profile for UG1= -1.1 V and UG2= -0.9 V is shown in the inset in Fig. 3). The interplay of the contributions upon variation of UG1 and for fixed UG2= -1.1 V is shown in the inset in Fig. 3(a). It is seen that for nonzero second gate voltage the circu- lar, jc, and linear, j2, photocurrent contributions change their direction with increasing UG1. Moreover, the inver- sions take place at different UG1 voltages. This fact can be used to switch on and off the circular photocurrent jC ∝ Pcirc contribution. To support the conclusion that j1 and j2 photocurrent contribution are caused by the linear polarized light com- ponent we carried out additional measurements applying linearly polarized light. The gate dependence of the nor- malized photocurrent jy/I measured for samples #A and #B for several azimuth angles α are shown in Fig. 4(a). The inset in this figure presents the dependence of jy/I on the electric field orientation. The polarization depen- dence is well described by the Eq. (1) taking into account that for linearly polarized light the last term vanishes and the Stokes parameters are given by s1(α) = cos 2α , s2(α) = sin 2α . Here α = 2β defines the orientation of the polarization plane and β is the angle between the initial polariza- tion plane and the optical axis of the half-wave plate. The magnitudes and signs of the coefficients j0, j1, and j2 used for the fit coincide with that applied for fitting of ϕ-dependencies obtained at the same gate voltages. These results demonstrate that photocurrents j1 and j2 measured in set-up applying quarter-wave plate are in- FIG. 4: (a) THz radiation induced normalized photocurrent jy/I excited by linearly polarized THz radiation in samples #A and #B as a function of the gate voltage UG1. The current is shown for UG2 = 0 and several in-plane orientations of the radiation electric field in respect to source-drain line defined by azimuth angles α. Inset shows dependence of jy on the angle α obtained for UG1 = −1.08 V and UG2 = 0. Full line shows fit to the total current calculated after Eq. (6). Arrows indicate electric field orientation for several angles α. (b) Photocurrent jy/I excited by linearly polarized microwave radiation (f = 95.5 GHz) in samples #A and #B as a function of the gate voltage UG1 (UG2 = 0). Inset shows dependence of jy/I on the azimuth angle α obtained in sample #B for UG1 = −1.14 V and UG2 = 0. Full line shows fit after jy ∝ cos2(α + θ) with the phase angle θ. fine the linear polarization of radiation in the (xy) and rotated by 45◦ coordinate frames, and s3 describes the degree of circular polarization or helicity of radiation. Consequently individual photocurrent contributions in Eq. (1) are induced by unpolarized, linearly or circularly polarized light components. While the polarization de- pendence given by Eq. (1) has been detected for arbitrary relations between voltages applied to the first and second gates, the magnitude and even the sign of the individual contributions can be controlled by the gate voltages. The inset in Fig. 2(a) shows a gate dependence of the polar- ization dependent contributions to the total photocur- rent [38]. The dependence on the gate voltage UG1 is obtained for zero biased second gate. Photocurrent mea- sured in the close circuit configuration with RL ≪ Rs shows a maximum amplitude for UG1 = −1.1 V. For open deed controlled by the degree of linear polarization of elliptically polarized radiation. The polarization sensitive photocurrent has been ob- served in all studied devices of similar design and ar- bitrary relation between second and first gate poten- tials. The photocurrent can always be well described by Eq. (1). Figure 3(b) summarizes the data on the he- licity driven photocurrent jC/I detected in three HEMT structures upon change of UG1 and for UG2 = 0. In all samples we detected similar dependencies of the pho- tocurrent characterized by close maximum positions but different signal magnitudes. The data of Fig. 3(b) as well as circles in its inset are obtained in the close cir- cuit configuration applying 50 Ω load resistance. The non-monotonic behavior of the photosignal measured in this geometry is caused by the interplay of the potential asymmetry, increasing with raising second gate voltage, and raising of the sample resistance for large gate volt- ages. For the open circuit geometry (signal is fed to the high input impedance of lock-in amplifier) the maximum of the signal is detected for gate voltages being equal to the threshold voltage, Uth, see squares in the inset in Fig. 3(b). Following Ref. [9] we estimate from the voltages measured in open circuit geometry the voltage responsivities for the signals corresponding to the pho- tocurrents j2 and jC as Rv = Us/P × S/St ≈ 0.3 V/W and 0.15 V/W, respectively. Here P the total power of the source at the detector plane, S radiation beam spot area, and St = 20 × 20 µm2 transistor area. The voltage responsivities, being rather low as compared to that typi- cally obtained for plasmonic FET detectors, indicates the necessity of further optimization of the structure design. Finally, we note that measurements applying microwave radiation show that for lower frequencies the polariza- tion behavior changes qualitatively. Instead of the sign- alternating dependencies discussed above the signal now varies after jy ∝ cos2(α + θ), see inset in Fig. 4(b). This observation is in a good agreement with the Dyakonov- Shur theory [12] and was reported for many conventional plasmonic FET detectors, see e.g. [1, 2]. The gate volt- age dependence of the response shown in Fig. 4 also re- produces well the results previously obtained for similar structures [9, 40]. Even the fact that the maximum of the signal in various structures has been obtained for dif- ferent directions of the electric field vector in respect to y-direction (source-drain) has already been reported for these transistors and attributed to the antenna coupling of MW radiation to transistor, see Ref. [40]. IV. DISCUSSION The observation of the circular photocurrent and the sign-alternating linear photocurrent j2 reveals that a mi- croscopic process actuating these photocurrents goes be- yond the plasmonic Dyakonov-Shur model typically ap- 5 plied to discuss operation of FETs THz detectors. In- deed, as addressed above, the latter implies an oscillating electric field along source-drain direction (y-direction) yielding sign conserving variation upon rotation of po- larization plane, jy ∝ cos2 α [41]. As recently shown in Ref. [42, 43], the Dyakonov-Shur model in fact may result in the circular photocurrent but only due to in- terference effects of two different channels and two inter- acting antennas in small size special design FETs - the model which can hardly be applied to the large DGG samples used in our experiments. At the same time, the observed polarization behavior is characteristic for the electronic ratchet effects excited in asymmetric pe- riodic structures [24 -- 27] and linear/circular plasmonic ratchet effects [22, 28]. The ratchet currents arise due to the phase shift between the periodic potential and the periodic light electric field resulting from near field diffraction in a system with broken symmetry. Micro- scopic theory developed in Ref. [26] shows that the helic- ity dependent photocurrent appear because the carriers in the laterally modulated quantum wells move in two directions and are subjected to the action of the two- component electric field. Symmetry analysis of the pho- tocurrent shows that in our DDG structures described by C1 point group symmetry [44] it varies with radia- tion polarization after Eq. (1), being in agreement with experimental observation shown in Figs. 2, 3 and 4(a). Moreover, as the ratchet photocurrents are proportional to the degree of the in-plane asymmetry, they reverse the sign upon inversion of static potential asymmetry. Exactly this behavior has been observed in experiment, see Fig. 2 (a) and (b). The proportionality to the de- gree of lateral asymmetry also explain the increase of the signal with raising voltage applied to one gate at con- stant voltage by the other. The interplay of the degree of lateral asymmetry and periodic modulation of THz electric field results in the complex gate-voltage depen- dence, in particular, for UG1 ≈ UG2. As the different individual contributions to the total current effect might imply different microscopic mechanisms of the photocur- rent formation, their behavior upon change of external parameters can distinct from each other. This would result in a sign-alternating gate-voltage behavior, in par- ticular for the range of comparable UG1 and UG2, like it is observed in experiment, see Fig. 2 (c). While all quali- tative features of the observed phenomena can be rather good described in terms of ratchet effects we would like to address another possible effect, which might trigger the helicity-driven photocurrent. It could be the differ- ential plasmonic drag effect in the two-dimensional struc- ture with an asymmetric double-grating gate considered in Refs. [22, 47]. As shown in Ref. [22] for a periodic AlAs/InGaAs/InAlAs/InP structure and linearly polar- ized THz radiation, photon drag effect can be comparable in strength with the plasmonic ratchet effect at THz fre- quencies. As the circular photon drag effect has been observed in different low dimensional materials [34, 48] we can expect that modification of the theory developed in [22] can also yield helicity driven plasmonic drag cur- rent compatible with the ratchet one. Finally, we note that the ratchet effects (either elec- tronic or plasmonic) can be greatly increased due to the resonant enhancement of the near-field in two- dimensional electron system at the plasmon resonance excitation as it was shown for the plasmonic ratchet in Refs. [23, 28]. The resonant plasmon condition ωτ > 1, see Ref. [12] can be well satisfied in our structure (ωτ = 4 at 2.54 THz). As shown in Ref. [23], the fundamental plasmon resonance is excited in a similar structure at fre- quency around 2 THz. Therefore, the plasmon resonance excitation can contribute to the observed ratchet effects independently of particular microscopic mechanisms of the ratchet photocurrent formation. The measurements in a broader THz frequency range could elucidate the role of the plasmonic resonance excitation in the ratchet photocurrent enhancement. V. SUMMARY To summarize, our measurements demonstrate that dual-grating-gate InAlAs/InGaAs/InAlAs/InP excited by terahertz radiation can yield a helicity sensitive pho- tocurrent response at THz frequencies. We show, that HEMTs with asymmetric lateral superlattice of gate fin- gers with unequal widths and spacing can be applied for generation of a photocurrent defined by linearly and circularly radiation polarization components. Moreover, one can obtain photoresponse being proportional to one of the Stokes parameters simply by variation of voltages applied to the individual gates. The photocurrent for- mations can be well described in terms of ratchet effects excited by terahertz radiation. By that the lateral grat- ing induces a periodical lateral potential acting on the 2D electron gas in QW. This grating also modulates the inci- dent radiation in the near field and hence in the plane of the 2DES, resulting in circular, linear and polarization- independent ratchet effects. While the responsivity of the polarization dependent response is lower than that reported for FET transistors it can be substantially im- proved by optimization of the structure design leading the resonant enhancement of the ratchet effects the plas- mon resonance excitation. We thank V. Popov for helpful discussions. The fi- nancial support from the DFG (SFB 689) is gratefully acknowledged. 6 edited by D. Saeedkia (Woodhead Publishing, Waterloo, Canada, 2013), pp. 121-155. [2] W. Knap, , S. Rumyantsev, M. S Pea Vitiello, D. Co- quillat, S. Blin, N. Dyakonova, M. Shur, F. Teppe, A. Tredicucci, T. Nagatsuma, Nanotechnology 24, 214002 (2013). [3] S. Boppel, A. Lisauskas, A. Max, V. Krozer, and H. G. Roskos, Opt. Lett. 37, 536 (2012). [4] V. M. Muravev and I. V. Kukushkin, Appl. Phys. Lett. 100, 082102 (2012). [5] M. Tonouchi, Nature Photon. 1, 97 (2007). [6] F. Schuster, D. Coquillat, H. Videlier, M. Sakowicz, F. Teppe, L. Dussopt, B. Giffard, T. Skotnicki, and W. Knap, Opt. Express 19, 7827 (2011). [7] G. C. Dyer, S. Preu, G. R. Aizin, J. Mikalopas, A. D. Grine, J. L. Reno, J. M. Hensley, N. Q. Vinh, A. C. Gossard, M. S. Sherwin, S. J. Allen, and E. A. Shaner, Appl. Phys. Lett. 100, 083506 (2012). [8] S. Preu, M. Mittendorff, S. Winnerl, H. Lu, A. C. Gos- sard, and H. B. Weber, Opt. Express 21, 17941 (2013). [9] T. Watanabe, S. A. Boubanga-Tombet, Y. Tanimoto, D. Fateev, V. Popov, D. Coquillat, W. Knap, Y. M. Meziani, Y. Wang, H. Minamide, H. Ito, and T. Otsuji, IEEE Sensors Journal 13, 89 (2013). [10] D. B. But, C. Drexler, M. V. Sakhno, N. Dyakonova , O. Drachenko, F. F. Sizov, A. Gutin, S. D. Ganichev, W. Knap, J. Appl. Phys. 115, 164514 (2014). [11] W. Knap, D. B. But, N. Dyakonova, D. Coquillat, A. Gutin, O. Klimenko, S. Blin, F. Teppe, M.S. Shur, T. Nagatsuma, S.D. Ganichev, and T. Otsuji, Recent Re- sults on Broadband Nanotransistor Based THz Detectors in NATO Science for Peace and Security Series B, Physics and Biophysics: THz and Security Applications, edited by C. Corsi, F. Sizov, (Springer, Dordrecht, Netherlands, 2014) pp.189 - 210. [12] M. Dyakonov and M. S. Shur, IEEE-Trans-ED 43(3), 380 (1996). [13] W. Knap, Y. Deng, S. Rumyantsev, J.-Q. Lu, M. S. Shur, C. A. Saylor and L. C. Brunel, Appl. Phys. Lett. 80, 3433 (2002). [14] W. Knap, V. Kachorovskii, Y. Deng, S. Rumyantsev, J.- Q. Lu, R. Gaska, M. S. Shur, G. Simin, X. Hu, M. Asif Khan, C. A. Saylor, and L. C. Brunel, J. Appl. Phys. 91, 9346 (2002). [15] T. Otsuji, M. Hanabe, T. Nishimura, and E. Sano, Opt. Exp. 14, 4815 (2006). [16] S. Sassine, Yu. Krupko, J.-C. Portal, Z. D. Kvon, R. Murali, K. P. Martin, G. Hill, and A. D. Wieck, Phys. Rev. B 78, 045431 (2008). [17] D. Coquillat , S. Nadar, F. Teppe, N. Dyakonova, S. Boubanga-Tombet, W. Knap, T. Nishimura, T. Otsuji, Y. M. Meziani, G. M. Tsymbalov, and V. V. Popov, Opt. Exp. 18, 6024 (2010). [18] V. V. Popov, J. Infr. Millim. THz Waves 32, 1178 (2011). [19] G. C. Dyer, G. R. Aizin, J. L. Reno, E. A. Shaner, and S. J. Allen, IEEE J. Sel. Topics Quantum Electron. 17, 85 (2011). [20] V. V. Popov, D. V. Fateev, T. Otsuji, Y. M. Meziani, D. Coquillat, and W. Knap, Appl. Phys. Lett. 99, 243504 (2011). [21] E. S. Kannan, I. Bisotto, J.-C. Portal, T. J. Beck, and L. Jalabert, Appl. Phys. Lett. 101, 143504 (2012). [1] W. Knap and M. Dyakonov, Plasma wave THz detec- tors and emitters in Handbook of Terahertz Technology [22] V. V. Popov, Appl. Phys. Lett. 102, 253504 (2013). [23] V.V. Popov, D.V. Fateev, T. Otsuji, Y.M. Meziani, D. Coquillat, and W. Knap, Appl. Phys. Lett. 99, 243504 (2011). [24] P. Olbrich, E.L. Ivchenko, T. Feil, R. Ravash, S.D. Danilov, J.Allerdings, D. Weiss, and S. D. Ganichev, Phys. Rev. Lett 103, 090603 (2009). [25] E. L. Ivchenko and S. D. Ganichev, JETP Lett. 93, 752 (2011). [26] P. Olbrich, J. Karch, E. L. Ivchenko, J. Kamann, B. Maerz, M. Fehrenbacher, D. Weiss, and S. D. Ganichev, Phys. Rev. B 83, 165320 (2011). [27] A. V. Nalitov, L. E. Golub, E. L. Ivchenko, Phys. Rev. B 86, 115301 (2012). [28] I. V. Rozhansky, V. Yu. Kachorovskii, and M. S. Shur, Phys. Rev. Lett. 114, 246601 (2015). [29] B. E. A. Saleh, M. C. Teich, Fundamentals of Photonics (John Wiley & Sons, New York, 2003). [30] S. N. Danilov, B. Wittmann, P. Olbrich, W. Eder, W. Prettl, L. E. Golub, E. V. Beregulin, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoretsky, V. A. Shalygin, N. Q. Vinh, A. F. G. van der Meer, B. Murdin, and S. D. Ganichev, J. Appl. Physics 105, 013106 (2009). [31] S. D. Ganichev, J. Kiermaier, W. Weber, S. N. Danilov, D. Schuh, Ch. Gerl, W. Wegscheider, D. Bougeard, G. Abstreiter, and W. Prettl, Appl. Phys. Lett. 91, 091101 (2007). [32] S. D. Ganichev, W. Weber, J. Kiermaier, S. N. Danilov, D. Schuh, W. Wegscheider, Ch. Gerl, D. Bougeard, G. Abstreiter and W. Prettl, J. Appl. Physics 103, 114504 (2008). [33] S.D. Ganichev, S.A. Tarasenko, V.V. Bel'kov, P. Olbrich, W. Eder,D.R. Yakovlev, V. Kolkovsky, W. Zaleszczyk, G. Karczewski, T. Wojtowicz, and D. Weiss, Phys. Rev. Lett. 102, 156602 (2009). [34] J. Karch, P. Olbrich, M. Schmalzbauer, C. Zoth, C. Brinsteiner, M. Fehrenbacher, U. Wurstbauer, M. M. Glazov, S. A. Tarasenko, E. L. Ivchenko, D. Weiss, J. Eroms, R. Yakimova, S. Lara-Avila, S. Kubatkin, S. D. Ganichev, Phys. Rev. Lett. 105, 227402 (2010). [35] J. Karch, C. Drexler, P. Olbrich, M. Fehrenbacher, M. Hirmer, M. M. Glazov, S. A. Tarasenko, E. L. Ivchenko, B. Birkner, J. Eroms, D. Weiss, R. Yakimova, S. Lara-Avila, S. Kubatkin, M. Ostler, T. Seyller, S. D. Ganichev, Phys. Rev. Lett. 107, 276601 (2011). [36] M.M. Glazov and S.D. Ganichev, Physics Reports 535, 101 (2014). [37] S. D. Ganichev and W. Prettl, Intense Terahertz Exci- tation of Semiconductors (Oxford University Press, Ox- ford, 2006). [38] While being detected in all reported measurements a po- larization independent offset given by the coefficient j0 will not be discussed in details. Instead, hereafter we fo- cus on helicity sensitive photocurrent, jC, and currents 7 driven by linearly polarized light, j1 and j2. [39] M. Sakowicz, M. B. Lifshits, O. A. Klimenko, F. Schuster, D. Coquillat, F. Teppe, and W. Knap, J. Appl. Phys. 110, 054512 (2011). [40] D. Coquillat, V. Nodjiadjim, A. Konczykowska, M. Riet, N. Dyakonova, C. Consejo, F. Teppe, J. Godin, W. Knap, Didgest of Int. Conf. on Infrared, Millimeter, and Tera- hertz Waves, Tucson, USA, (2014). [41] Note that signal variation with polarization is, apart the offset, identical with that of s1, therefore this polarization dependence can also be used to describe the j1-related photocurrent behavior. [42] C. Drexler, N. Dyakonova, P. Olbrich, J. Karch, M. Schafberger, K. Karpierz, Yu. Mityagin, M. B. Lifshits, F. Teppe, O. Klimenko, Y. M. Meziani, W. Knap, and S. D. Ganichev, J. Appl. Physics 111, 124504 (2012). [43] K. S. Romanov and M. I. Dyakonov, Appl. Phys. Lett. 102, 153502 (2013). [44] All previous works aimed to the radiation induced ratchet effects discuss the case of unconnected parallel metal stripes: a system belonging to Cs point group symme- try consisting of the identity element and the reflection in the plane perpendicular to the stripes [22, 24 -- 28]. For this symmetry circular photocurrent jC and the pho- tocurrent j2 can be generated along stripes only whereas polarization independent offset and j0 and photocurrent j1 are allowed in the perpendicular to that direction (source-drain). Design of our DDG structures with in- terconnected metal stripes in each of gates excludes re- flection plane reducing the point group symmetry to C1. As a result the symmetry does not imply any restrictions and the photocurrent includes all four individual con- tributions (j0, j1, j2 and jC ) which are allowed in any in-plain direction. More details on the symmetry analy- sis of photocurrents in quantum wells of C1 symmetry can be found in [45, 46]. [45] B. Wittmann, S.N. Danilov, E.G. Novik, V.V. Bel'kov, H. Buhmann, S.A. Tarasenko, C. Brune, L.W. Molenkamp, E.L. Ivchenko, Z.D. Kvon, N.N. Mikhailov, S.A. Dvoretsky, N. Q. Vinh, A. F. G. van der Meer, B. Murdin, and S.D. Ganichev Semicond. Sci. and Technology 25, 095005 (2010). [46] V.V. Bel'kov, and S.D. Ganichev, Semicond. Sci. Tech- nol. 23, 114003 (2008). [47] V.V. Popov, D.V. Fateev, E.L. Ivchenko, and S.D. Ganichev, Phys. Rev. B 91, 235436 (2015). [48] V.A. Shalygin, H. Diehl, Ch. Hoffmann, S.N. Danilov, T. Herrle, S.A. Tarasenko, D. Schuh, Ch. Gerl, W. Wegscheider, W. Prettl and S.D. Ganichev, JETP Lett. 84, 570 (2006). 

1908.07252

1

1908

Thermal generation of shift electric current

It is shown that the dissipation of energy in an electron gas confined in a quantum well made of non-centrosymmetric crystal leads to a direct electric current. The current originates from the real-space shift of the wavepackets of Bloch electrons at the electron scattering by phonons, which tends to restore thermal equilibrium between the electron and phonon subsystems. We develop a microscopic theory of such a phonogalvanic effect for narrow band gap zinc-blende quantum wells.

cond-mat.mes-hall

cond-mat

a Thermal generation of shift electric current G. V. Budkin and S. A. Tarasenko Ioffe Institute, 194021 St. Petersburg, Russia It is shown that the dissipation of energy in an electron gas confined in a quantum well made of non-centrosymmetric crystal leads to a direct electric current. The current originates from the real-space shift of the wavepackets of Bloch electrons at the electron scattering by phonons, which tends to restore thermal equilibrium between the electron and phonon subsystems. We develop a microscopic theory of such a phonogalvanic effect for narrow band gap zinc-blende quantum wells. I. INTRODUCTION The currents of charge carriers, electrons or holes, in solid-state systems are commonly generated by gradients of electric or chemical potentials, temperature, etc. In structures of low spatial symmetry, direct currents can also emerge when the system is driven out of thermal equilibrium by an undirected force with zero average driving [1, 2]. Well known examples are ratchet and photogalvanic effects [3 -- 11] and the spin-galvanic effect [12 -- 14], when non-equilibrium spin polarization drives an electric current in gyrotropic structures. Here, we study the effect of generating direct electric current from the kinetic energy of hot electrons. We show that the breaking of thermal equilibrium between the electron and phonon subsystems in semiconductor structure of sufficiently low symmetry is enough to drive an electric current. Microscopically, the current originates from the shift of Bloch elec- trons in real space at the emission or absorption of phonons, which tend to restore thermal equilibrium in the system. While the spin-dependent shift of electrons at momentum scatter- ing (side jump contribution) has been intensively studied, particularly, in the physics of the anomalous and spin Hall effects [15 -- 20], less is known about the charge shift which occurs at inelastic scattering [21, 22]. Here, we study this effect for narrow gap two-dimensional (2D) systems made of zinc-blend-type semiconductors, such as HgTe/CdHgTe quantum wells (QWs), which naturally lack the center of space inversion. These 2D structures, except for QWs grown along the high-symmetry axes [001] and [111], are polar in the plane and sup- port the generation of electric current in the presence of energy transfer between the electron subsystem and the crystal lattice. In analogy to the (linear) photogalvanic effect [2, 23], where the absorption of photons gives rise to a shift electric current, the effect we study can be named phonogalvanic. 2 II. MICROSCOPIC PICTURE We consider a two-dimensional electron gas in a QW structure, see Fig. 1. The electron gas is initially heated. Frequent electron-electron collisions thermalize the electrons and establish the Fermi-Dirac distribution with the effective electron temperature Te that is above the lattice temperature T . The electron gas is being cooled down by emitting phonons which transfer the energy from the QW region. As will be calculated below, the processes of phonon emission/absorption in zinc-blend QWs are accompanied by the in-plane shift of the Bloch electrons involved in the quantum transitions. The shifts have a preferable direction, except for the QWs grown along the high-symmetry axes [001] and [111], although the phonons can be emitted in all directions. Therefore, the energy relaxation of electrons gives rise to a direct electric current j. The current is generated until the electron temperate reaches the lattice temperature. FIG. 1. Shift mechanism of phonogalvanic effect in quantum wells. Emission of phonons by hot electrons is accompanied by the electron shift in the QW plane, which leads to a net electric current. At thermodynamic equilibrium, the net current vanishes because the electron shifts at the phonon emission and absorption are opposite and compensate each other. If, however, the electron temperature is higher (or lower) than the lattice temperature, the processes of phonon emission (absorption) dominate, which leads to an electric current. 3 III. THEORY Now we develop a microscopic theory of the phonogalvanic effect. We present a brief derivation of the electron shift in 2D systems, describe the electron states in narrow gap QWs and interaction with acoustical phonons in zinc-blende structures, calculate the electron shift and the density of electric current, and discuss the results. A. Electron shift in 2D structures The Bloch states of electrons with the in-plane wave vectors k and the spin index s in a two-dimensional crystalline structure are described by the functions ψks = Uks exp(ik · ρ) , (1) where Uks contains the Bloch amplitude and the envelope function along the quantization axis z and ρ = (x, y) is the in-plane coordinate. To calculate the electron displacement at the quantum transition between the initial state ν = (k, s) and the final state ν(cid:48) = (k(cid:48), s(cid:48)) we consider that these states are described by narrow wave packets of the form Ψν = c(ν) ks ψks (2) (cid:88) ks centered at the wave vectors k and k(cid:48), respectively. The displacement in the real space is determined by the difference of the position mean values in the initial and final states. For the wave packets (2) constructed from the Bloch functions (1), the displacement is given by (cid:20) (cid:20) (cid:88) −(cid:88) ks Rν(cid:48)ν = (cid:21) ic(ν(cid:48))∗ ks d dk ks + c(ν(cid:48)) c(ν(cid:48)) ks 2Ωks (cid:21) ic(ν)∗ ks d dk ks + c(ν) c(ν) ks 2Ωks where (cid:90) ks Ωks = i U∗ ks d dk Uksdr , , (3) (4) also called the Berry connection in the reciprocal space. Taking into account that the wave packets are narrow, the displacement can be expressed in terms of the gradient of the phase Φk(cid:48)s,ks which the wave function acquires at the transition, Rν(cid:48)ν = − (∇k(cid:48) + ∇k) Φk(cid:48)s(cid:48),ks + Ωk(cid:48)s(cid:48) − Ωks , (5) 4 or, equivalently, in terms of the transition matrix elements Vk(cid:48)s(cid:48),ks, (∇k(cid:48) + ∇k) Φk(cid:48)s(cid:48),ks = k(cid:48)s(cid:48),ks (∇k(cid:48) + ∇k) Vk(cid:48)s(cid:48),ks Vk(cid:48)s(cid:48),ks2 Im(cid:2)V ∗ (cid:3) . Equation (5) for the displacement of Bloch electrons at quantum transitions in crystals was obtained by Luttinger [15] and Belinicher, Ivchenko, and Sturman [21]. Note that the displacement (5) does not dependent on the choice of the phases of the Uks and Uk(cid:48)s(cid:48) functions while the individual contributions are non-invariant. B. Electron states in narrow gap QWs To be specific we consider the class of (0lh)-oriented QWs, where l and h are integer numbers, which includes (001)-, (011)-, and (013)-grown structures. The ground electron e1, s(cid:105) and heavy-hole h1, s(cid:105) subbands at k = 0 are described in the 6-band k·p model by the wave functions e1, +(cid:105) = f1(z)Γ6, +1/2(cid:105) + f4(z)Γ8, +1/2(cid:105) , h1, +(cid:105) = f3(z)Γ8, +3/2(cid:105) , e1,−(cid:105) = f1(z)Γ6,−1/2(cid:105) + f4(z)Γ8,−1/2(cid:105) , h1,−(cid:105) = f3(z)Γ8,−3/2(cid:105) , (6) where fj(z) are the envelope functions, Γ6, m(cid:105) (m = ±1/2) and Γ8, m(cid:105) (m = ±1/2,±3/2) are the Bloch amplitude of the Γ6 and Γ8 states in the center of the Brillouin zone, re- spectively. At k (cid:54)= 0, the states are mixed. The mixing in narrow gap QWs, such as HgTe/CdHgTe, is described by the k-linear Bernevig-Hughes-Zhang Hamiltonian [24]  H =  , δ iAk+ −iAk− −δ 0 0 0 0 0 0 0 0 δ −iAk− iAk+ −δ (7) 5 where 2δ is the energy gap between the e1 and hh1 subbands and A is a real parameter determining the dispersion. Admixture of excited electron and hole subbands and zero-field spin splitting are neglected in this model Hamiltonian [25, 26]. In QWs with symmetric confinement potential, the functions f1(z) and f3(z) are even while f4(z) is odd with respect to the QW center. Solution of the Schrodinger equation with the Hamiltonian (7) yields the dispersion of the conduction subband and the wave functions Uks = √ δ2 + A2k2 εk = where s = ±1, ϕ = arctan(ky/kx) is the polar angle of the wave vector k, and 1√ 2 (cid:0)ake1, s(cid:105) − ise−isϕbkh1, s(cid:105)(cid:1) , (cid:114)εk + δ (cid:114) εk − δ . εk ak = , bk = εk For the states in the valence subband in the electron representation one obtains and k = − ε(v) √ δ2 + A2k2 (cid:0)bke1, s(cid:105) + ise−isϕakh1, s(cid:105)(cid:1) U (v) ks = 1√ 2 with the coefficients ak and bk given by Eqs. (10). (8) (9) (10) (11) (12) C. Electron-phonon interaction Consider now the deformation interaction of electrons with acoustic phonons (DA mech- anism) [27]. The matrix elements of the deformation interaction within the Γ6 and Γ8 bands in zinc-blend-type crystals can be presented in the form (cid:104)Γ6, mV Γ6, m(cid:48)(cid:105) = Ξc Tr uαβ δmm(cid:48) , (cid:104)Γ8, mV Γ8, m(cid:48)(cid:105) = Ξv Tr uαβ δmm(cid:48) , (13) where Ξc and Ξv are the deformation-potential constants and uαβ are the strain tensor components. Here, we use the simplified Hamiltonian for the Γ8 band with the single deformation-potential constant Ξv. Ξv = a + (5/4)b in the Bir-Pikus notation [28], the other terms are neglected. 6 Importantly, in zinc-blend crystals the strain also mixes the states of the Γ6 and Γ8 bands [29]. Such an interband mixing in the coordinated frame x (cid:107) [100], y (cid:107) [0h¯l], z (cid:107) [0lh], relevant for (0lh)-oriented QWs, is described by the matrix elements [30] (cid:20) (cid:104)Γ8,±3/2V Γ6,±1/2(cid:105) = = ∓Ξcv√ 2 (uyz ∓ iuxz) cos 2θ + 3(cid:104)Γ8,±1/2V Γ6,∓1/2(cid:105) (cid:18) uzz − uyy (cid:19) ± iuxy sin 2θ , 2 (cid:21) (cid:104)Γ8,±1/2V Γ6,±1/2(cid:105) = [uxy cos 2θ + uxz sin 2θ] , (14) √ (cid:114)2 3 where θ is the angle between the [001] and [0lh] axes and Ξcv is the interband deformation- potential constant [29]. Note that Ξcv vanishes in centrosymmetric crystals. The matrix elements of the scattering of electrons confined in the QW by bulk acoustic phonons can be readily calculated using the deformation Hamiltonians (13,14) and the wave functions (9). Since Ξc,Ξv > Ξcv, the scattering occurs mainly with the conservation of the spin index s and by longitudinal (LA) phonons. Further, the out-of-plane component qz of the wave vector q of the phonons involved is typically much larger than the in-plane component q(cid:107) = k − k(cid:48). In these approximations, the matrix elements of the scattering assisted by emission or absorption of LA phonons assume the form k(cid:48)s,ks = ∓i V (±) q2 z 2q (ΞcZ11 + ΞvZ44)ak(cid:48)ak + ΞvZ33 (cid:35) × bk(cid:48)bkeis(ϕ(cid:48)−ϕ) − i ΞcvZ13(bk(cid:48)akeisϕ(cid:48) − ak(cid:48)bke−isϕ) δk(cid:48),k∓q(cid:107), (15) q = Nq + (1 ± 1)/2, Nq is the phonon occupation number, ρ is where q(cid:107) = ±(k − k(cid:48)), N (±) the crystal density, ωq = csq is the phonon frequency, cs is the speed of longitudinal sound, and Zij(qz) =(cid:82) fi(z)fj(z)eiqzzdz. D. Electric current (cid:115)N (±) q (cid:34) 2ρωq √ sin 2θ 2 2 The electric current caused by the shift of electrons at quantum transitions is calculated after the equation j = e (cid:88) ks,k(cid:48)s(cid:48) Wk(cid:48)s(cid:48),ksRk(cid:48)s(cid:48),ks , (16) where (cid:88) q,± Wν(cid:48),ν = 2π  V (±) ν(cid:48),ν 2fν(1 − fν(cid:48))δ(εν(cid:48) − εν ± ωq) 7 (17) is scattering probability, Rν(cid:48)ν is the shift given by (5), and fν is the electron distribution function. We assume that electron thermalization governed by electron-electron collisions is estab- lished at a time scale which is much shorter than the energy relaxation time determined by in- elastic electron-phonon scattering. As a result, the electron distribution is quasi-equilibrium and can be described by the Fermi-Dirac function fν = {exp[(εν − µ)/kBTe] + 1}−1, where Te is the effective electron temperature and µ is the chemical potential. The phonon dis- tribution is described by the Bose-Einstein function Nq = [exp(ωq/kBT ) − 1]−1 with the lattice temperature T . At ∆T = Te − T (cid:28) T , the difference between the rates of photon emission and absorption is given by (cid:2)fν(cid:48)(1 − fν)N + (cid:3) q − fν(1 − fν(cid:48))N− q εν(cid:48) =εν +ωq ≈ fν(cid:48)(1 − fν) ∆T T . (18) To proceed further, we calculate the current contributions j(1) and j(2) related to the phase gradients and the Berry connections in Eq. (5), respectively, separately. However, we stress that only the sum of them, j = j(1) + j(2), is invariant with respect to the wave function choice and is a physical observable. The contribution to the current stemming from the phase gradients is determined by the vectors Λ(±) = k(cid:48)s(cid:48),ks(∇k(cid:48) + ∇k)V (±) V (±)∗ k(cid:48)s(cid:48),ks , (19) where the angle brackets denote averaging over the directions of the wave vectors k and k(cid:48). For the matrix elements (15), they have the form [ΞcZ11(qz) + ΞvZ44(qz)]× x sin 2θ 2 √ 32 Λ(±) = (cid:20) ak(cid:48)akak(cid:48) (cid:18) d dk (cid:20) − ΞvZ33(qz) bk(cid:48)bkbk(cid:48) q2 z Z13(qz)ΞcvN (±) q (cid:19) ρωq bk − ak(cid:48)akak (cid:18) d dk(cid:48) + dk(cid:48) − bk(cid:48)bk(cid:48)ak dak(cid:48) (cid:19) (cid:18) d 1 k(cid:48) dk 1 k − bk(cid:48)bkbk + dak dk bk(cid:48) − ak(cid:48)bkak(cid:48) dak dk (cid:19) + akbk(cid:48)ak (cid:18) d + 1 k bk + bkbkak(cid:48) dk(cid:48) + (cid:21) (cid:21)(cid:41) bk(cid:48) , (20) (cid:19) dak(cid:48) dk(cid:48) 1 k(cid:48) (cid:69) (cid:68) (cid:40) where x is the unit vector along the x axis. This results in the following contribution to the 8 electric current j(1) x = e sin 2θ √ 2π2 64 Ξcv∆T ρA3d3 (cid:20) +∞(cid:90) δ (Ξcζ1 + Ξvζ4) a4 k + Ξvζ3 (cid:18) a4 k − b4 kb2 k + 2a2 2 k (cid:19)(cid:21) fk(1 − fk) T dεk , (21) where d is the QW width, ζj are dimensionless parameters determined by the functions of size quantization, ζj = d3 +∞(cid:90) dqzq2 z Z13(qz)Zjj(qz) = 2πd3 +∞(cid:90) −∞ −∞ d f1(z)f3(z) dz d f 2 j (z) dz dz , (22) and it is assumed that the energy of phonons involved is much smaller than the mean electron energy. FIG. 2. Distribution of the Berry connection Ωks in the k space in the spin subbands s = " ± " plotted after Eq. (23). Red and blue dotted circles show isoenergetic curves. Black arrows sketch the electron transitions with the emission of phonons. Asymmetry of the electron transitions in the spin and momentum spaces (shown by black arrows of different thickness) at energy relaxation leads to a variation of the mean value of Ωks, i.e., to the contribution j(2) to the shift current. The second contribution to the current is determined by the Berry connection which, for the wave functions (9), has the form Ωks = z × k , sb2 k 2k (23) Ωk'--Ωk-Ωk'+-Ωk+s=+s=-kxkyΩk+Ωk'+Ωk-Ωk'-εkεk'εkεk'9 where z and k are the unit vectors along z and k, respectively. The Berry connection lies in the QW plane and points in the direction perpendicular to k. Its direction depends on the spin index s and its amplitude is proportional to b2 k/k. The distribution of the Berry connection in the k space for the both spin subbands is shown in Fig. 2. Black arrows in the same figure sketch the electron transitions with the initial k and final k(cid:48) wave vectors with the emission of phonons. Owing to the term ∝ Ξcv, the rate of such transitions is asymmetric in the k space, which is illustrated by the arrows of different thickness. Importantly, the sign of the asymmetry also depends on the spin index s. Therefore, the value of Ωk(cid:48)s − Ωks summed up over the initial and final wave vectors is the same for the spin subbands and contributes to the shift current. Straightforward calculations yields the Berry connection contribution to the current j(2) x = e sin 2θ √ 2π2 64 Ξcv∆T ρA3d3 (cid:20) +∞(cid:90) δ (Ξcζ1 + Ξvζ4) (cid:18)a4 k − 2a2 k − b4 kb2 2 k (cid:19) − Ξvζ3b4 k (cid:21) fk(1 − fk) T dεk . (24) Finally, summing up the partial contributions (21) and (24) we obtain the shift electric current caused by the energy relaxation of electrons in the conduction subband jx = e sin 2θ √ 2π2 32 Ξcv∆T ρA3d3 (cid:20) +∞(cid:90) δ (cid:18) (cid:19) δ εk (cid:19)(cid:21) δ (cid:18) 2 − δ εk (Ξcζ1 + Ξvζ4) 2 + + Ξvζ3 fk(1 − fk) εk T dεk . (25) The shift current emerges also in p-type QWs at energy relaxation of holes. Similar calculations carried out for the case when the chemical potential lies in the valence subband give also Eq. (25) where the integral is taken from −∞ to −δ. Equation (25) is the main result of this paper. Below, we discuss it and analyze for some particular cases. IV. DISCUSSION The shift current depends on the band structure parameters, the QW crystallographic orientation, and the carrier distribution function. Its dependence on the QW orientation is given by sin 2θ. The case of θ = πn/2 with integer n, when the current vanishes, corresponds to {001}-grown QWs. Such structures have high symmetry described by the D2d or C2v point groups depending on whether the confinement potential is symmetric or asymmetric. The structures contain two mirror planes orthogonal to each other and to the QW plane, which 10 forbids the generation of an in-plane electric current by a scalar perturbation of the system, such as breaking the thermal equilibrium between electrons and phonons. At θ (cid:54)= πn/2, the QW structures have lower symmetry (our model Hamiltonian corresponds to the Cs point group, the actual symmetry can be even lower). They belong to the class of pyroelectric structures with the in-plane component of the polar vector, and the symmetry allows the current generation. The integral in Eqs. (25) can be calculated analytically in the case of low temperature and degenerated electron gas when fk(1 − fk)/T ≈ δ(εk − εF ), where εF is the Fermi level measured from the center of the band gap. Then, the expression for the current assumes the form jx = e sin 2θ √ 2π2 32 Ξcv∆T ρA3d3 δ ε2 F [(Ξcζ1 + Ξvζ4)(2εF + δ) + Ξvζ3(2εF − δ)] Θ(εF − δ) , (26) where Θ is the Heaviside step function. Equation (26) is valid for both n-type and p-type structures. A prominent example of narrow gap QWs with the zinc-blend lattice is HgTe/CdHgTe structures. Therefore, we make quantitative calculations of the shift current for the pa- rameters relevant to HgTe [31]: the crystal density ρ = 8 g/cm3, the deformation-potential constants Ξc = −3.8 eV and Ξv = −2 eV, and the ratio Ξcv/Ξc = 1/3 (this ratio is not known for HgTe, we use the ratio for GaAs [29]). The band gap δ and overlap integrals ζj are calculated numerically in the k · p-model [32]. Figure 3 shows the dependence of the shift current on the Fermi level position calculated after Eq. (26). Curves of different colors correspond to the QWs with different band gap which can be controlled in HgTe/CdHgTe structures by the QW thickness. Moreover, in HgTe/CdHgTe structures the Fermi level position can be tuned from the conduction to the valence subband by applying the gate voltage. Figure 3 shows that the current flows in the opposite directions in n-type and p-type structures. At small densities of electrons or holes, when the Fermi level εF is close to ±δ, respectively, the electric currents reach the values jc,v = ±e sin 2θ √ 32 2π2 Ξcv∆T ρA3d3 [(2 ± 1)(Ξcζ1 + Ξvζ4) + (2 ∓ 1)Ξvζ3] . (27) 11 FIG. 3. Dependence of the shift current on the Fermi level position calculated after Eq. (26) for different band gaps δ, the parameters relevant to the HgTe/CdHgTe QWs, ∆T = 10 K, and θ = π/4. Black line correspond to δ = 5 meV (QW width 5 nm), blue line to δ = 10 meV (width 5.6 nm) and red line to δ = 20 meV (width 6.0 nm). The theory of the shift current is developed here for a small deviation between the effective It gives j ∼ nA/cm for ∆T = 10 K and the energy relaxation by acoustic phonons. If fact, the current magnitude is proportional to the rate electron and lattice temperatures. of energy transfer between the electron and lattice subsystems. The estimations above correspond to the energy transfer rate E ∼ 0.3 mW/cm2. In real experiments on electron E can be orders of magnitude larger giving rise to gas heating by THz or optical pulses, much stronger electric current. EFEFδ-δ2δ2δ-40-2002040-1.6-1.2-0.8-0.400.40.81.2Fermilevel,εF(meV)Shiftcurrent,j(nA/cm)V. SUMMARY 12 We have shown that the energy relaxation of a heated (or cooled) electron gas in low- symmetry quantum wells by phonons drives a direct electric current and developed a mi- croscopic theory of such a phonogalvanic effect. The current originates from the shifts in the real space that the Bloch electrons experience in the course of quantum transitions at emission or absorption of phonons. When the thermal equilibrium between the electron and phonon subsystems is broken and the subsystems are characterized by different effective temperatures, the shifts get a preferable direction, which results in a net electric current. We have calculated the electric current for the deformation mechanisms of electron-phonon interaction in narrow band gap zinc-blende quantum wells. This work was supported by the Government of the Russian Federation (contract # 14.W03.31.0011 at Ioffe Institute). G.V.B also acknowledges the support from the BASIS foundation. [1] R. P. Feynman, R. B. Leighton, and M. Sands, "The feynman lectures on physics vol 1," (San Francisco, CA: Pearson, 2006) Chap. 46. [2] P. Sturman and V. Firdkin, Photovoltaic and Photo-refractive Effects in Noncentrosymmetric Materials, Ferroelectricity and related phenomena (Taylor & Francis, 1992). [3] P. Reimann, Physics Reports 361, 57 (2002). [4] P. Hanggi and F. Marchesoni, Reviews of Modern Physics 81, 387 (2009). [5] J. Bang, R. Pan, T. M. Hoang, J. Ahn, C. Jarzynski, H. T. Quan, and T. Li, New Journal of Physics 20, 103032 (2018). [6] E. M. Hohberger, A. Lorke, W. Wegscheider, and M. Bichler, Applied Physics Letters 78, 2905 (2001). [7] E. L. Ivchenko and S. D. Ganichev, JETP Letters 93, 673 (2011). [8] A. D. Chepelianskii, M. V. Entin, L. I. Magarill, and D. L. Shepelyansky, Physical Review E 78 (2008), 10.1103/physreve.78.041127. [9] S. A. Tarasenko, Physical Review B 83 (2011), 10.1103/physrevb.83.035313. [10] C. Drexler, S. A. Tarasenko, P. Olbrich, J. Karch, M. Hirmer, F. Muller, M. Gmitra, J. Fabian, 13 R. Yakimova, S. Lara-Avila, S. Kubatkin, M. Wang, R. Vajtai, P. M. Ajayan, J. Kono, and S. D. Ganichev, Nature Nanotechnology 8, 104 (2013). [11] G. V. Budkin and S. A. Tarasenko, Phys. Rev. B 93, 075306 (2016). [12] S. D. Ganichev, E. L. Ivchenko, V. V. Bel'kov, S. A. Tarasenko, M. Sollinger, D. Weiss, W. Wegscheider, and W. Prettl, Nature 417, 153 (2002). [13] G. Seibold, S. Caprara, M. Grilli, and R. Raimondi, Physical Review Letters 119 (2017), 10.1103/physrevlett.119.256801. [14] D. Smirnov and L. Golub, Physical Review Letters 118 (2017), 10.1103/phys- revlett.118.116801. [15] J. M. Luttinger, Physical Review 112, 739 (1958). [16] N. A. Sinitsyn, Journal of Physics: Condensed Matter 20, 023201 (2007). [17] N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. 82, 1539 (2010). [18] L. N. Oveshnikov, V. A. Kulbachinskii, A. B. Davydov, B. A. Aronzon, I. V. Rozhansky, N. S. Averkiev, K. I. Kugel, and V. Tripathi, Scientific Reports 5 (2015), 10.1038/srep17158. [19] I. A. Ado, I. A. Dmitriev, P. M. Ostrovsky, and M. Titov, Phys. Rev. B 96, 235148 (2017). [20] C. Xiao, Y. Liu, M. Xie, S. A. Yang, and Q. Niu, Phys. Rev. B 99, 245418 (2019). [21] V. I. Belinicher, E. L. Ivchenko, and B. I. Sturman, Sov. Phys. JETP 56, 359 (1982). [22] L. E. Golub and E. L. Ivchenko, JETP 112, 152 (2011). [23] E. M. Baskin, L. I. Magarill, and M. V. Entin, Fizika Tverdogo Tela 20, 2432 (1978). [24] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). [25] R. Winkler, L. Wang, Y. Lin, and C. Chu, Solid State Communications 152, 2096 (2012). [26] S. A. Tarasenko, M. V. Durnev, M. O. Nestoklon, E. L. Ivchenko, J.-W. Luo, and A. Zunger, Phys. Rev. B 91, 081302 (2015). [27] V. Gantmakher and I. Levinson, Carrier scattering in metals and semiconductors, Modern problems in condensed matter sciences (North-Holland, 1987). [28] G. Bir and G. Pikus, Symmetry and Strain-induced Effects in Semiconductors, A Halsted Press book (Wiley, 1974). [29] G. E. Pikus, V. A. Marushchak, and A. N. Titkov, Sov. Phys. Semicond. 22, 115 (1988). [30] P. Olbrich, C. Zoth, P. Vierling, K.-M. Dantscher, G. V. Budkin, S. A. Tarasenko, V. V. Bel'kov, D. A. Kozlov, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoretsky, and S. D. Ganichev, 14 Phys. Rev. B 87, 235439 (2013). [31] S. Adachi, Handbook on Physical Properties of Semiconductors. Volume 3: II -- VI Compound Semiconductorst (Kluwer Academic Publishers, 2004). [32] K.-M. Dantscher, D. A. Kozlov, P. Olbrich, C. Zoth, P. Faltermeier, M. Lindner, G. V. Budkin, S. A. Tarasenko, V. V. Bel'kov, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoretsky, D. Weiss, B. Jenichen, and S. D. Ganichev, Phys. Rev. B 92, 165314 (2015). 

1501.02644

1

1501

Discrete relaxation of exciton-polaritons in an inhomogeneous potential

We present indications, that the wave function-stiffness condition during energy-relaxation as observed in single-phase state quantum systems manifests also in a single particle ensemble. This is demonstrated for exciton-polaritons in the strong coupling regime in a ZnO-based microcavity at T = 10 K for non-resonant excitation. It is well known that the pump-induced spatially inhomogeneous background potential leads to nearly equally spaced energy levels in the k-space distribution for propagating polariton Bose-Einstein condensates. Surprisingly this particular pattern is also observable for uncondensed exciton-polaritons.

cond-mat.mes-hall

cond-mat

Discrete relaxation of exciton-polaritons in an inhomogeneous potential T. Michalsky, H. Franke, C. Sturm, M. Grundmann, and R. Schmidt-Grund Universitat Leipzig, Linn´estrasse 5, D-04103 Leipzig, Germany (Dated: May 8, 2019) Abstract We present indications, that the wave function-stiffness condition during energy-relaxation as ob- served in single-phase state quantum systems manifests also in a single particle ensemble. This is demonstrated for exciton-polaritons in the strong coupling regime in a ZnO-based microcavity at T = 10 K for non-resonant excitation. It is well known that the pump-induced spatially inhomoge- neous background potential leads to nearly equally spaced energy levels in the k -space distribution for propagating polariton Bose-Einstein condensates. Surprisingly this particular pattern is also observable for uncondensed exciton-polaritons. 5 1 0 2 n a J 2 1 ] l l a h - s e m . t a m - d n o c [ 1 v 4 4 6 2 0 . 1 0 5 1 : v i X r a 1 I. INTRODUCTION In microcavities (MCs) exciton-polaritons (polaritons) are composite particles with partly photonic and excitonic properties. Their bosonic character allows them to undergo the phase transition to a macroscopically coherent state known as a dynamic Bose-Einstein condensate (BEC) which was first experimentally shown in 2006 in CdTe-based MCs [1] at liquid helium temperatures. Since then BECs in MCs have evolved to be a model system for research in many topics of modern physics, like superfluidity [2], vortex-generation [3] and ballistic transport phenomena [4] leading to novel devices like all-optical polariton switches [5] and transistors [6]. In this context often a varying background potential is used to accelerate the condensed particles. Relaxation processes connected to this background potential have been observed by several groups [4, 7 -- 10]. Wouters et al. [11] developed a description of the occuring phenomena in terms of energy conservation and scattering properties. In this work, we present experimental results suggesting that this model can also be applied to a certain extent to an ensemble of uncondensed polaritons. II. EXPERIMENTAL METHODS The investigated ZnO-based MC is the same as has been described in Ref. [10]. The MC contains a wedge-shaped cavity layer giving access to a wide range of detuning values ∆ between the uncoupled cavity-photon mode and the excitonic transition energy. We used a frequency tripled and modelocked Ti:Sa laser at a wavelength of 266 nm for non resonant excitation. The pulse width was 2 ps at a repetition rate of 76 MHz. The laser beam was focused on an area of about 5 µm2 through an UV objective with a magnification of 50 and a numerical aperture (NA) of 0.4 corresponding to a detectable angular range of ±23◦. The detection of the light emitted from the sample was realized in a confocal configuration. The lens setup allows for the imaging of the momentum (k -) as well as of the real space. In order to investigate relaxation processes which take place in the region where the pump-induced background potential is strongly inhomogeneous, a pinhole (PH) was put in an intermediate image plane so that only the emission from the central excitation area of about 3 µm2 was detected. If the PH is used the lower polariton branch (LPB) emission is spectrally 2 broadened and smeared out. This is caused by diffraction at the PH as follows from the Heisenberg principle: ∆x∆k ≥ 0.5. Furthermore the detectability of high-k emission is suppressed as the cryostat window introduces spherical aberrations which strongly increase with the emission angle causing these rays to be blocked at the PH. III. EXPERIMENTAL RESULTS AND DISCUSSION The following results are taken from a sample position corresponding to ∆ = −30 meV (∼ 0.6× the coupling constant of V = 50 meV) at T = 10 K. Figure 1 shows the polariton emission below (a) and above (b) the condensation threshold without PH and includes the calculated dispersion relations of the uncoupled exciton (X) and cavity-photon mode (C) [12]. In the uncondensed case the lower polariton branch (LPB) with a homogeneous broadening of γuncond = (2.8 ± 0.2) meV is clearly visible without any recognizable further features. Above the condensation threshold (Fig. 1 (b) and (c)) several dispersionless states appear bounded by the dispersion relation of the uncondensed polaritons. These states are roughly equally spaced in energy (see Fig. 1 (c)) with a distance of ∆Er,cond = (1.7 ± 0.3) meV and a homogeneous broadening of γcond = (1.5 ± 0.2) meV. Thus one can conclude that the condensate lifetime is roughly twice as high as for the uncondensed polaritons. The relative occupation of the condensate states changes with in-plane momentum and does not show a regular pattern. Please note that the spectral broadening of the condensate states shown in Fig. 1 (b) and (c) is increased additionally to the lifetime broadening by temporal potential fluctuations due to the short Ti:Sa pump-laser pulses [13]. The homogeneous broadening of the condensate emission was determined via fitting Voigt-oscillators to PL spectra obtained from experiments [10, 14] with a diode laser with a pulse duration of 500 ps acting as quasi continuous excitation and yielding spectrally narrower condensate states. The lowest energy condensate state at 3.320 eV is excluded in Fig. 1 (c) because of its very strong inhomogeneous broadening. This is caused by the relatively long effective lifetime of this condensate state whose energy follows the decreasing background potential as time resolved measurements (not shown here) reveal. Regarding the k -space pattern for increasing excitation power, the situation changes when only photons are detected which were emitted from the central excitation area where the 3 Figure 1. Energy resolved k -space image of the LPB for ∆ = −30 meV at T = 10 K (a) below (0.04 Pth) and (b) above (1.6 Pth) the condensation threshold. One has to note that there is a pale ghost image of the LPB shiftes by about 8 µm−1 to negative k values produced by the monochromator in (a). The modeled uncoupled exciton and the cavity-photon dispersion are marked with X and C, respectively. (c) shows the PL spectrum at k = 0 µm−1 from the red marked region in (b). ∆Er,cond marks the energy spacing between the dispersionless condensate states. In order to specify the spectral position of these states the measured data (black symbols) were fitted with Voigt oscillators (gray lines). pump-induced background potential is strongly inhomogeneous. This is shown in Fig. 2, where one can clearly see that in addition to the dominating LPB two further polariton branches become visible at higher energies even far below the condensation threshold. It is important to note that for increasing excitation density the distance in energy of ∆Er,uncond = (3.3 ± 0.1) meV between these additional branches stays constant as demonstrated more clearly by the k = −2 µm−1 spectra in Figure 2 (b). As the excitation density increases the LPB branches are shifted towards higher energies caused by the repulsive Coloumb interaction between the polaritons according to their excitonic parts and the interaction with hot carriers [15]. One should be aware of the fact that these additional branches are not symmetrical to k = 0 µm−1 because the irregular bar-like laser spot profile, a consequence of the birefringent frequency tripling crystals, destroys rotational symmetry, as shown in Fig. 3 (a). At high excitation densities (10 Pth in Fig. 2 (a) and 2 (b)) the bare cavity-photon mode becomes visible as the particle density reaches the Mott transition. This is accompanied by the disappearance of condensate emission. As mentioned above, at 4 Figure 2. (a) shows the normalized energy resolved k -space images for different excitation densities. The red bar in the first image marks the k channel the spectra in (b) are taken from. Note that the intensity scale is logarithmic. In (b) E0 represents the polariton emission of lowest energy, ∆Er,uncond marks the energy spacing between two emission states which correspond to blueshifted polaritons and C represents the energy of the uncoupled cavity mode. the condensation threshold new dispersionless states appear in the energy resolved k -space images being roughly equally spaced in energy. They belong to the coherent emission of condensates [16] being expelled from the central excitation area [4, 17]. The appearance of these states in the condensed phase has been observed also by other groups [4, 7 -- 9] and was explained in [11, 17] as a consequence of the spatially inhomogeneous background potential induced by the optical pumping. The roughly equal energy distances between these states were attributed to the dynamical balance of in- and outscattering rates for bosonic states of different energy [11]: Γin = Γout (1) Following the ansatz of Wouters et al. [11] the inscattering rate Γin for a polaritonic (or bosonic) state is proportional to the energy distance to the initial state of the scattering process: Γin = κ × ∆Er 5 (2) Here the scattering constant κ has a unit of an inverse action, for simplicity not being normalized to a unit density as in the original paper of Wouters et al. [17]. The outscat- tering rate Γout of a state is limited by its radiative lifetime and is therefore assumed to be proportional to the homogeneous spectral broadening γ: Γout ≃ γ  (3) This means that for states with a doubled outscattering rate (∝ doubled spectral broadening) the energetical distance ∆Er to the next state has to be twice as high to reach a compensation of in- and outscattering rates, as can be seen from equations (1) and (2). Exactly this is the case in our sample when we compare the condensate emission with the emission from the uncondensed polariton ensemble. From our experimental results we can deduce the bosonic scattering constant to be κ = (0.8 ± 0.4) −1. This result, namely that the spectral broadening of a condensate state is roughly the same as the energy spacing between relaxing condensate states was also observed in [4, 7, 8]. The consistent applicability of this simple model to both, the condensed and the uncondensed phase, leads to the conclusion that in a varying background potential discrete relaxation processes take place in both phases which can be described with one scattering constant κ being independent of the particle density within the observed range of more than two orders of magnitude. The relaxation in the potential landscape is accompanied with an effective acceleration towards the peripheral region of the excitation spot where the local potential energy is smaller. To show this, we recorded the laser spot intensity distribution on and the emission of uncondensed polaritons from the sample surface (Fig. 3). It can be clearly seen that the main emission of uncondensed polaritons is coming from the spatial region where the laser light intensity is smaller. Here the uncondensed polaritons could be trapped in the area of the first minimum of the excitation laser spot (see inset of Fig. 3 (b)), providing another explanation for the appearance of the observed discrete polariton states, namely the quantization in a potential trap created by first diffraction minimum of the pump laser spot. But this seems not reasonable as both, the uncondensed and the condensed polaritons should feel more or less the same trapping potential resulting in equal energy spacings for the trapped uncondensed and condensed polaritons (∆Er,uncond ! = ∆Er,cond), which is obviously not the case. 6 Figure 3. The real space image of the laser spot on the sample surface is shown in (a) whereas the LPB emission below the condensation threshold is shown in (b). The inset in (b) shows the normalized intensity profile of the laser spot (red) and the polariton emission (black) along the dashed green line. IV. CONCLUSION We have shown experimental results giving hints that the model [11] of relaxation pro- cesses describing condensed polaritons in a spatially varying potential landscape is also valid for the uncondensed phase. This assumption is supported by real and k -space images revealing discrete polariton energy levels of constant (energy-) spacing in k -space and an accumulation of polaritons in the peripheral region of the pump-induced potential hill shown by real space emission imaging. From our experimental results we could derive a value for the bosonic scattering constant in ZnO MC which is close to −1. V. ACKNOWLEDGEMENT We thank H. Hochmuth and M. Lorenz for their support during the growth of the sample. Funding by DFG within Gru1011/20-2 and FOR1616 (SCHM2710/2-1) is acknowledged. [1] J. Kasprzak, M. Richard, S. Kundermann, A. Baas, P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. Szymanska, R. Andr´e, J. L. Staehli, V. Savona, P. B. Littlewood, B. Deveaud, and L. S. Dang, Nature 443, 409 (2006). 7 [2] A. Amo, J. Lefr`ere, S. Pigeon, C. Adrados, C. Ciuti, I. Carusotto, R. Houdr´e, E. Giacobino, and A. Bramati, Nature Phys. 5, 805 (2009). [3] K. G. Lagoudakis, M. Wouters, M. Richard, A. Baas, I. Carusotto, R. Andr´e, L. S. Dang, and B. Deveaud-Pl´edran, Nature Phys. 4, 706 (2008). [4] E. Wertz, L. Ferrier, D. D. Solnyshkov, R. Johne, D. Sanvitto, A. Lemaıtre, I. Sagnes, R. Grousson, A. V. Kavokin, P. Senellart, G. Malpuech, and J. Bloch, Nature Phys. 6, 860 (2010). [5] F. Li, L. Orosz, O. Kamoun, S. Bouchoule, C. Brimont, P. Disseix, T. Guillet, X. Lafosse, M. Leroux, J. Leymarie, M. Mexis, M. Mihailovic, G. Patriarche, F. R´everet, D. Solnyshkov, J. Z´uniga-P´erez, and G. Malpuech, Phys. Rev. Lett. 110, 196406 (2013). [6] C. Ant´on, T. C. H. Liew, G. Tosi, M. D. Mart´ın, T. Gao, Z. Hatzopoulos, P. S. Eldridge, P. G. Savvidis, and L. Vina, Appl. Phys. Lett. 101, 261116 (2012). [7] T. Guillet, M. Mexis, J. Levrat, G. Rossbach, C. Brimont, T. Bretagnon, B. Gil, R. Butt´e, N. Grandjean, L. Orosz, F. R´everet, J. Leymarie, J. Z´uniga-P´erez, M. Leroux, F. Semond, and S. Bouchoule, Appl. Phys. Lett. 99, 161104 (2011). [8] S. Christopoulos, G. Baldassarri Hoger von Hogersthal, A. J. D. Grundy, P. G. Lagoudakis, A. V. Kavokin, J. J. Baumberg, G. Christmann, R. Butt´e, E. Feltin, J.-F. Carlin, and N. Grandjean, Phys. Rev. Lett. 98, 126405 (2007). [9] D. N. Krizhanovskii, K. G. Lagoudakis, M. Wouters, B. Pietka, R. A. Bradley, K. Guda, D. M. Whittaker, M. S. Skolnick, B. Deveaud-Pl´edran, M. Richard, R. Andr´e, and L. S. Dang, Phys. Rev. B 80, 045317 (2009). [10] H. Franke, C. Sturm, R. Schmidt-Grund, G. Wagner, and M. Grundmann, New J. Phys. 14, 013037 (2012). [11] M. Wouters, T. C. H. Liew, and V. Savona, Phys. Rev. B 82, 245315 (2010). [12] C. Sturm, H. Hilmer, B. Rheinlander, R. Schmidt-Grund, and M. Grundmann, Phys. Rev. B 83, 205301 (2011). [13] D. Sanvitto and V. Timofeev, eds., Exciton Polaritons in Microcavities (Springer, 2012). [14] H. Franke, PLD-grown ZnO-based microcavities for Bose-Einstein Condensa- tion of Exciton-Polaritons, Ph.D. thesis, Universitat Leipzig (2012), http://nbn- resolving.de/urn:nbn:de:bsz:15-qucosa-98174. 8 [15] C. Ciuti, V. Savona, C. Piermarocchi, A. Quattropani, and P. Schwendimann, Phys. Rev. B 58, 7926 (1998). [16] M. Thunert, A. Janot, H. Franke, C. Sturm, T. Michalsky, M. D. Mart´ın, L. Vina, B. Rosenow, M. Grundmann, and R. Schmidt-Grund, arXiv:1412.8667 (2014). [17] M. Wouters, I. Carusotto, and C. Ciuti, Phys. Rev. B 77, 115340 (2008). 9 

1702.03119

1

1702

Relative weight of the inverse spin Hall and spin rectification effects for metallic Py,Fe/Pt and insulating YIG/Pt bilayers estimated by angular dependent spin pumping measurements

We quantify the relative weight of inverse spin Hall and spin rectification effects occurring in RF-sputtered polycrystalline permalloy, molecular beam epitaxy-grown epitaxial iron and liquid phase epitaxy-grown yttrium-iron-garnet bilayer systems with different capping materials. To distinguish the spin rectification signal from the inverse spin Hall voltage the external magnetic field is rotated in-plane to take advantage of the different angular dependencies of the prevailing effects. We prove that in permalloy anisotropic magnetoresistance is the dominant source for spin rectification while in epitaxial iron the anomalous Hall effect has an also comparable strength. The rectification in yttrium-iron-garnet/platinum bilayers reveals an angular dependence imitating the one seen for anisotropic magnetoresistance caused by spin Hall magnetoresistance.

cond-mat.mes-hall

cond-mat

a Relative weight of the inverse spin Hall and spin rectification effects for metallic Py,Fe/Pt and insulating YIG/Pt bilayers estimated by angular dependent spin pumping measurements S. Keller,1 J. Greser,1 M. R. Schweizer,1 A. Conca,1 B. Hillebrands,1 and E. Th. Papaioannou1 Fachbereich Physik and Landesforschungszentrum OPTIMAS, Technische Universitat Kaiserslautern, Erwin-Schrodinger-Str. 56, 67663 Kaiserslautern, Germany (Dated: 13 February 2017) We quantify the relative weight of inverse spin Hall and spin rectification effects occurring in RF-sputtered polycrystalline permalloy, molecular beam epitaxy-grown epitaxial iron and liquid phase epitaxy-grown yttrium-iron-garnet bilayer systems with different capping materials. To distinguish the spin rectification signal from the inverse spin Hall voltage the external magnetic field is rotated in-plane to take advantage of the different angular dependencies of the prevailing effects. We prove that in permalloy anisotropic magne- toresistance is the dominant source for spin rectification while in epitaxial iron the anomalous Hall effect has an also comparable strength. The rectification in yttrium-iron-garnet/platinum bilayers reveals an angular dependence imitating the one seen for anisotropic magnetoresistance caused by spin Hall magnetoresistance. Spintronic bilayers composed of a ferromagnetic (FM) and a nonmagnetic (NM) layer with large spin-orbit- interaction are promising devices for the spin-to-charge conversion for future applications. At ferromagnetic res- onance (FMR) the spin pumping (SP) effect allows for the injection of a pure spin current from the FM into the NM layer1. There, the spin current is converted into a charge current by the inverse spin Hall effect (ISHE)2. A wide range of metallic, semiconducting or insulating ferro-3 and ferrimagnets4 and NM5 materials, like Au, Pd, Ta, W, and Pt, have been investigated up to this point. In metallic FM layers, an overlapping additional effect take place, the so called spin rectification (SR) ef- fect, which hinders the access to the pure ISHE signal. Different approaches for separation have been thoroughly investigated6–8. Thickness variations of the FM and the NM layers show different dependencies for ISHE and SR, but require a lot of effort for producing whole sample series or wedged microstructures. Another method is a sweep of the excitation frequency, which cannot be ap- plied to all experimental setups and requires a careful calibration of the microwave transmission properties of the setup. Also the minimization of the electrical mi- crowave field at the location of the sample by using a microwave cavity is possible, but in most cases only fixed frequencies can be applied. The rotation of the magneti- zation angle by rotating the external magnetic field in- or out-of-plane is one of the most common and practicable methods, with out-of-plane rotation normally requiring larger magnetic fields for thin films7,8. Here, we quan- tify with the help of the in-plane angular dependent spin pumping measurements the ISHE and the SR contribu- tions, mainly anisotropic magnetoresistance (AMR) and anomalous Hall effect (AHE). Therefore, bilayers com- posed of magnetic (Fe, Py, and YIG) and non-magnetic (Pt, Al and MgO) materials have been used. Capping layers with significant spin Hall angle ΘSH (Pt) should show a large ISHE, while materials with small ΘSH (Al) and insulating materials (MgO) should not. All bilayer FIG. 1. Experimental setup and coordinate system: x, y and z are the lab fixed coordinates. The external field ~H rotates in-plane, while the angle ΘH is defined as the angle between z and ~H. The bilayer films are lying in the x and z plane and y is the out-of-plane coordinate. The exciting stripline antenna is parallel to z and is generating an in-plane dynamic magnetic field hx, an out-of-plane field hy and also a dynamic electrical field ez, which induces an electrical current jz in the samples in z direction. Eddy currents jeddy potentially can flow transverse to the microwave electrical field in x direction. The electrical contacts for measuring the DC voltage are ei- ther transverse (Vx) or parallel (Vz) to the stripline antenna. samples with metallic FM (Py/Al, Py/Pt, Fe/MgO and Fe/Pt) have the dimensions of (10 × 10) mm2, while the YIG/Pt sample is of smaller dimensions (2 × 3) mm2. We first address the measurements on polycrystalline Py/Pt and Py/Al bilayers, that is, with presence and ab- sence of ISHE voltage, respectively. We will use the data of this model system to illustrate the angular dependence of the measured signal and the analysis method used 2 FIG. 2. Theoretical in-plane magnetization angular depen- dencies of spin rectification effects and ISHE with contacts transverse to the microwave antenna (x direction) and differ- ent dynamic magnetic field geometries, adapted from Harder et. al.8. ΘH is the magnetic field angle (defined in Fig. 1), AL and AD are the amplitudes of the effects contributing to the symmetric voltage (L: Lorentzian) and the antisymmetric voltage (D: Dispersive). to separate the different contributions. Second, we will present the data for epitaxial Fe/Pt and Fe/MgO samples and we will apply again the same analysis method com- paring the weights of the different contributions with the Py case. Finally, results in YIG/Pt bilayers are presented to compare the situation for a system with an insulating magnetic layer where no AMR or AHE can be present. Prior to concluding, we will present some important re- marks about the validity and limitations of the analysis method based on angular measurements. In the experiment for a fixed excitation frequency and external field angle, the external field amplitude is swept. The voltage measured by lock-in-amplification technique exhibits peaks consisting of symmetric and antisymmet- ric components which are fitted by the following equation for each individual external magnetic field sweep5: Vmeas(H) =Vsym (∆H)2 (H − HFMR)2 + (∆H)2 + Vasym −2∆H(H − HFMR) (H − HFMR)2 + (∆H)2 , (1) where Vsym and Vasym are the amplitude of the symmetric and antisymmetric components, respectively. ∆H is the FIG. 3. Angular dependent spin pumping measurements of Py(12nm)/Al(10nm) (top graph) and Py(12nm)/Pt(10nm) (bottom graph) at 13 GHz excitation frequency with contacts transverse to the direction of the stripline antenna. Black and orange arrows are highlighting the side-maxima/minima orig- inating from AMR. linewidth, H is the applied magnetic field, and HFMR is the corresponding FMR field value. While the SP/ISHE effect contributes only to Vsym, the SR effects contribute to both voltage amplitudes. The relative contribution of AMR to Vsym and Vasym and AHE to Vsym and Vasym is determined by the phase difference between the dy- namic magnetization ~m(t) and the microwave electrical field induced AC current ~j(t) inside the FM layer. This phase difference is not easily accessible5 and the relative contribution of AMR does not necessarily have to be the same as the one of AHE8. To fit the measured voltage amplitudes it is needed to calculate the angular depen- dencies of SP/ISHE and SR (a detailed derivation can be found in7,8). The symmetric as well as the antisymmetric voltage will then be fitted. First let us consider the coordinate system (see Fig. 1), where x and z are the in-plane and y the out-of-plane lab fixed coordinates, ΘH is the angle between the ex- ternal magnetic field ~H and the z axis. The electrical contacts are either in x (transverse to the stripline an- tenna) or z (parallel to the stripline antenna) direction. jz induced by the microwave electrical field ez and jeddy in x direction (explained later) are the in-plane current components. The dynamic magnetic microwave fields hx (in-plane) and hy (out-of-plane) are determined by the microwave stripline antenna. At first the model of the measurements, where the DC voltage is measured in x direction (transverse to the an- tenna, shown in Fig. 1), is discussed: For this measure- ment configuration significant values for jz and hx (in- plane dynamic magnetic field component), and smaller values for hy (out-of-plane dynamic magnetic field com- ponent), which is estimated a magnitude smaller than the in-plane field components, are considered. The the- oretical angular dependencies of the underlying effects are graphically shown in Fig. 27,8. It can be recog- nized that in-plane excited AHE is similar to in-plane excited ISHE bearing only one maximum and one mini- mum, but with different slopes at zero crossing. In-plane excited AMR is showing three maxima/minima where one is of higher amplitude (referred to as main maxi- mum/minimum in the following) and two of smaller am- plitude (referred to as side maxima/minima). Out-of- plane excited AMR is showing two maxima/minima with equal amplitude. Out-of-plane AHE will generate a con- stant offset and out-of-plane ISHE has an identical shape as in-plane AHE and can therefore not be distinguished from it. As to be shown later in the measurements for the Py samples, an additional AMR effect also takes place. This AMR effect is shown in Fig. 2 in blue and is the only one antisymmetric around 0◦. This AMR scales with an electrical current jx perpendicular to the microwave in- duced currents and with an out-of-plane microwave field component hy and is affiliated to eddy currents9. To fit the experimental data all considered effects are linear su- perimposed: ISHE cos3(ΘH) + V hy ,hx sym = V hx V x V hy AHE + V hx,jz AMR cos(2ΘH) + V hy,jeddy V hy ,jz AMR AMR cos(2ΘH) cos(ΘH) + ISHE,AHE cos(ΘH) + asym = V hx V x AHE + AHE cos(ΘH) + V hy V hx,jz AMR cos(2ΘH) cos(ΘH) + AMR cos(2ΘH) + V hy,jeddy V hy ,jz AMR sin(2ΘH). (2) sin(2ΘH). Equations 2 were then used to fit the angular depen- dent spin pumping measurements shown in Fig. 3 and 4 for the of Vsym and Vasym of Py/Al, Py/Pt, Fe/MgO and Fe/Pt bilayers. The voltage amplitudes from the fits of the Py and Fe sample measurements have been summa- rized in Table I for comparison. After familiarizing with the angular dependencies of the ISHE and SR effects the measurements for the Py bi- layers are now discussed: In Fig. 3 we see for the Py/Al 3 FIG. 4. Angular dependent spin pumping measurements of Fe(12nm)/MgO(10nm) (top graph) and Fe(12nm)/Pt(10nm) (bottom graph) at 13 GHz excitation frequency with contacts transverse to the direction of the stripline antenna. sample that the signal is mainly consisting of AMR in the symmetric as well as in the antisymmetric ampli- tude, since the signals exhibit pronounced side-maxima (arrows). The antisymmetric voltage amplitude of Py/Pt has almost identical shape as the one of Py/Al. For both samples the AMR to AHE ratio of the antisymmet- ric voltage is approximately 1 to 4 (see Table I). Py/Al and Py/Pt also show that their side-maxima (arrows) are having not the same amplitudes. This is correlated to AMR caused by eddy currents with an out-of-plane dynamic magnetic field component (see Table I). The measurements of Fe/MgO and Fe/Pt can be seen in Fig. 4. Since epitaxial Fe has a strong magneto- crystalline anisotropy the magnetization will in general not be aligned to the external magnetic field due to the anisotropy fields. The ISHE and SR effects are, however, only dependent on the angle of magnetization ΘM. For this reason, an additional rescaling of the angle axis is re- quired. A numerical analysis has been performed where ΘM has been calculated for the data measured at ΘH. For this K1/Ms (K1: cubic anisotropy constant, Ms: Vsym/Vasym Sample V hx ISHE (µV ) V hy ,hx † (µV ) Vsym Vasym Py/Al Py/Pt Fe/MgO Fe/Pt Py/Al Py/Pt Fe/MgO Fe/Pt 0 amb. 0 amb. 0 0 0 0 ISHE,AHE 1.43 ± 0.04 amb. 6.85 ± 0.12 amb. -1.49 ± 0.05 -0.61 ± 0.03 4.07 ± 0.15 3.55 ± 0.09 amb. V hy AHE (µV ) V hx,jz AMR (µV ) 0.15 ± 0.02 5.63 ± 0.06 0.02 ± 0.02 -0.01 ± 0.05 5.12 ± 0.15 0.12 ± 0.05 0.03 ± 0.03 -5.95 ± 0.07 0.01 ± 0.01 -2.36 ± 0.04 0.07 ± 0.06 -6.20 ± 0.19 0.01 ± 0.04 -7.88 ± 0.11 amb. 4 V hy ,jz AMR (µV ) 0 0 0.18 ± 0.09 -0.25 ± 0.08 (µV ) hy ,jeddy AMR V -1.68 ± 0.03 -0.61 ± 0.02 0 0 0 0 1.02 ± 0.04 0.44 ± 0.02 0.18 ± 0.10 0.13 ± 0.05 0 0 TABLE I. Results of the angular spin pumping measurements: symmetric and antisymmetric voltage amplitudes of Py/Al, Py/Pt, Fe/MgO and Fe/Pt. Items marked with amb. are ambiguous (see text). The voltage of the effects mainly contributing are marked in bold. The voltage marked with † corresponds to the term ∝ cos(ΘH), which is comprised of in-plane AHE and out-of-plane ISHE in the symmetrical voltage and only of in-plane AHE in the antisymmetric voltage (to be seen in Fig. 1). Absolute values between samples are not comparable because of different excitation frequencies. saturation magnetization) has been extracted from the dependence of HFMR on the frequency (Kittel fit10). For RF-sputtered polycrystalline samples which are isotropic ΘH and ΘM are identical. However, in the case of epi- taxial Fe they can differ more than 10◦. After the angle rescaling the angular dependent mea- surements of the Fe bilayers can be discussed: In these bi- layers (Fig. 4) in-plane excited AHE seems to be equally prominent as in-plane excited AMR, as can be recognized from the lack of side-maxima and also from the voltage amplitudes of the fits shown in Table I. This is a main dif- ference to the Py case where AMR is strictly dominant. The AHE excited by the out-of-plane dynamic magnetic field is rather small (as it also is in the case for Py). The difference is not due to the difference in growth (poly- or single-crystalline) of the FM layers. For instance, recent results on also polycrystalline CoFeB/Pt and CoFeB/Ta layers show that there AHE is the only dominant effect while AMR is almost negligible11. The weight of the different spin rectification contribution is reflecting only the strength of the different effects (AHE, AMR) in the ferromagnetic material. The different capping materials (insulator, respectively metal) is changing the relative contribution of the SR effects onto Vasym. Additionally the epitaxial Fe samples, especially Fe/MgO, show char- acteristic features around angles, where the external field is oriented equidistant between the magnetic hard (e.g. 45◦) and easy axis (e.g. 0◦) of Fe. This is due to the intrinsic magnetic anisotropy influencing the angular de- pendencies of ISHE and SR. In addition, to compare with the measurements of the bilayers with metallic FM, a bilayer with an insulator magnetic material was measured with the same setup. For this YIG(100 nm)/Pt(10nm), where AMR and AHE are suppressed, was chosen and the results are shown in Fig. 5 (top graph) and Table II. A surprisingly non- vanishing antisymmetric voltage with angular dependen- cies similar to AMR and AHE can be seen. The sym- metric voltage amplitude seems to be consisting mainly of an ISHE contribution and of a contribution ∝ cos(ΘH) which can be either in-plane ISHE or out-of-plane AHE, as shown in Fig. 2. In order to understand this behav- FIG. 5. Angular dependent spin pumping measurements of YIG(100nm)/Pt(10nm) at 6.4 GHz excitation frequency with contacts transverse (top graph) and parallel (bottom graph) to the direction of the stripline antenna. ior we performed a second measurement with electrical contacts parallel to the stripline antenna (z-direction), measurements shown in Fig. 5 (bottom graph). In this contact geometry the ISHE and SR effects have differ- ent angular dependencies as shown in Fig. 6. Here in- Vsym/Vasym Contacts V hx ISHE (µV ) V hy ISHE (µV ) -1.35⋆ V hx AHE (µV ) amb. Vsym Vasym transverse parallel transverse parallel amb. -1.48† 0 0 -1.82 ± 0.03 -0.19 ± 0.04 0.53 ± 0.03 0.11 ± 0.03 0 0 AMR (µV ) V hy AHE (µV ) V hx,jz -0.02 ± 0.01 0.05 ± 0.02 0.00 ± 0.01 0.20 ± 0.03 -0.01 ± 0.02 0.65 ± 0.04 amb. 0† 5 V hy ,jz AMR (µV ) -0.04 ± 0.02 -0.09 ± 0.02 0.03 ± 0.02 -0.08 ± 0.03 TABLE II. Results of the angular spin pumping measurements: symmetric and antisymmetric voltage amplitudes of YIG/Pt with contacts transverse and parallel to the microwave antenna. Values marked with * are ambiguous (see text). The voltage of the effects mainly contributing are marked in bold. The out-of-plane ISHE voltage with transverse contacts marked with ⋆ has the same shape as the AHE and could easily be confused with it, but the comparison with the measurement with parallel contacts confirms it as an ISHE voltage. In-plane ISHE in the parallel contacts case marked with † cannot be distinguished from in-plane AMR. In-plane AMR and AHE in the symmetrical voltage are estimated small since out-of-plane AMR and AHE are also small despite of relatively high out-of-plane excitation fields. plane excited ISHE and AMR have the same angular de- pendence, but the out-of-plane excited ISHE exhibits an unique cos(ΘH) dependence. To fit the measured data with contacts parallel to the antenna following equations has been used: sym = V hx V z ISHE,AMR sin(2ΘH) cos(ΘH) + V hy ISHE sin(ΘH) + V hx AHE + V hy ,jz V hy AMR sin(2ΘH). AHE cos(ΘH) + (3) AHE cos(ΘH) + V hy asym = V hx V z V hx AMR sin(2ΘH) cos(ΘH) + V hy ,jz AMR sin(2ΘH). AHE + Looking at Table II the out-of-plane ISHE contribution V hy ISHE in transverse contacts (−1.82µV ) has almost the same value as the one of parallel contacts (−1.35µV ). The discrepancy is roughly reflecting the difference in sample width (2 mm) and length (3 mm) where the DC contacts have been applied to. The term ∝ cos(ΘH) used for the fit of Vsym in Fig. 5 (top graph) is therefore con- firmed as out-of-plane ISHE contribution. To understand the enhancement of the out-of-plane magnetic microwave field component it is needed to con- sider that the size of the YIG/Pt sample is much smaller than the Fe and Py samples, its dimensions being closer to the width of the stripline antenna. This changes the distribution of the microwave fields and therefore en- larges the out-of-plane field components to the extent that they can be comparable in magnitude to the in- plane fields. In Fig. 5 (bottom graph) we also see a non-vanishing antisymmetric voltage with AMR-like de- pendence. Other authors also reported rectification ef- fects in YIG/Pt bilayers in spin pumping experiments at room temperature caused by spin Hall magnetoresis- tance (SMR)12–14. SMR is a rectification effect occuring in bilayers consisting of a FM insulator and a NM layer, where a spin current induced by spin Hall effect (SHE) forms a spin accumulation at the interface. When the magnetization is aligned parallel to the polarization of the accumulation, fewer spin currents can enter the FM FIG. 6. Theoretical in-plane magnetization angular depen- dencies of spin rectification effects and ISHE with contacts parallel to the microwave antenna and different dynamic ex- ternal magnetic field geometries, adapted from Harder et. al.8. ΘH is the magnetic field angle (defined in Fig. 1), AL and AD are denoting the amplitudes of the effects contributing to the symmetric voltage (L: Lorentzian) and the antisymmetric voltage (D: Dispersive). layer and spin back-flow induces an additional charge cur- rent by ISHE reducing the resistivity of the NM layer. This in-plane angular dependent change in resistivity in- duces effects similar to AMR and AHE. Additionally the magnetic proximity effect can also contribute to spin rec- tification in YIG/Pt bilayers15,16: a ferromagnetic layer in contact to Pt can induce a finite magnetic moment in Pt near the interface because of the high paramagnetic susceptibility of Pt. This thin ferromagnetic Pt film can also exhibit spin rectification by itself with the same an- gular dependence. This is also true in metallic systems but there the spin rectification generated by the FM layer is dominating. Summarizing this section we have shown that the symmetric voltage of YIG/Pt is mainly consist- ing of ISHE contributions and the antisymmetric voltage is indeed small but not negligible and consisting of SMR induced rectification. Furthermore, the results from the analysis of the YIG/Pt can be used to interpret the ISHE contribution in Py/Pt, seen in Fig. 3: There the side-maxima of the symmetric amplitude are stronger pronounced than the ones of Py/Al. This is due to the reduction of the am- plitude of the main-maximum of Vsym of Py/Pt. The reason for this is the opposite sign of the ISHE to the SR contributions. As shown in Table II ISHE from the YIG/Pt measurements shows a negative sign. The sign of the ISHE voltage is determined by the sign of spin Hall angle, the direction of the spin polarization and the direc- tion of the spin current. They are all the same for both Py/Pt and YIG/Pt, therefore, the voltages generated by ISHE in Py/Pt and YIG/Pt should have the same sign. In Tables I and II some values of the fits have not been shown, indicated by "amb.". An intrinsic limita- tion of the analysis procedure is present when rotating the external magnetic field in-plane and investigating in- plane excited effects. According to Equation 4 an ambi- guity exists with the main in-plane contributions of this measurement configuration: ISHE, AMR and AHE are mathematically linearly dependent. Therefore, the abso- lute values obtained from the fits for the Py/Pt, Fe/Pt and YIG/Pt from Vsym may not be relevant. Neverthe- less, the overall angular dependence and the Vasym data, where no ambiguity is present, support the interpreta- tions shown in this paper. cos(2ΘH) cos(ΘH) =[2 cos2(ΘH) − 1] cos(ΘH) =2 cos3(ΘH) − cos(ΘH). (4) In summary, we have shown that the spin rectification effect does scale differently in Fe, Py and YIG bilayer systems, as summarized in Table I and II: While AMR is more pronounced than AHE in RF magnetron sputtered Py, AHE seems to be equal in magnitude for epitaxial Fe systems. Spin rectification with an angular depen- dence similar to AMR is appearing in the antisymmetric Lorentzian shape in nanometer thin YIG/Pt bilayer films originating from the spin Hall magnetoresistance. The symmetric signal of YIG/Pt is mainly consisting of equal ISHE contributions excited by in- and out-of-plane dy- namic magnetic fields. In epitaxial Fe systems the effects due to non-collinearity between the external field and the 6 magnetization needs to be taken into account. The Carl Zeiss Stiftung is gratefully acknowledged for financial support. 1Y. Tserkovnyak, A. Brataas, and G. E. Bauer, Enhanced Gilbert Damping in Thin Ferromagnetic Films, Phys. Rev. Lett. 88 117601 (2002). 2E. Saitoh, M. Ueda, H. Miyajima, and G. Tatara, Conversion of spin current into charge current at room temperature: Inverse spin-Hall effect, Appl. Phys. Lett. 88 182509 (2006). 3A. Conca, S. Keller, L. Mihalceanu, T. Kehagias, G. P. Dimi- trakopulos, B. Hillebrands, E. Th. Papaioannou, Study of fully epitaxial Fe/Pt bilayers for spin pumping by FMR spectroscopy, Phys. Rev. B 93, 134405 (2016). 4F. D. Czeschka, L. Dreher, M. S. Brandt, M. Weiler, M. Altham- mer, I.-M. Imort, G. Reiss, A. Thomas, W. Schoch, W. Limmer, H. Huebl, R. Gross, and S. T. B. Goennenwein, Scaling behav- ior of the spin pumping effect in ferromagnet/platinum bilayers, Phys. Rev. Lett. 107 046601 (2011). 5A. Azevedo, L.H. Vilela-Leao, R.L. Rodr´ıguez-Su´arez, A.F. Lac- erdo Santos, S.M. Rezende, Spin pumping and anisotropic mag- netoresistance voltages in magnetic bilayers: Theory and exper- iment, Phys. Rev. B 83, 144402 (2011). 6E.Th. Papaioannou, P. Fuhrmann, M.B. Jungfleisch, T. Bracher, P. Pirro, V. Lauer, J. Losch, B. Hillebrands, Optimizing the spin- pumping induced inverse spin Hall voltage by crystal growth in Fe/Pt bilayers, Appl. Phys. Lett. 103, 162401 (2013). 7R. Iguchi, and E. Saitoh, Measurement of spin pumping voltage separated from extrinsic microwave effects, arXiv:1607.04716v1 (2016). 8M. Harder, Y. Gui, and C.-M. Hu, Electrical detec- tion of magnetization dynamics via spin rectification effects, arXiv:1605.00710v1 (2016). 9V. Flovik, and E. Wahlstroem, Eddy current interactions in a Ferromagnet-Normal metal bilayer structure, and its impact on ferromagnetic resonance lineshapes, J. Appl. Phys. 117 143902 (2015). 10C. Kittel, On the Theory of Ferromagnetic Resonance Absorp- tion, Phys. Rev. 73, 155 (1948). 11A. Conca, B. Heinz, M. R. Schweizer, S. Keller, E. Th. Pa- paioannou, and B. Hillebrands, Lack of correlation between the spin mixing conductance and the ISHE-generated voltages in CoFeB/Pt,Ta bilayers, arXiv:1701.09110v1 (2017). 12R. Iguchi, K. Sato, D. Hirobe, S. Daimon, and E. Saitoh, Effect of spin Hall magnetoresistance on spin pumping measurements in insulating magnet/metal systems, Appl. Phys. Express 7, 013003 (2014). 13P. Wang, S. W. Jiang, Z. Z. Luan, L. F. Zhou, H. F. Ding, Y. Zhou, X. D. Tao, and D. Wu, Spin rectification induced by spin Hall magnetoresistance at room temperature, J. Appl. Phys. 109, 112406 (2016). 14Z. Fang, A. Mitra, A. L. Westerman, M. Ali, C. Ciccarelli, O. Cespedes, B. J. Hickey and A. J. Ferguson, Thickness de- pendence study of current-driven ferromagnetic resonance in Y3Fe5O12/heavy metal bilayers, arXiv:1612.06111 (2016). 15M. Caminale, A. Ghosh, S. Auffret, U. Ebels, K. Ollefs, F. Wil- helm, A. Rogalev, and W. E. Bailey, Spin pumping damping and magnetic proximity effect in Pd and Pt spin-sink layers, Phys. Rev. B. 94, 014414 (2016). 16H. Nakayama, M. Althammer, Y.-T. Chen, K. Uchida, Y. Kaji- wara, D. Kikuchi, T. Ohtani, S. Geprags, M. Opel, S. Takahashi, R. Gross, G. E. W. Bauer, S. T. B. Goennenwein, and E. Saitoh, Spin Hall magnetoresistance induced by a nonequilibrium prox- imity effect, Phys. Rev. Lett. 110, 206601 (2013). 

1006.1851

1

1006

Simulating multiple quantum well solar cells

The quantum well solar cell (QWSC) has been proposed as a route to higher efficiency than that attainable by homojunction devices. Previous studies have established that carriers escape the quantum wells with high efficiency in forward bias and contribute to the photocurrent. Progress in resolving the efficiency limits of these cells has been dogged by the lack of a theoretical model reproducing both the enhanced carrier gen- eration and enhanced recombination due to the quantum wells. Here we present a model which calculates the incremental generation and recombination due to the QWs and is verified by modelling the experimental light and dark current-voltage characteristics of a range of III-V quantum well structures. We find that predicted dark currents are significantly greater than experiment if we use lifetimes derived from homostructure devices. Successful simulation of light and dark currents can be obtained only by introducing a parameter which represents a reduction in the quasi-Fermi level separation.

cond-mat.mes-hall

cond-mat

SIMULATING MULTIPLE QUANTUM WELL SOLAR CELLS James P. Connolly a Jenny Nelson a Keith W.J. Barnham a Ian Ballard a C.Roberts a J.S. Roberts c C.T.Foxon d a Blackett Laboratory, Imperial College of Science, Technology and Medicine, London SW7 2BZ b Centre for Electronic Materials and Devices, Imperial College, London SW7 2BZ UK c EPSRC III-V Facility, University of Sheffield, Sheffield S1 3JD UK d University of Nottingham, Nottingham N67 2RD UK Key words: Quantum well solar cell, dark current, light current, efficiency 1 abstract The quantum well solar cell (QWSC) has been proposed as a route to higher efficiency than that attainable by homojunc- tion devices. Previous studies have established that carriers escape the quantum wells with high efficiency in forward bias and contribute to the photocurrent. Progress in resolving the efficiency limits of these cells has been dogged by the lack of a theoretical model reproducing both the enhanced carrier gen- eration and enhanced recombination due to the quantum wells. Here we present a model which calculates the incremental generation and recombination due to the QWs and is verified by modelling the experimental light and dark current-voltage characteristics of a range of III-V quantum well structures. We find that predicted dark currents are significantly greater than experiment if we use lifetimes derived from homostructure de- vices. Successful simulation of light and dark currents can be obtained only by introducing a parameter which represents a reduction in the quasi-Fermi level separation. 2 Introduction The quantum well solar cell (QWSC) [1] consists of a p − i − n structure with nm-wide regions of low bandgap (quantum wells) in the nominally undoped intrinsic region where carriers occupy discrete energy levels. Experiment has shown ([2], [3]) that photo-excited electrons and holes in these quantum wells escape with high efficiency in forward bias and contribute to the photocurrent. If the pho- tocurrent enhancement exceeds the dark current increase this may be a highly efficient solar cell design. Detailed balance arguments in the radiative limit [4] have sug- gested the Voc is determined by the lowest absorption energy which is the well bandgap. Further work [5] however indicates that efficiency enhancements are possible over single bandgap designs in non-radiative dominated structures. This has been confirmed by work ([6] [7]) which showed smaller quasi-Fermi level (QFL) separation in the wells than in the barriers. This suggests that significant efficiency enhancements are possible with this solar cell design. This paper describes QWSC photocurrent and dark current calculations with a view to establishing whether the effect of enhanced photocurrent overcompensates the dark current losses and secondly modelling the overall device performance. The photocurrent is calculated from the diffusion equation applied to photogenerated minority carriers in bulk and well material and is well understood. The dark current aspect is a 1 Electronic mail: [email protected] Shockley-Read-Hall (SRH) calculation including an ideal diode component. The model examines the extent to which the dark current behaviour of QWSCs can be characterised in terms of the quantum well density of states without considering es- cape and capture rates from the quantum wells. Furthermore, it explores the extent to which the QWSC dark current is de- termined by the SRH recombination rates at the centre of the intrinsic region, and for many wells where this methodology is expected to be accurate. This is explored by modelling a range of experimetally measured dark currents in the well charac- terised GaAs/AlxGa1−xAs system. 3 QWSC current voltage model The program "SOL" calculates the incremental light and dark currents due to the quantum wells. It will be shown that the changes in photocurrent and dark current can both be reduced to functions of the quantum well densities of states. The photocurrent model of the QWSC consists of a spectral response calculation as a function of bias, cell dimensions and quantum well energy spectrum. Cell photocurrents can then be calculated for different incident light spectra and power levels. This has been described for the GaAs/AlxGa1−xAs GaAs/InxGa1−xAs InGaPx GaxAsP and InGa0.53xAsxP materials systems [2], [3]. The dark currents measured in QWSC structures in these ma- terials show idealities close to 2. This is characteristic of SRH recombination in the depletion layer. The dark current model applied here is a simplified implementation of a self consistent model by Nelson et al [8] which has succesfully described the dark currents of single quantum well cells for a range of single well positions and i region background doping levels. The carrier generation rate G at wavelength λ, position x, in- cident photon flux F , R the surface reflectivity and absorption coefficient α is defined as G(x, λ) = F (λ)(1 − R(λ))α(x)e−R x (1) In the quantum wells the generation is calculated from Fermi's golden rule governing transition probabilities between electron and hole well states. These are found by solving the effective mass equation for electrons and holes confined in the QW within the envelope function approximation. Omitting con- stants for brevity it is written as αxdx 0 αqw ∼ Nstates X jn=1,jp=1 M 2H(Ejn − Ejp − ¯hω) (2) where the double sum is over initial and final valence and conduction band states labelled jn and jp, M is the transition Hamiltonian, H is the Heaviside step function delta, Ejn and over quantum well allowed states. These are written in terms of two functions θp and θn. Electron and hole populations nq and pq in the wells are then given as follows in terms of the barrier concentrations (5) (6) nq = nbeθn pq = pbeθp The θ functions are as follows θn = kT lnh 1 θp = kT lnh 1 L ( 2 NC L ( 2 NV )1/3 mnq )1/3 mpq mn PN mp PN jn=1 e( jp=1 e( Ecq −Ejn KT Ejp−Ejq KT )i )i where L is the quantum well width. mn,p are electron and hole masses in the barrier and mnq,pq are the masses in the well. These functions contain all information necessary about the quantum wells with the sole exception of carrier lifetimes. We note that they are symmetrical to the well photocurrent en- hancement term of equation 2. Both involve a sum over well states and the necessary information is the same as in equa- tion 2 and reduces to band structure data about the well rela- tive to the barriers. The functions will not be symmetrical for electrons and holes because of different valence/conduction offsets and effective masses. The contributions from neutral layers are included in terms of the ideal diode formalism The electron and hole Fermi-levels are set equal to the valence and conduction band edges. The depletion approximation is made with the result that photo- generated carriers are omitted in the dark current calculation. The dark current of the device in this picture is given by the integral of the SRH recombination rate over the depletion re- gion including p and n layer depletion layers if present, incre- memented by the ideal diode contribution from the charge neutral regions. An example SRH profile is shown in figure 2. The quantum wells show are step-like peaks in the profile, and clearly dominate the dark current of a QWSC. Ejp are the energy levels of the initial and final quantum well states. The photocurrent Ji due to the QW at any λ is found by inte- grating G across the QW layers and assuming 100% escape. The QW spectral response is added on to that of the bulk regions of the device calculated as in [2]. Figure 1 shows the calculated and measured QE for an Alx- Ga1−xAs QWSC comprising 30 GaAs wells together with the same for a control sample which is identical but with- out wells. Under an standard AM1.5 spectrum the theoretical short circuit current (Jsc) for this sample is 108Am−2 versus 116Am−2 calculated from the experimental SR. These results agree with calibrated results from NREL to within 3% in the worst case. Fig. 1. Experimental and calculated SR of a 30well AlxGa1−xAs sample with aluminium x = 30% The SRH [9] recombination rate U (x) in the i region is given by U (x) = τn(p + pt) + τp(n + nt) pn − n2 i (3) where where pt and nt are hole and electron populations in trap levels in the bandgap, p and n are hole and electron populations in the valence and conduction bands and τp and τn are their respective lifetimes. Electron and hole populations in barrier regions of the intrinsic region are expressed in terms of the intrinsic population ni, Fermi levels Ef and the intrinsic energy Ei KT (cid:17) nb = nie(cid:16) Ef n−Ei KT (cid:17) pb = nie(cid:16) Ei−Ef p (4) where ni is given by the effective masses of electrons and holes via the conduction and valence band effective densities of states and the law of mass action in equilibrium, and for nb the subscript b refers to barrier material. It has been shown [8] that the variation in electron and hole populations in well and barrier regions can be reduced to a difference in effective mass within the parabolic band approx- imation and a constant factor modifying the conduction and valence band densities of states which is determined by a sum Fig. 2. SRH recombination profile for the 30 well QWSC of figure 1 for a biasses 0.8V, 1.15 and 1.5V, showing different θn and θp and the reduction of SCR width at high bias. 4 Experiment and Modelling Samples were grown by MBE and MOVPE in the Centre for Electronic Materials and Devices in Imperial College, Phillips Laboratories (Redhill) and the Sheffield III-V Semiconductor Facility. They take the form of p − i − n diodes. Structures having GaAs wells in the i region will be referred to as QWSCs and identical p − i − n structures of homogeneous composi- tion grown in the same growth run are termed controls. The experimental configuration has been described previously [2]. The fitting parameters for dark currents are the non radiative lifetimes. Estimates for these are drawn from studies of barrier and well controls which are GaAs/AlxGa1−xAs and GaAs controls. These lifetimes are then applied to QWSC structures and the resulting prediction compared with experiment. Figure 3 shows modelled and calculated IV for GaAs and Alx- Ga1−xAs controls. The upper curves shows a GaAs control. The model shows SRH and ideal dark currents and their sum. The onset of radiatively dominated behaviour is visible below the built in voltage of 1.44V and is well described. The pa- rameters for the ideal Shockley fit are the same as those used for the SR leaving the non radiative lifetimes as the only pa- rameters. They are chosen equal in this case. Making them different shifts the SRH peak with position and changes the ideality of the dark current in disagreement with experiment. The lifetime of 10ns is in close agreement with lifetimes used in previous work [8]. The lower curves shows similar data and theory for an Alx- Ga1−xAs control with a much shorter SRH lifetime of 0.6ns. Again this is in agreement with previous work by the same factor in a different sample. Modelling an identical control with aluminium mole fractions x = 20% yields a lifetime of 0.05ns. Fig. 4. 30 well QWSC Al 20% (dashed line modified δEf see text) Fig. 3. GaAs control dark current and lifetimes Figures 4 and 5 show dark current modelling for 30 well MBE grown QWSCs with Al composition of 20% and 33% and with lifetimes derived from relevant controls assuming similar material quality. Figure 6 shows the same data for a 50 well QWSC with Al fractions 33%. Fitting the data can again be achieved by re- ducing the QFL separation by the same shift. We have observed that experiment systematically records a lower dark current than predicted by the control lifetimes. This Fig. 5. 30 well QWSC Al 33% (dashed line modified δEf see text) can be explained in a variety of ways by the model. Longer lifetimes in the wells linearly decrease the dark current until the well lifetime is much longer than in the barrier. However there is no reason why the lifetimes in the QW should be greater than in bulk material. The approach we have investigated is the possibility of a lower QFL separation δEf in the quantum wells. We find that a narrowing of the QWFL separation in the wells by a factor symmetrical for holes and electrons models all samples well, as shown by the dashed lines in graphs 4 to 6 despite the fact that the barrier bandgap varies significantly, and that one sample has 50 wells. This reaffirms work [10] on the QFL separation which was found to be significantly smaller in SQW samples at 10meV REFERENCES [1] Barnham KWJ and Duggan G, A New Approach to High Efficiency High-Bandgap Solar Cells, J. Appl. Phys, 67 (7), 1990 [2] M.Paxman, J.Nelson, K.W.J.Barnham, B.Braun, J.P.Connolly, C.Button, J.S.Roberts and C.T.Foxon Modelling the Spectral Response of the Quantum Well Solar Cell J.Appl.Phys. 74, 614 (1993). [3] J.Barnes, J.Nelson, K.W.J.Barnham, J.S.Roberts, M.A.Pate, S.S. Dosanjh, R.Grey, M.Masseur and F.Ghiraldo, Characterisation of GaAs/InGaAs quantum wells using photocurrent spectroscopy, J. Appl. Phys. 79 7775-9 (1996) [4] G. L. Araujo et al., Proc. 12th European Photovoltaic Solar Energy Conference, pp. 1481- 1484 (1994) [5] Keith Barnham, James Connolly, Paul Griffin, Guido Haarpaintner, Jenny Nelson, Alexander Zachariou, Jane Osborne voltage enhancement in Quantum Well Solar Cells, J. apply. Phys 80 1201 (1996) [6] Jenny Nelson, Benjamin Kluftinger, Ernest S.M.Tsui, Keith Barnham et al., Quasi-Fermi Level Separation in Quantum Well Solar Cells, 13th European Photovoltaic Energy Conference, Nice, pp150 (1996) [7] E.S.M Tsui, J.Nelson, K.W.J. Barnham, C.Button and J.S.Roberts, Determination of the Quasi Fermi level separation in single quantum well p-i-n diodes J. Appl. Phys. 80 (7) (1996) [8] Jenny Nelson, Ian Ballard, Keith Barnham, James P. Connolly, Effect of quantum well location on single quantum well p -- i -- n photodiode dark currents, J. Appl. Phys, 86 (10), 1999 [9] Shockley, W. T. Read, and r. N. Hall, Phys. Rev. Lett. 87, 835 (1952) [10] Jenny Nelson, Jenny Barnes, Nicholas Ekins-Daukes, Benjamin Kluftinger, Ernest Tsui and Keith barnham Observation of suppressed radiative recombination in single quantum well p − i − n photodiodes J.Appl.Phys. 82, 6240 (1997). Fig. 6. 50 well QWSC Al 33% (dashed line modified δEf see text) 5 Conclusions The QWSC benefits from an increase in photogeneration in the wells which is counterpointed by recombination in the lower bandgap well regions. In order to study which effect is greater we study photocurrent and dark current and express the modifications to dark and photocurrent in terms of the quantum well density of states. The photocurrent from the wells is determined with no free parameters to good accuracy. The dark current for the control homojunctions is well under- stood. Modelling these gives us an estimate of carrier lifetimes in the SCR, together with verifying the transport parameters used in the QE calculation at the onset of ideal diode be- haviour, if present. The QWSC dark current depends on four lifetimes. We reduce this to two by assuming that hole and electron non radiative lifetimes are equal. In the absence of direct measurements we derive the barrier lifetimes from controls with the barrier composition. Similarly, controls with the well composition set an upper limit on the well lifetimes employed. We see a systematic overestimation of the dark current, in- dicating that the well lifetimes in a QWSC system are longer than expected by one order of magnitude to a first approxi- mation. This can be explained in terms of longer well lifetimes or a smaller QFL separation in the wells. This strongly sug- gests that these structures have the potential to be efficient photoconverters. Finally we note that this formulation of the MQW dark current problem shows shortcomings because of the approximations made. This is particularly noticeable in the case of single well samples. A full treatment needs a self consistent approach to the Poisson equation. This would be noticeable for single quantum well samples where the position of the quantum wells and the background doping is important. We note however that the treatment ap- plies reliably to MQW systems studied here since dark current is determined mainly by the balance between QW and barrier material, in that on average the position of the wells is not critical. This is borne out by the range of GaAs/AlxGa1−xAs samples examined. 

1106.2106

1

1106

Discrete-time quadrature feedback cooling of a radio-frequency mechanical resonator

We have employed a feedback cooling scheme, which combines high-frequency mixing with digital signal processing. The frequency and damping rate of a 2 MHz micromechanical resonator embedded in a dc SQUID are adjusted with the feedback, and active cooling to a temperature of 14.3 mK is demonstrated. This technique can be applied to GHz resonators and allows for flexible control strategies.

cond-mat.mes-hall

cond-mat

Discrete-time quadrature feedback cooling of a radio-frequency mechanical resonator M. Poot∗,1 S. Etaki,1, 2 H. Yamaguchi,2 and H. S. J. van der Zant1, † 1Kavli Institute of Nanoscience, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands 2NTT Basic Research Laboratories, NTT Corporation, Atsugi-shi, Kanagawa 243-0198, Japan (Dated: September 11, 2018) We have employed a feedback cooling scheme, which combines high-frequency mixing with digital signal processing. The frequency and damping rate of a 2 MHz micromechanical resonator embedded in a dc SQUID are adjusted with the feedback, and active cooling to a temperature of 14.3 mK is demonstrated. This technique can be applied to GHz resonators and allows for flexible control strategies. Mechanical systems in the quantum regime [1 -- 3] can be used to answer fundamental questions about quantum measurement, decoherence, and the validity of quantum mechanics in macroscopic objects. This requires a me- chanical resonator which is cooled to such a low tem- perature that it is in its ground state for most of the time. In the past few years tremendous progress has been made in actively cooling resonators [3], mainly by using sideband cooling [2, 4, 5] and active feedback cool- ing [6 -- 8]. The latter technique has mainly been applied to low-frequency (kHz) resonators combined with opti- cal detection. The largest cooling factors have been ob- tained using velocity-proportional feedback, i.e., by feed- ing back the differentiated displacement signal. However, at higher frequencies, delays in the feedback system seri- ously degrade the cooling performance. Here, we demon- strate a feedback cooling technique [9] with a nearly un- limited bandwidth, based on fast digital signal processing (DSP) in combination with single-sideband mixing. A 2 MHz micromechanical resonator with inductive readout is cooled to 14.3 mK using this scheme. Figure 1a shows the device, which consists of a dc SQUID with a part of its loop suspended. This forms a 50 µm long flexural resonator with its fundamental mode around f0 ∼ 2 MHz. The chip is glued onto a piezo el- ement for feedback and actuation, and cooled in a di- lution refrigerator with a minimum bath temperature of 15 mK. By applying an in-plane magnetic field B (green), a displacement of the beam u changes the amount of flux through the dc SQUID loop. When a bias current IB is applied, a change in flux results in a change in the SQUID voltage V . This way, the dc SQUID is a sensi- tive displacement detector [10]. In all measurements pre- sented here, the same working point for the dc SQUID is used, to avoid backaction-induced changes in frequency and damping [11]. The thermal noise of the resonator is used to calibrate the dc SQUID detector. Figure 1b shows the displace- ment noise spectrum Suu measured at two different cryo- stat temperatures T . The thermal motion of the res- ∗Present address: Department of Electrical Engineering, Yale Uni- versity, New Haven, CT 06520, USA †Electronic address: [email protected] FIG. 1: (a) Schematic overview of the SQUID detector (red) with the integrated flexural resonator. (b) Displacement noise spectra without feedback. (c) The resonator temperature ex- tracted from the thermal noise spectra plotted against the mixing chamber temperature. (d) Generic linear system rep- resentation [12] of feedback cooling. (e) The feedback filter consists of a digital signal processor with a single-sideband mixer at the input and output. onator shows up as a peak on top of the imprecision noise floor Sunun . The cryogenic-amplifier-limited dis- placement noise is 2 fm/√Hz at T = 15 mK. The area under the peak is the amplitude of the Brownian mo- tion of the resonator squared. When the temperature of the refrigerator is increased to T = 0.3 K, the spectrum changes: Firstly, the noise floor is higher due to a de- crease in the transduction, dV /du, as the critical current decreases with increasing temperature [10]. Secondly, the peak is higher and wider (the intrinsic damping rate γ0 increases with temperature), indicating that the thermal motion is larger at higher temperatures. The resonator temperature T0 = k0hu2i/kB (k0 = 110 N/m is the spring constant) is plotted against T in Fig. 1c: It follows the cryostat temperature for T > 50 mK (solid line) and sat- urates below this value. 0 /(f 2 To further lower the resonator temperature, active feedback is employed, where the displacement of the res- onator is fed back to it to damp its thermal motion. Fig. 1d illustrates the generic process [3]: The thermal force noise Fth ≡ k0fth drives the resonator whose response is 0 − f 2 + if γ0/2π). Note, that fth and the HR = f 2 other signals are scaled to have the unit of position. The force results in a displacement which is measured by the dc SQUID detector, and imprecision noise un is added to its output v. This signal is fed to the feedback filter with transfer function Gf b. The actuation a is multiplied by A, which consists of the SQUID transduction, an atten- uation (-40 dB) and the piezo responsivity. Finally, the resulting piezo displacement up exerts an inertial force on the resonator. Note, that in practise crosstalk (X) exists between the applied feedback and the detector output, which modifies the system response. To fully characterize the linear system an ac signal is applied to a (see Fig. 1d) and the response at v is mea- sured at the same frequency, while sweeping the driving frequency across the resonance. In this case, the feed- back Gf b is disabled. From this network-analyzer mea- surements the elements A = 1.94· 10−4 exp(−0.73i), and X = 0.26 exp(2.56i) of the linear systems are obtained as well as the parameters of HR: f0 and γ0. The non-zero phase of A is due to the time it takes for the signal to travel through the whole system. If an analog differentia- tor would be used for Gf b, this delay causes the feedback to not be purely velocity proportional thus degrading the cooling performance. The DSP-based feedback presented here can compensate for this effect as demonstrated be- low. Our implementation of the feedback filter Gf b is shown in Fig. 1e. The high-frequency input signal v is split and both branches are mixed with local oscillator (LO) signals with a 90o phase difference between them. This IQ mixer gives both quadratures vs and vc of the input signal. The LO frequency is fLO = 2.0492 MHz so the down-mixed signals oscillate at fR−fLO = 8.9 kHz. They are digitized and the DSP (Adwin Pro II at a sampling rate fs = 820 kS/s) applies the following transformation to the input signals to generate two output signals ac and as: 2 tures do not change faster than the sampling rate, which is equivalent to γR/2π . fs/2. The operation is thus not limited to resonators with frequencies within the band- width of the DSP, allowing feedback cooling of radio and microwave frequency resonators. Note, that the impreci- sion noise floor is still determined by the cryogenic ampli- fier; the contributions from the mixers and discretization are negligible. f /2 = 0 π 2 138 424 Hz 18 Hz 36 Hz γ π0/2 = 88 Hz T = 37 mK 0T = 3 0 0 π 2π θfb FIG. 2: Feedback phase dependence of the resonator fre- quency (a), damping rate (b), and resonator temperature (c) for gf b = 0.1. The dashed lines indicate their measured val- ues in the absence of feedback, (f0, γ0, and T0 resp.); the solid line is the phase dependence calculated using indepen- dent measurements. The feedback modifies the resonator response from HR to its closed-loop form H ′ R [3]: as (cid:19) = gf b(cid:18) cos θf b − sin θf b (cid:18) ac cos θf b (cid:19)(cid:18) vc vs (cid:19) sin θf b (1) H ′ R = These quadratures are then up-converted by the LO fre- quency with a second IQ mixer. The final result is a signal a at the original frequency that is phase-shifted by the feedback phase θf b and multiplied by the feedback gain gf b, i.e. Gf b = gf b exp(iθf b). The only frequency requirement for this mixing scheme is that the quadra- f b = g′ where G′ f b) = Gf b/(1 − XGf b) is the feedback filter modified by the crosstalk. The real part of G′ f bA modifies the resonance frequency from f0 to fR ≈ f0(1 − Re[G′ f bA]/2), whereas the imaginary part changes the damping from γ0 to γR ≈ γ0−2πf0Im[G′ f bA]. Both the frequency shift and the change in damping de- f 2 0 0 − f 2 + if γ0/2π − f 2 f 2 f b exp(iθ′ , (2) 0 G′ f bA pend periodically on the phase of G′ f bA and the maxi- mum frequency shift is half the maximum damping rate change. In order to achieve optimal cooling the feedback phase is varied for a fixed feedback gain as shown in Fig. 2. The feedback gain is chosen sufficiently small so that G′ f b ≈ Gf b. At every point the thermal noise spectra are measured and fitted to obtain the resonance frequency (top), the damping rate (center), and the resonator tem- perature (bottom). The resonance frequency and damp- ing rate show the expected sinusoidal dependence on the feedback phase. The amplitude of the frequency shift is half of that of the change in damping, consistent with the discussion above. The phase where the damping is max- imized, coincides with the lowest resonator temperature and zero frequency shift. At this phase the system delay is compensated and a pure velocity-proportional feedback is applied to the resonator (i.e. ∠AGf b = −π/2). The phase dependencies can also be calculated without any free parameters by using the values from the network characterization (Fig. 1d). Figure 2 show that these are in good agreement with the feedback results. ) r s m V m ( n o i t a u t c a 100 10 1 0.1 Full bandwidth Δf = 50 kHz Δf = 10 kHz 0.01 0.0 0.5 g fb 1.0 fb θ = 1.69 π Δf = 10 kHz fb 3 To further cool the resonator, the feedback gain is in- creased at the optimal phase as indicated in Fig. 3. First the resonator temperature decreases rapidly with increas- ing gain due to the increased damping rate. However, by increasing the gain further, more of the imprecision noise un is fed back as force noise. This causes a steady in- crease in TR for large gf b. The minimum temperature that can be reached is set by Sunun and the solid line shows the predicted curve for velocity-proportional feed- back [7, 13] calculated with the experimental parame- ters. The achieved minimum of 14.3 mK is close to the predicted lowest temperature of 14.0 mK. Note, that a temperature of 14.3 mK corresponds to an average ther- mal phonon occupation of ¯n ≈ kBTR/hfR = 138 for a 2 MHz resonator. The heterodyne DSP-based technique employed in this work thus successfully reaches the low- est temperature possible for the standard fully-analog ap- proach, but now applied to a high-frequency resonator. Another advantage of DSP-based feedback is that the transformation of vs and vc to ac and as can be designed with almost arbitrary transfer characteristics, allowing implementation of optimal control strategies [14]. In the measurements in Fig. 3, the input signal is digitally fil- tered using a Fourier transform filter which is centered around f0 and a tunable filter bandwidth ∆f . The fil- ter reduces the bandwidth of the feedback which pre- vents excess signal output outside the resonator band- width that may overload the detector or the amplifiers. The inset of Fig. 3 shows the root-mean-square output voltage as a function of gf b for three values of ∆f . For the full bandwidth (fs/2 = 410 kHz) an instability oc- curs around gf b = 0.23, which affects the cooling. The 10 kHz bandwidth, which is used for the cooling curve of Fig. 3, has two orders of magnitude less actuation compared to the full bandwidth, enabling efficient cool- ing without affecting the closed-loop response as long as ∆f ≫ γR/2π. This again illustrates the versatility of our digital quadrature feedback cooling platform. FIG. 3: Resonator temperature as a function of feedback gain, showing both the feedback measurements (symbols) and cal- culations for pure velocity-proportional feedback (solid line). The inset shows the root-mean-squared value of the actuation signal (pha2 si) as a function of gain for different filter bandwidths. ci + ha2 We thank Hidde Westra, Khasahir Babei Gavan, Abah Obinna and Joris van der Spek for their help with the measurements. This work was supported in part by FOM, NWO (VICI grant), NanoNed, a EU FP7 STREP project (QNEMS), and JSPS KAKENHI (20246064 and 23241046). [1] A. D. O'Connell, M. Hofheinz, M. Ansmann, R. C. Bial- czak, M. Lenander, E. Lucero, M. Neeley, D. Sank, H. Wang, M. Weides, et al., Quantum ground state and single-phonon control of a mechanical resonator , Nature 464, 697 (2010). [2] J. D. Teufel, T. Donner, D. Li, J. H. Harlow, M. S. All- man, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds, Sideband cooling mi- cromechanical motion to the quantum ground state, arXiv:1103.2144v1 (2011). [3] M. Poot and H. S. J. van der Zant, Mechanical systems in the quantum regime, Phys. Rep., submitted (2011). [4] O. Arcizet, P.-F. Cohadon, T. Briant, M. Pinard, and A. Heidmann, Radiation-pressure cooling and optome- chanical instability of a micromirror , Nature 444, 71 (2006). [5] T. Rocheleau, T. Ndukum, C. Macklin, J. B. Hertzberg, A. A. Clerk, and K. C. Schwab, Preparation and detec- tion of a mechanical resonator near the ground state of motion, Nature 463, 72 (2010). [6] D. Kleckner and D. Bouwmeester, Sub-kelvin optical cool- ing of a micromechanical resonator , Nature 444, 75 (2006). [7] M. Poggio, C. L. Degen, H. J. Mamin, and D. Rugar, Feedback Cooling of a Cantilever's Fundamental Mode be- low 5 mK , Phys. Rev. Lett. 99, 017201 (pages 4) (2007). [8] B. A. et al., Observation of a kilogram-scale oscillator near its quantum ground state, New J. Phys. 11, 073032 (2009). [9] T. E. Kriewall, J. L. Garbini, J. A. Sidles, and J. P. Jacky, Heterodyne Digital Control of a High-Frequency Micromechanical Oscillator , Journal of Dynamic Sys- tems, Measurement, and Control 128, 577 (2006). [10] S. Etaki, M. Poot, I. Mahboob, K. Onomitsu, H. Yam- aguchi, and H. S. J. van der Zant, Motion detection of 4 a micromechanical resonator embedded in a d.c. SQUID, Nat Phys 4, 785 (2008). [11] M. Poot, S. Etaki, I. Mahboob, K. Onomitsu, H. Yam- aguchi, Y. M. Blanter, and H. S. J. van der Zant, Tunable Backaction of a DC SQUID on an Integrated Microme- chanical Resonator , Phys. Rev. Lett. 105, 207203 (2010). [12] A. V. Oppenheim, A. S. Willsky, and S. Hamid, Signals and Systems, Prentice Hall signal processing series (Pren- tice Hall, Englewood Cliffs, NJ, 1997), 2nd ed. [13] K. H. Lee, T. G. McRae, G. I. Harris, J. Knittel, and W. P. Bowen, Cooling and Control of a Cavity Opto- electromechanical System, Phys. Rev. Lett. 104, 123604 (2010). [14] K. J. Bruland, J. L. Garbini, W. M. Dougherty, and J. A. Sidles, Optimal control of force microscope cantilevers. II. Magnetic coupling implementation, J. Appl. Phys. 80, 1959 (1996). 

1901.06973

1

1901

Coherent anti-Stokes Raman Spectroscopy of single and multi-layer graphene

We report stimulated Raman spectroscopy of the G phonon in both single and multi-layer graphene, through Coherent anti-Stokes Raman Scattering (CARS). The signal generated by the third order nonlinearity is dominated by a vibrationally non-resonant background (NVRB), which obscures the Raman lineshape. We demonstrate that the vibrationally resonant CARS peak can be measured by reducing the temporal overlap of the laser excitation pulses, suppressing the NVRB. We model the observed spectra, taking into account the electronically resonant nature of both CARS and NVRB. We show that CARS can be used for graphene imaging with vibrational sensitivity.

cond-mat.mes-hall

cond-mat

Coherent anti-Stokes Raman Spectroscopy of single and multi-layer graphene A. Virga1,2,∗, C. Ferrante3,1,∗,†, G. Batignani1, D. De Fazio4, A. D. G. Nunn2, A. C. Ferrari4, G. Cerullo5, T. Scopigno1,3,‡ 1Dipartimento di Fisica, Universit´a di Roma "La Sapienza", I-00185, Roma, Italy 2Istituto Italiano di Tecnologia, Center for Life Nano Science @Sapienza, Roma, I-00161, Italy 3Istituto Italiano di Tecnologia, Graphene Labs, Via Morego 30, I-16163 Genova, Italy 5IFN-CNR, Dipartimento di Fisica, Politecnico di Milano, P.zza L. da Vinci 32, 20133 Milano, Italy 4Cambridge Graphene Centre, Cambridge University, 9 JJ Thomson Avenue, Cambridge CB3 OFA, UK ∗These authors contributed equally to this work †[email protected] and ‡[email protected] We report stimulated Raman spectroscopy of the G phonon in both single and multi-layer graphene, through Coherent anti-Stokes Raman Scattering (CARS). The signal generated by the third order nonlinearity is dominated by a vibrationally non-resonant background (NVRB), which obscures the Raman lineshape. We demonstrate that the vibrationally resonant CARS peak can be measured by reducing the temporal overlap of the laser excitation pulses, suppressing the NVRB. We model the observed spectra, taking into account the electronically resonant nature of both CARS and NVRB. We show that CARS can be used for graphene imaging with vibrational sensitivity. Single-layer graphene (SLG) has a high nonlinear χ(3) ∼ 10−10 e.s.u for har- third-order susceptibility: monic generation[1] and χ(3) ∼ 10−7 e.s.u. for fre- quency mixing[2], where one electrostatic unit of charge (1 e.s.u), in standard units (SI) is[3]: χ(3)(SI)/χ(3)(e.s.u.) = 4π/(3× 104)2. This is up to seven orders of magnitude greater than those of dielectric materials such as silica (χ(3) = 1.4×10−14 e.s.u[4]). This property is due to opti- cal resonance with interband electronic transitions[5] and has led to the observation of gate-tunable third-harmonic generation[1] and nonlinear four-wave mixing[2, 6, 7] (FWM, i.e. the third-order processes whereby an elec- tromagnetic field is emitted by the nonlinear polariza- tion induced by three field-matter interactions[3]). FWM can be exploited for for graphene imaging, with an im- age contrast of up to seven orders of magnitude[2] higher than that of optical reflection microscopy[8]. However, FWM-based imaging reported to date in graphene[2] lacks chemical selectivity and does not provide the same wealth of information brought about by the vibrational sensitivity of Raman spectroscopy[9, 10]. Coherent anti-Stokes Raman scattering (CARS)[11 -- 14] is a FWM process that exploits the nonlinear interac- tion of two laser beams, the pump field EP at frequency ωP and the Stokes field ES at frequency ωS < ωP , to ac- cess the vibrational properties of a material. As shown in Fig.1a, when the energy difference between the two pho- tons matches a phonon energy (ωP −ωS = ωv), the in- teraction of the laser pulses and the sample results in the generation of vibrational coherences[4]. While sponta- neous Raman (SR) scattering is an incoherent signal[15], since the phases of the electromagnetic fields emitted by individual scatterers are uncorrelated[15], in CARS, atomic vibrations are coherently stimulated, i.e. atoms oscillate with the same phase[4], potentially leading to a signal enhancement of several orders of magnitude de- pending on incident power and scatterer density[16, 17]. The same combination of optical fields used for CARS can generate another FWM signal, a non-vibrationally resonant background (NVRB)[2], Fig.1b. In both pro- cesses, the optical response consists of a field emitted at the anti-Stokes frequency ωas = 2ωP − ωs[4]. However, the interference of the two effects usually generates an an additional contribution which is dispersive with re- spect to the emitted optical frequency, i.e. shaped as the first derivative of a peaked function (resembling the real part of the refractive index around a resonance), which introduces an asymmetric distortion of the Raman peak profile in the region ωas = ωP + ωv [18]. In the biological field[16, 19], a wealth of studies has demonstrated the potential of CARS for fast imaging[16, low as∼ 17, 20], with pixel acquisition times as 0.16µs[19], thus allowing for video-rate microscopy[19]. By contrast, there are only a few reports to date of CARS imaging of micro-structured materials (such as polyethy- lene blend[21], multicomponent polymers[22], cholesterol micro-crystals[23]) and nano-structured ones (patterned gold surfaces[24], single wall nanotubes[25, 26], highly oriented pyrolytic graphite[27]). Such studies, performed with pixel acquisition times down to∼ 2µs[28], have shown the ability of CARS to identify chemical hetero- geneities on sub-micrometer scales and characterize sin- gle particles that are part of a larger domain, enabling e.g. to visualize microscopic domains (polystyrene, poly- methyl methacrylate, and poly-ethylene terephthalate) in the case of the above mentioned polymer mixtures[29], or to provide detailed maps of microcrystal orientation in organic matrices (e.g. cholesterols in atherosclerotic plaques[23]). In graphene, despite the large χ(3)[1, 2], no CARS peak 9 1 0 2 n a J 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 3 7 9 6 0 . 1 0 9 1 : v i X r a 2 NVRB under electronically resonant conditions, which allows vibrational imaging with signal levels as large as those of the third-order nonlinear response. SLG is grown on a 35µm copper (Cu) foil, follow- ing the process described in Ref.33. The substrate is heated up to 1000◦C and annealed in hydrogen atmo- sphere (H2, 20 sccm) for 30 minutes at ∼200 mTorr. Then, 5 sccm of methane (CH4) are let into the cham- ber for the following 30 minutes to enable growth[33, 34]. The sample is then cooled back to room temperature in vacuum (∼1 mTorr) and unloaded from the chamber. SLG is subsequently transferred on a glass substrate by a wet method. Poly-methyl methacrylate (PMMA) is spin coated on the SLG/Cu and floated on a solution of ammonium persulfate (APS) and deionized water. When Cu is etched[34, 35], the PMMA membrane with attached SLG is then moved to a beaker with deionized water for cleaning APS residuals. The membrane is subsequently lifted with the target substrate. After drying, PMMA is removed in acetone leaving SLG on glass. SLG is char- acterized by SR after transfer using a Renishaw InVia spectrometer at 514nm. The position of the G peak, Pos(G), is∼1582cm−1, while FWHM(G)∼14cm−1. The 2D to G peak area ratio is∼5.3, indicating a p-doping af- ter transfer∼250meV[36, 37], which corresponds to a car- rier concentration∼5·1012cm−2. FLG flakes are produced by micromechanical cleavage from bulk graphite[38]. The bulk crystal is exfoliated on Nitto Denko tape. The FLG G peak is∼1580cm−1. The D peak is negligible. The 2D peak shape indicates this is Bernal stacked FLG[9, 10]. For CARS experiments, we use a two-modules Toptica FemtoFiber Pro source, with two Erbium-Doped fiber amplifiers (EDFA) at∼ 1550nm generating 90fs pulses at 40MHz, seeded by a common mode-locked Er:fiber oscillator[39], Fig.2. In the first branch (FemtoFiber pro NIR), 1ps pulses at 784nm (pump pulse, PP) are pro- duced by second-harmonic spectral compression[40] in a 1cm Periodically Poled Lithium Niobate (PPLN) crys- tal. In the second branch (FemtoFiber pro TNIR), the amplified laser passes through a nonlinear fiber (NLF), wherein a supercontinuum (SC) output is generated. The SC spectral intensity can be tuned with a motorized Si- prism-pair compressor. A PPLN crystal with a fan-out grating (a poling period changing along the transverse direction) is exploited to produce broadly tunable (from 840 to 1100nm) narrowband 1ps Stokes pulses (SP), with a power< 10mW. A dichroic mirror (DM) is used to combine the two beams, whose relative temporal delay is tuned with an optical delay line (DL). A long-working distance 20x objective (O, numerical aperture NA=0.4) focuses the pulses onto the sample (S). The generated and transmitted light is collected by a condenser (C) and the PP and SP are filtered out by a short-pass filter (F). The total FWM signal is collected with an optical mul- tichannel analyser (OMA, Photon Control SPM-002-E). A dichroic mirror reflects the SP in order to measure its Figure 1. Schematic of CARS and NRVB third-order non- linear processes. Interaction with pulses ωP , ωS, results in blue-shifted a) CARS and b) NVRB contributions at ωas = 2ωP − ωS. Since in CARS a vibrational coherence is stim- ulated by two consecutive interactions with the pump and Stokes fields, their frequency difference must correspond to a Raman active mode, ωP − ωS = ωv. c,d) Constraints for the temporal sequence of the field-matter interactions (repre- sented by circles at the top of the pulse envelopes), for CARS and NVRB. In NVRB, the 3 interactions generating the χ(3) signal happen within the few fs electronic dephasing time[15]. In CARS, the third interaction can occur over the much longer vibrational dephasing time (a few ps)[15], within the pump pulse (PP) temporal envelope. profiles, equivalent to those measured in SR, have been observed to date, to the best of our knowledge. We re- ported SR with single-color pulsed excitation[30], using the same picosecond lasers usually adopted for CARS[19]. However, in order to measure CARS, a combination of pulses with different colors must be adopted[31]. By scanning the pulse frequency detuning in a two-color ex- periment, a dip has been observed in the third order nonlinear spectral response of SLG at the G phonon fre- quency. This was interpreted as an anomalous antires- onance and phenomenologically described in terms of a Fano lineshape[32]. Here we use two 1ps pulses (see inset of Fig.2) to ex- plore FWM in SLG and few-layer graphene (FLG). We experimentally demonstrate and theoretically describe how the inter-pulse delay, ∆T (Figs.1c-d) can be used to modify the relative weight of CARS and NVRB compo- nents that simultaneously contribute to the FWM, thus recovering the G-band Raman peak profile. We show that the dip in the nonlinear optical response around the vibrational resonance is due to the interplay of CARS and 3 intensity (Is) with a powermeter (P). FWM spectra are obtained by scanning ωS around ωP −ωv (at fixed ωP ) to probe the G-band phonon frequency, ωv = ωG. Fig.3 dis- plays the FWM intensity, normalized to Is, for different ∆T . In both SLG and FLG measurements, for ∆T shorter than the vibrational dephasing time τ ∼1ps[41], i.e. the characteristic time of coherence loss[15], a Lorentzian dip at ωas = ωP + ωG appears on top of a background[32]. For ∆T > 2ps, while the total FWM signal decreases by nearly two orders of magnitude, the dip observed in FLG at ∆T ∼ 0ps evolves into a Raman peak shape at the G-phonon energy. No dispersive features are observed at any ∆T , unlike what normally expected for the interfer- ence between NVRB and CARS[18]. Here we use pulses with duration δt ∼ 1ps since this allows us to scan the inter-pulse delay across the vibrational dephasing time τ to suppress the NVRB cross section more than the vibra- tional contribution, while minimizing the spectral broad- ening due to the finite pulse duration 1/δt (cid:39)15cm−1 [15]. Since both CARS and NVRB signals depend quadrat- ically on the number of scatterers[29], the SLG signal intensity is significantly reduced with respect to FLG (Fig.3), with a lower signal-to-noise ratio, hampering the detection of peak-shaped vibrational resonances expected for ∆T > 1.4ps. The data in Fig.3 can be qualitatively understood as follows. The anti-Stokes signal, I(ωas), generated by CARS and NVRB, is proportional to the square modulus of the electric field emitted by the third-order polariza- tion, P (3)[4], as: I(ωas) ∝ P (3) CARS(ωas) + P (3) N V RB(ωas)2. (1) P E∗ CARS and NVRB signals are simultaneously generated by two light-matter interactions with the PP and a sin- gle interaction with the SP, with different time order- ing, Figs.1a-b. Consequently, P (3) ∝ E2 S, where * indicates the complex conjugate. However, the temporal constraints for such interactions are significantly differ- ent for the two cases. As shown in Figs.1c-d, in the case of NVRB the three interactions must take place within the dephasing time of the involved electronic excitation, which in SLG is∼10fs[42, 43], i.e. much shorter than the pulses duration (δt ∼1ps). Hence, P (3) N V RB(ωas) is only generated during the temporal overlap between the two pulses P (3) S(t) (the three field inter- actions, in a representative NVRB event, are indicated by three nearly coincident dots in Fig.1d). In CARS, the electronic dephasing time only constrains the lag between the first two interactions that generate the vibrational coherence (the two stimulating-field interactions are rep- resented by the two nearly coincident dots in Fig.1c). This can be read out by the third field interaction within the phonon dephasing time, τ ∼ 1ps[41] (indicated, for a representative CARS event, by the third dot in P (t− ∆T )E∗ N V RB ∝ E2 Figure 2. CARS setup. EDFA, erbium-doped fiber ampli- fied; NLF, nonlinear fiber for SC generation; DL, delay line; DM, dichroic mirror; O, objective; S, sample; C, condenser; P, powermeter; F, filter; OMA, optical multichannel analyzer. Purple, green, and red lines represent the beam pathways of 1550nm, 784 nm (PP) and tunable SP. The second harmonic autocorrelation of PP (green line) and SP (red line) are re- ported in the inset. The black dashed line simulates the au- tocorrelation obtained by using the profile from the best fit (colored dashed lines) in Fig.3. Figure 3. CARS spectra of (a) FLG and (b) SLG as a function of Raman shift (ν − νP ) at different ∆T between the beams at tunable ωS and fixed ωP . In (a), colored dashed lines are fits to the data using Eq.1 and the nonlinear polarization obtained from Eqs.19-20. Vertical black dashed lines indicate three energies (ν1,2,3 − νP = 1545, 1576, 1607 cm−1), taken as reference for the FLG CARS images in Fig.5. 155016001650Raman Shift (cm-1)100101102I (arb.units)FLGa)155016001650Raman Shift (cm-1)100I (arb.units)SLGb)0.00 ps1.07 ps1.74 ps2.08 ps2.41 ps1.41 ps2.75 ps0.00 ps0.40 ps0.60 ps0.80 ps1.01 ps1.21 ps1.41 psS(t)(cid:82) t CARS ∝ EP (t − ∆T )E∗ −∞ EP (t(cid:48) − Fig.1d). Thus, P (3) ∆T )e−t(cid:48)/τ dt(cid:48)[44]. Therefore, ∆T can be used to control the relative weights of P (3) N V RB[17, 44 -- 48]. For positive time delays within a few τ , P (3) is progressively enhanced, as shown in Fig.4a,b,c. CARS and P (3) CARS/P (3) N V RB The system response can be evaluated through a density-matrix description of P (3)(ω, ∆T )[15]. In the presence of a temporal delay between PP and SP, 4 their electric fields can be written as[3]: EP (t, ∆T ) = AP (t, ∆T )e−iωP t and ES(t, 0) = AS(t, 0)e−iωS t, where AP/S(t, ∆T ) indicates the PP/SP temporal envelope. By Fourier transform, the fields can be expressed in the fre- −∞ EP (t, ∆T )eiωtdt −∞ ES(t, 0)eiωtdt, which can be used quency domain as: EP (ω, ∆T ) = (cid:82) +∞ and ES(ω, 0) = (cid:82) +∞ to calculate P (3) CARS(ω, ∆T ) as[15, 49]: (cid:90) ∞ (cid:90) ∞ CARS(ω, ∆T ) ∝ −ηCARS P (3) (cid:90) ∞ dω3 −∞ dω2 −∞ −∞ AP (ω3, ∆T ) AP (ω1, ∆T ) A∗ S(ω2, 0)δ(ω − 2ωP + ωS − ω3 − ω1 + ω2) dω1 (ωP + ω3 − ¯ωba) (ωP − ωS + ω3 − ω2 − ¯ωca) (2ωP − ωS + ω3 − ω2 + ω1 − ¯ωda) (2) where ηCARS = nCARSµbaµcbµcdµad, µij is the transi- tion dipole moment between the i and j states, nCARS is the number of scatterers involved in the CARS process, ¯ωij = ωij − iγij = ωi − ωj − iγij, ωij = ωi − ωj is the en- ergy difference between the levels i and j, and γij = τ−1 is the dephasing rate of the i(cid:105)(cid:104)j coherence[15]; a and c denote the vibrational ground state g, 0(cid:105), and the first vibrational excited level, g, 1(cid:105), with respect to the elec- tronic ground state g(cid:105) (π band). In our experiments, c corresponds to the G phonon at q ∼ 0, b and d indicate the vibrational ground state, e, 0(cid:105), and the first vibra- tional excited level, e, 1(cid:105), with respect to the excited electronic state e(cid:105) (π∗ band). ij Using the conservation of energy represented by the δ-distribution in Eq.2 and integrating over ω2, we get: CARS(ω, ∆T ) ∝ −ηCARS P (3) AP (ω3, ∆T ) AP (ω1, ∆T ) A∗ (cid:90) ∞ (cid:90) ∞ S(2ωP − ωS − ω + ω3 + ω1, 0) −∞ −∞ dω1 dω3 (ωP + ω3 − ¯ωba) (ω − ωP − ω1 − ¯ωca) (ω − ¯ωda) (3) Defining ν = ω/(2πc), the third-order nonlinear polariza- tion can be expressed as a function of the Raman shift (ν − νP ) as P (3)(ω, ∆T ) = P (3)(2πcν, ∆T ). The ωca level in the denominator of Eq.3 is the fre- quency of the Raman mode coherently stimulated in CARS, while ωba and ωda are frequency differences be- tween the electronic levels. In the case of real levels, resonance enhancement occurs[15]. In view of the opti- cal nature of the involved phonons (q ∼ 0), and due to momentum conservation, only one electronic level must be included in the calculation and, consequently, the non- linear response can be derived for ωba = ωdc = ωP . In a similar manner, P (3) N V RB can be expressed as[15]: N V RB(ω, ∆T ) ∝ −ηNVRB P (3) AP (ω1, ∆T ) AP (ω2, ∆T ) A∗ (cid:90) ∞ (cid:90) ∞ S(2ωP − ωS − ω + ω1 + ω2, 0) −∞ −∞ dω1 dω2 (ωP + ω1 − ¯ωea) (2ωP + ω1 + ω2 − ¯ωea) (ω − ¯ωea) (4) where ηNVRB = nNVRBµea4, nNVRB is the number of scatterers involved in the NVRB process, and ωea is the energy of the electronic excited level involved in the NVRB process. Since the cross section of third-order nonlinear processes in graphene is enhanced by increasing the photon energy[50, 51], we consider only the dominant case, i.e., νea = 2νP . We describe the spectral FWM response assuming monochromatic fields with no inter-pulse delay: EP (ω) = EP · δ(ω − ωP ), ES(ω) = ES · δ(ω − ωS). From Eqs.3,4, the CARS and NVRB nonlinear polarizations can be ex- pressed as[4, 15]: CARS(ω) ∝ − P (3) = χ(3) (ωP − ¯ωba)(ω − ωP − ¯ωca)(ω − ¯ωda) CARSE2 ηCARSE2 P E∗ P E∗ = S S N V RB(ω) ∝ − P (3) = χ(3) (ωP − ¯ωea)(2ωP − ¯ωea)(ω − ¯ωea) N V RBE2 P E∗ S ηN V RBE2 P E∗ S (5) = (6) which can be used to calculate the total FWM spectrum according to Eq.1. Fig.4a plots the electronically non- resonant case. The CARS polarization, defined by Eq.5, is a complex quantity: the real part has a dispersive line- shape, while the imaginary part peaks at the phonon fre- quency ωca. The NVRB polarization, defined by Eq.6, is 5 Figure 4. CARS and NVRB spectral profiles for (a,b,c) electronically non-resonant and (d,e,f) resonant regimes, as derived from Eqs.5,6, considering τba = τda = τea = 10fs[43], γca = F W HM (G)/2 = 6cm−1[41]. (a,c) Normalized (cid:60)(P (3) CARS) and (cid:60)(P (3) In (c,e), selected spectra corresponding to three ηNVRB/ηCARS from the colormap are reported. N V RB). Colormaps in (b,e) generalize (a,c) for different ηNVRB/ηCARS, as for Eq.7 and 8. N V RB), (cid:61)(P (3) CARS), (cid:61)(P (3) a positive real quantity. Accordingly, the FWM spectrum in the electronically non-resonant condition, I(ωas)N R, can be written as[15]: I(ωas)N R ∼ P (3) ∝ χ(3) CARS2 + 2(cid:60)(P (3) CARS2 + 2(cid:60)(χ(3) N V RB2 + P (3) N V RB2 + χ(3) N V RB)(cid:60)(P (3) CARS) N V RB)(cid:60)(χ(3) CARS) (7) and it can be significantly distorted by the third term in Eq.7 depending on the relative weight of the two cor- responding susceptibilities. The maximum of the signal, when the dispersive contribution is dominant, can be fre- quency shifted from the phonon frequency. This is the most common scenario, in which the dispersive lineshapes hampering a direct access to the vibrational characteri- zation of the sample in terms of phonon frequencies and lifetimes. Such limitation is particularly severe when χ(3) N RV B is comparable to χ(3) CARS and the NVRB and CARS contributions have the same order of magnitude. This condition is common in the case of a weak vibra- tional resonant contribution ( µbaµcbµcdµad << 1), as in the case of low concentrations of oscillators ( µea4 nCARS nNVRB << 1). Hence, this produces an intense NVRB signal and reduces the vibrational contrast, hindering the imaging of electronically non-resonant samples. This is the case for cells and tissues which need to be excited in the near infrared to avoid radiation damage[52]. For SLG, the linear dispersion of the massless Dirac Fermions makes the response always electronically res- onant. In the case of FLG, absorption has a com- plex dependence on wavelength, as well as on the num- ber of layers and their relative orientation, exhibiting, for instance, a tunable band gap in twisted bilayer graphene[53]. This is also reflected in the resonant na- ture of SR[54, 55]. However, approaching visible wave- lengths, the absorption spectrum flattens above∼ 0.8eV and it is ∼ (1 − πe2/2h)N for Bernal-stacked N -layer graphene[56]. Our exfoliated FLG are Bernal stacked, as also confirmed by the measured SR 2D peak shape in SR[9, 10]. Accordingly, at the typical CARS wavelengths used here (784 and 894nm), SLG and Bernal FLG are electronically resonant, unlike the situation for most bio- logical samples[52]. This results in an opposite sign in the CARS response, i.e. a spectral dip in (cid:61)(χ(3) CARS), related to two additional imaginary unit contributions in the de- nominator of Eq.5, (ωP − ¯ωba) and (ω− ¯ωda), wherein the iγba, iγda components dominate. Further, the −i contri- bution from (2ωP − ¯ωea) results in a NVRB dominated by the imaginary part, as illustrated in Fig.4c,d,e. matter interactions (NVRB and CARS) rather than from a matter-only Hamiltonian coupling the electronic con- tinuum and a discrete phonon state (implying a reso- nance between the corresponding energies), resulting in the Fano resonance[57] suggested in Ref.[32]. 6 Thus, the third term in Eq.7 must be replaced with the contribution from the interference of the spectral dip CARS) with the imaginary part (cid:61)(χ(3) (cid:61)(χ(3) N V RB). This leads to a signal that, under the electronically resonant regime, becomes[15]: I(ωas)R ∼ P (3) ∝ χ(3) CARS2 + 2(cid:61)(P (3) CARS2 + 2(cid:61)(χ(3) N V RB2 + P (3) N V RB2 + χ(3) N V RB)(cid:61)(P (3) N V RB)(cid:61)(χ(3) CARS) CARS) (8) CARS/χ(3) which indicates that the total FWM, at the phonon fre- quency, can be either a negative dip or a positive peak de- pending on the ratio between the vibrationally resonant and the non-resonant susceptibilities (χ(3) N V RB), which depends only on the material under examination and not on the pulses used in the experiment. Such a qualitatively different interplay between NVRB and CARS, compared with the experimental lineshapes for ∆T = 0 in Fig.3, unambiguously indicates the presence of electronic resonance in SLG and Bernal FLG. For a given material, the relative weight of the two FWM con- tributions can be in general modified by using pulsed excitation and tuning the temporal overlap between the PP and SP fields[44], i.e. changing ∆T . The experimen- tally observed evolution of the FWM signal in FLG as a function of PP-SP delay in Fig.3 has a trend similar to that shown in Fig.4e,f as function of ηN RV B/ηCARS, validating the resonance-dominated scenario. A more quantitative picture can be derived from Eqs.3,4, where the PP and SP temporal profiles are taken into account, matching those retrieved from the experi- mentally measured autocorrelation (Fig.2). As model parameters for the FLG we use the ex- perimental SR value νca=1580cm−1, with fitted τca = 1.1±0.1ps[9, 41] (corresponding to FWHM(G)= 10cm−1) and τda = τba = τea = 10 ± 2fs in agreement with the value measured for SLG[43]. The ratio between NVRB = (3.0 ± 0.7) × 10−5 and CARS contributions ηCARS ηN V RB is obtained by fitting to the experimental data in Fig.3 with Eqs.1,19,20. The resulting spectra (colored dashed lines in Fig.3), evaluated by tuning only the PP-SP de- lays, are in good agreement with the experimental data, with ηCARS as the only adjustable parameter. This ratio, ηN V RB combined with Eqs.5,6, allows us to extract the ratio be- tween the third-order nonlinear susceptibilities for CARS and NVRB: χ(3) CARS N V RB ∼ 1.3 at the G-phonon resonance. χ(3) The dependence of our spectral profiles on the inter- pulse delay, ∆T , indicates that the peculiar FWM line- shapes for SLG and FLG originates from the inter- ference between two electronically resonant radiation- In the electronically non-resonant case, CARS provides access to the real part of χ(3)[18]. However, due to the dispersive nature of the χ(3) real part[18], it distorts the phonon lineshapes[3], unlike SR. In SLG and FLG the FWM signal arises from the imaginary (non dispersive) CARS susceptibility, and is amplified by its NVRB (third term in Eq.8). Thus, the signal can be used for vibra- tional imaging, unlike the non-resonant case[18]. The vibrationally resonant contribution I can be iso- lated by subtracting from the I2 FWM signal at ν2−νP ∼ νG, the NRVB obtained by linear interpolation of the spectral intensities measured at the two frequencies at the opposite sides of vibrational resonance: (I3 − I1) I = I1 − I2 + (9) ν2 − ν1 ν3 − ν1 where the indexes i=1,2,3 refer to data at ν1 = νP + 1545cm−1, ν2 = νP + 1576cm−1, ν3 = νP + 1607cm−1 (i.e. with ν2 near to the G phonon frequency and ν1,3 − νG greater than two half-widths at half maximum of the measured profiles, as shown in Fig.3). This combination of electronically resonant NVRB and CARS non-linear responses gives CARS images (i.e. retaining vibrational sensitivity) with signal intensities comparable to those of NVRB, for which sub-ms pixel dwell times have been demonstrated with the use of a point detector, e.g. photomultiplier[2]. In our case, the images in Fig.5 are obtained with a pixel dwell time∼ 200ms using a Si-Charge-Coupled Device (CCD) array, aiming at a complete spectral characterization, and scanning the sample at fixed ∆T with stepper-motor translation stages. Figs.5(a -- c) display nonlinear optical images measured at two different ωS, corresponding to vibrationally non- resonant and resonant conditions. Extracting for each image pixel I1 (Fig.5a), I2 (Fig.5c), and I3, required for Eq.9, we obtain an image with suppression of the signal not generated by FLG, as in Fig.5e. To obtain a quantitative comparison of the differ- intensity histogram in ent images, we plot the pixel Figs.5b,d,f. This gives a bimodal distribution: one peak corresponds to the most intense pixels, associated with FLG (Ig) and the other is related to the weaker substrate signal (Is). The ability to discriminate sample from sub- strate can be quantified in terms of 1) Ig compared to Is, evaluated as the difference Ig − Is, and 2) the proxim- ity of Is to I = 0 in the histogram (dashed black line in Figs.5b,d,f). These two features can be quantified by the contrast C in order to compare the images: C = I g−I s , where I g and I s are the mean FLG and substrate intensi- ties, corresponding to the local maxima in the histograms I s 7 Figure 5. Nonlinear optical images of FLG measured under conditions of (a) a non-vibrationally resonant λS at 891.5nm and (c) a resonant λS at 894 nm and ∆T =1.7 ps. e) CARS image of two FLG flakes, obtained by the spectral dip (see Eq.9). b,d,f) Intensity histograms of a,c,e). The corresponding contrast C is also reported. The black dashed lines represent the colormap boundaries of a,c,e). g,h) Intensity profiles along the scanning paths in and out of a FLG flake as highlighted in a,c,e by dashed and full lines, respectively. in Figs.5b,d,f. In Figs.5g,h we plot the intensity profiles along two scanning paths, one inside (dashed) and the other adjacent to (full line) the FLG flake. Comparing the three histograms (Figs.5b,d,f), the vibrationally off resonant FWM image (NVRB only, Fig.5b) has the highest I g. The visibility of the flakes is limited by the noise of the detector and by χ(3) non- linearity of the substrate. NVRB, lacking vibrational specificity, can also originate from the glass substrate outside the FLG flake (I s (cid:29) 0), as indicated by the scanning profile in Fig.5h (red line). This may become a critical limitation in those substrates with χ(3) much larger than Si (χ(3) ∼ 2.5 × 10−10 e.s.u.[29]), such as Au (χ(3) = 4 × 10−9e.s.u[58, 59]). Similarly, the vibra- tionally resonant FWM, I2, originating from concurrent CARS and NVRB processes (Fig.5d), has a I s (cid:29) 0 re- lated to NRVB. The depth of the FWM dip (Fig.5f) is related to the CARS signal intensity, and its vibrational sensitivity brings about a substantial contrast increase, as demonstrated by the close-to-zero average value of the (green) scanning profile in Fig.5h. In summary, by using an experimental time-delayed FWM scheme, CARS peaks equivalent to those seen in spontaneous Raman were obtained from graphene. By explaining the physical mechanism responsible for the FWM signal, we demonstrated that the spectral response can be described in terms of joint CARS and NVRB contributions concurring to the overall signal. Unlike non-resonant FWM, where dispersive lineshapes hamper vibrational imaging of biological systems, the resonant nature of FWM in graphene, which can be traced back to its peculiar electronic properties, mixes CARS and NVRB, resulting in Lorentzian profiles which are either peaks or dips depending on their relative strength. We also demonstrated that CARS can be used for vibrational imaging with contrast equivalent to spontaneous Raman microscopy and signal levels as large as those of the third order nonlinear response. 8 ERC grant Hetero2D, EPSRC grants EP/L016087/1, EP/K01711X/1, EP/K017144/1. This project has re- ceived funding from the European Unions Horizon 2020 research and innovation programme under grant agree- ment No. 785219 - GrapheneCore2. METHODS The third -order the SLG and FLG samples can be obtained from the third-order polarization[15]: response for (cid:90) ∞ (cid:90) ∞ (cid:90) ∞ P (3)(t) ∝ N E(t − τ2 − τ3)E(t − τ1 − τ2 − τ3)S(3)(τ1, τ2, τ3) dτ1E(t − τ3) dτ3 dτ2 0 0 0 (10) (cid:110) where N is the number of scatterers, S(3)(τ1, τ2, τ3) may be expressed as[15]: µ(τ1 + τ2 + τ3) S(3)(τ1, τ2, τ3) ∝ (i)3 T r (cid:2)µ(τ1 + τ2),(cid:2)µ(τ1),(cid:2)µ(0), ρ(−∞)(cid:3)(cid:3)(cid:3)(cid:111) (cid:2)Ei(t, ∆ti)+c.c.(cid:3) = and E(t) is the total electric field on the sample E(t) = (cid:88) (cid:88) (cid:2) Ai(t, ∆it)e−iωit+c.c.(cid:3) (11) ACKNOWLEDGEMENTS We thank M. Polini for useful discussions. We ac- knowledge funding from the EU Graphene Flagship, i=P,S i=P,S (12) Consider a SP at ∆tS = 0 with ∆t = ∆tP . The energy level diagrams in Fig.1schematically illustrate the CARS and NVRB processes[15]: CARS(t) ∝ (i)3 nCARSµbaµcbµcdµad P (3) (cid:90) ∞ N V RB(t) ∝ (i)3 nNVRBµea4 P (3) (cid:90) ∞ (cid:90) ∞ dτ3 dτ2 (cid:90) ∞ 0 0 0 (cid:90) ∞ (cid:90) ∞ dτ1AP (t − τ1 − τ2 − τ3, ∆t)A∗ S(t − τ2 − τ3)AP (t − τ3, ∆t) e−iωP (t−τ1−τ2−τ3,∆t)e+iωS (t−τ2−τ3)e−iωP (t−τ3,∆t)e−i¯ωbaτ1e−i¯ωcaτ2e−i¯ωdaτ3 dτ2 dτ1AP (t − τ1 − τ2 − τ3)AP (t − τ2 − τ3)A∗ dτ3 e−iωP (t−τ1−τ2−τ3,∆t)e−iωP (t−τ2−τ3,∆t)e+iωS (t−τ3)e−i¯ωeaτ1 e−i¯ωeaτ2 e−i¯ωeaτ3 S(t − τ3) 0 0 0 (13) (14) (cid:90) ∞ dτ2 By Fourier transform, the frequency dispersed signal where ¯ωij = ωi − ωj − iγij. can be expressed as: P (3)(ω) = (cid:90) ∞ −∞ P (3)(t)eiωtdt. (15) In order to reduce the computational effort to calculate Eqs.13,14, we also write the pulse fields in terms of their (cid:90) ∞ (cid:90) ∞ Fourier transforms, obtaining: (cid:90) ∞ (cid:90) ∞ (cid:90) ∞ CARS(ω) ∝ ηCARS (i)3 P (3) (cid:90) ∞ dt eiωt dτ3 −∞ dω3 AP (ω1, ∆t) A∗ 0 0 0 dω1 −∞ S(ω2, 0) dτ1 AP (ω3, ∆t)e−i(ωP +ω1)(t−τ1−τ2−τ3)e+i(ωS +ω2)(t−τ2−τ3) −∞ −∞ dω2 e−i(ωP +ω3)(t−τ3)e−i¯ωbaτ1e−i¯ωcaτ2 e−i¯ωdaτ3 (16) where ηCARS = nCARSµbaµcbµcdµad. In this way all the temporal integrals can be solved analytically: CARS(ω) ∝ −ηCARS P (3) (cid:90) ∞ −∞ dω1 (cid:90) ∞ −∞ dω2 (cid:90) ∞ −∞ dω3δ(ω − 2ωP + ωS + ω1 − ω2 + ω3) (ωP + ω1 − ¯ωba)(ωP − ωS + ω1 − ω2 − ¯ωca)(2ωP − ωS + ω1 − ω2 + ω3 − ¯ωda) AP (ω1, ∆t) A∗ S(ω2, 0) AP (ω3, ∆t) 9 (17) using the energy conservation, represented by the delta distribution: δ(ω−2ωP +ωS−ω1+ω2−ω3) = ei(ω−2ωP +ωS−ω1+ω2−ω3)t, (cid:90) ∞ the ω2 integral can be simplified: CARS(ω, ∆t) ∝ −ηCARS P (3) AP (ω3, ∆t) AP (ω1, ∆t) A∗ −∞ (18) (cid:90) ∞ (cid:90) ∞ S(2ωP − ωS − ω + ω3 + ω1, 0) −∞ −∞ dω1 dω3 (cid:90) ∞ (cid:90) ∞ dω1 −∞ dω2 −∞ (ωP + ω3 − ¯ωba) (ω − ωP − ω1 − ¯ωca) (ω − ¯ωda) (19) In a similar way, using ηNVRB = nNVRBµea4, Eq.14 can be written as: N V RB(ω, ∆t) ∝ −ηNVRB P (3) AP (ω1, ∆t) AP (ω2, ∆t) A∗ S(2ωP − ωS − ω + ω1 + ω2, 0) (ωP + ω1 − ¯ωea) (2ωP + ω1 + ω2 − ¯ωea) (ω − ¯ωea) (20) [1] G. Soavi et al. Nat. Nanotechnol., 15, 1748 (2018) [2] E. Hendry et al. Phys. Rev. Lett. 105, 097401 (2010) [3] R. W. Boyd, Nonlinear Optics, (Academic Press, 2008) [4] E. O. Potma and S. Mukamel, in Coherent Raman Scat- tering Microscopy, edited by J. Cheng and X. S. Xie (CRC Press, 2012) [5] J.B. Khurgin, Appl. Phys. Lett. 104, 161116 (2014) [6] R. Wu et al. Nano Lett. 11, 5159 (2011) [7] T. Gu et al. Nat. Photonics 6, 554 (2012) [8] D.L. Duong et al. Nature 490, 235 (2012) [9] A.C. Ferrari et al. Phys. Rev. Lett. 97, 187401 (2006) [10] A.C. Ferrari, D.M. Basko, Nat. Nanotechnol. 8, 235 (2013). [11] R.F. Begley et al. Appl. Phys. Lett. 25, (1974). [12] M.D. Duncan et al. Opt. Lett. 7, 350 (1982) [13] A. Zumbusch et al. Phys. Rev. Lett. 82, 4142 (1999) [14] J.X. Cheng et al. J. Opt. Soc. Am. B 19, 1363 (2002) [15] S. Mukamel, Principles of Nonlinear Optical Spec- troscopy, (Oxford University Press, New York, 1999) [16] M. Cui et al. Opt. Lett. 34, 773 (2009) [17] D. Pestov, et al. Science 316, 265 (2007) [18] C.L. Evans, X.S. Xie, Annu. Rev. Anal. Chem. 1, 883 (2008) [19] C. L. Evans et al. Proc. Natl. Acad. Sci. U.S.A 102, 16807 (2005) [20] C. H. Camp Jr and M. T. Cicerone, Nat. Photonics, 5, 295 (2015) [21] Y. J. Lee et al. ACS Macro Lett. 1 (11), 1347 (2012) [22] S. H. Lim et al. J. Phys. Chem. B 110 (11), 5196 (2006) [23] R.S. Lim, et al. J. Lipid Res. 52 2177 (2011) [24] C. Steuwe et al. Nano Lett. 11, 5339 (2011) [25] K. Ikeda, K. Uosaki Nano Lett. 9, 1378 (2009) [26] Y.J. Lee et al. Phys. Rev. B 82, 165432 (2010) [27] G. Dovbeshko et al. Nanoscale Res. Lett. 9, 263 (2014) [28] T. Baldacchini, R. Zadoyan, Opt. Express 18, 19219 (2010) [29] J. Brocious, E.O. Potma, Mater. Today 16, 344 (2013) [30] C. Ferrante et al. Nat. Commun., 9, 308 (2018) [31] J. Koivistoinen et al. J. Phys. Chem. Lett., 8, 4108 (2017) [32] L. Lafeta et al. Nano Lett., 17 (6), pp 34473451 (2017) [33] X. S. Li, W. W. Cai, J. H. An, S. Kim, J. Nah, D. X. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, S. K. Banerjee, L. Colombo, R. S. Ruoff, Science 324, 1312 (2009). [34] S. Bae, H. Kim, Y. Lee, X. Xu, J. S. Park, Y. Zheng, J. Balakrishnan, T. Lei, H. R. Kim, Y. I. Song, Y. J. Kim, K. S. Kim, B. Ozyilmaz, J. H. Ahn, B. H. Hong, S. Iijima, Nat. Nanotechnol. 5, 574 (2010). [35] F. Bonaccorso, A. Lombardo, T. Hasan, Z. P. Sun, L. Colombo, A. C. Ferrari, Mater. Today 15, 564 (2012). [36] D. M. Basko, S. Piscanec, A. C. Ferrari, Phys. Rev. B 80, (2009). [37] A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. K. Saha, U. V. Waghmare, K. S. Novoselov, H. R. Krish- namurthy, A. K. Geim, A. C. Ferrari, A. K. Sood, Nat. Nanotechnol. 3, 210 (2008). [38] K. S. Novoselov et al. Proc. Natl. Acad. Sci. USA 102 (30), 10451 (2008) [39] G. Krauss et al. Opt. Lett., 34, 2847 (2009) [40] M. Marangoni et al. Opt. Lett., 34, 3262 (2009) [41] N. Bonini et al. Phys. Rev. Lett. 99, 176802 (2007) [42] A. Tomadin, et al. Phys. Rev. B 88, 035430 (2013) [43] D. Brida et al. Nat. Commun. 4, 1987 (2013) [44] A. Volkmer et al. Appl. Phys. Lett. 80, (2002) [45] F.M. Kamga, M.G. Sceats, Opt. Lett. 5, 126 (1980) [46] D.A. Sidorov-Biryukov, E.E.Serebryannikov, A.M. Zheltikov, Opt. Lett. 31, 2323 (2006) [47] R. Selm, et al. Opt. Lett. 35, 3282 (2010). [48] V. Kumar et al. Opt. Express 19, 15143 (2011). [49] G. Batignani et al. Sci. Rep. 6 (2016) [50] P. Wang et al. Chem. Phys. Lett. 556, 146 (2013) [51] L.G. Can¸cado, et al. Phys. Rev. B 76, 064304 (2007) [52] M. Muller, A. Zumbusch, ChemPhysChem 8, 2156 (2007) [53] Y. Zhang, et al. Nature 459, 820 (2009) [54] J. B. Wu, et al. Nat. Commun. 5, 5309 (2014) [55] J. B. Wu, et al. ACS Nano 9, 7, 7440 (2015) [56] K. F. Mak, et al. Proc. Natl. Acad. Sci. USA 107, 14999 (2010) [57] U. Fano, Phys. Rev. 124, 1866 (1961) [58] R. W. Boyd et al. Opt. Commun., 326, 74 (2014) [59] E. Xenogiannopoulou et al. Opt. Commun., 275, 217 (2007) 10 

1612.02867

1

1612

Brightening of dark excitons in monolayers of semiconducting transition metal dichalcogenides

We present low temperature magneto-photoluminescence experiments which demonstrate the brightening of dark excitons by an in-plane magnetic field $B$ applied to monolayers of different semiconducting transition metal dichalcogenides. For both WSe$_2$ and WS$_2$ monolayers, the dark exciton emission is observed at $\sim$50 meV below the bright exciton peak and displays a characteristic doublet structure which intensity is growing with $B^2$, while no magnetic field induced emission peaks appear for MoSe$_2$ monolayer. Our experiments also show that the MoS$_2$ monolayer has a dark exciton ground state with a dark-bright exciton splitting energy of $\sim$100 meV.

cond-mat.mes-hall

cond-mat

Brightening of dark excitons in monolayers of semiconducting transition metal dichalcogenides M. R. Molas,1, ∗ C. Faugeras,1 A. O. Slobodeniuk,1 K. Nogajewski,1 M. Bartos,1 D. M. Basko,2 and M. Potemski1, † 1Laboratoire National des Champs Magnétiques Intenses, CNRS-UGA-UPS-INSA-EMFL, 25, avenue des Martyrs, 38042 Grenoble, France 2Laboratoire de Physique et Modélisation des Milieux Condensés, Université de Grenoble-Alpes and CNRS, 25 rue des Martyrs, 38042 Grenoble, France (Dated: December 12, 2016) 6 1 0 2 c e D 8 ] l l a h - s e m . t a m - d n o c [ 1 v 7 6 8 2 0 . 2 1 6 1 : v i X r a We present low temperature magneto-photoluminescence experiments which demonstrate the brightening of dark excitons by an in-plane magnetic field B applied to monolayers of different semiconducting transition metal dichalcogenides. For both WSe2 and WS2 monolayers, the dark exciton emission is observed at ∼50 meV below the bright exciton peak and displays a characteris- tic doublet structure which intensity is growing with B2, while no magnetic field induced emission peaks appear for MoSe2 monolayer. Our experiments also show that the MoS2 monolayer has a dark exciton ground state with a dark-bright exciton splitting energy of ∼100 meV. I. INTRODUCTION Monolayers (MLs) of semiconducting transition metal dichalcogenides (S-TMDs) MX2 where M=Mo or W and X=S, Se or Te, are direct band gap semiconductors [1] with the minima (maxima) of conduction (valence) band located at the inequivalent K+ and K− points of their hexagonal Brillouin zone (BZ). These two-dimensional semiconductors host tightly bound excitons with uncon- ventional properties such as binding energies as large as few hundreds of meV and non Rydberg excitation spec- trum [2–4]. The lack of inversion symmetry together with the strong spin-orbit interaction lift the degeneracy be- tween spin levels in the conduction (CB) and valence (VB) bands at the K+ and K− points related by time reversal symmetry. The spin-orbit interaction leads to well separated spin subbands in each valley and to the possibility of initializing a defined valley population with circularly polarized optical excitation [5–8] or generation of valley coherence [9, 10]. The spin-orbit splitting ∆so,vb in the valence is as large as few hundreds of meV [3, 11– 22] while its counterpart in the conduction band ∆so,cb is predicted to be of the order of few tens of meV only. What is however important is that ∆so,cb can be posi- tive or negative [23, 24] and in consequence, two distinct ordering of the spin orbit split CB subbands are feasi- ble [25–27]. Because optical transitions in S-TMDs do conserve the spin, different orderings of electronic bands in the con- duction band have profound consequences on their op- tical properties. Depending on the sign of ∆so,cb, the excitonic ground state can be bright (parallel spin con- figuration for the top VB and the lowest CB subbands between which the optical transition is allowed) or dark (anti-parallel spin configuration and optically forbidden ground state interband transition). The ordering of the ∗ [email protected] † [email protected] electronic bands, characteristic for these two monolayer families, referred to as bright and darkish ones, are il- lustrated in Fig. 1(a). Theoretical studies [25–27] indeed predict that monolayers of MoSe2 and of MoTe2 should be bright (∆so,cb>0 to set a convention) while WSe2 and WS2 monolayers are darkish (∆so,cb<0). Yet, there is no general consensus concerning the bright of darkish character of a MoS2 monolayer. The theoretical works reported in Refs 26, 27 classify the MoS2 monolayer as a bright system: ∆so,cb>0 but as small as 3 meV. In- stead, another theoretical study [28] indicates that MoS2 monolayers are rather darkish (∆so,cb ∼ −40meV ). A detailed knowledge of the exciton fine structure is crucial for S-TMD based optoelectronic devices and for valleytronic applications, as i) optical properties strongly depend on the type of excitonic ground state, and ii) scat- tering mechanisms, and in particular intervalley scatter- ing mechanisms, can have much different efficiencies for bright and dark excitons [29, 30]. On the experimen- tal point of view, recent optical studies of WSe2 have shown that the temperature dependence of its PL inten- sity is consistent with a dark excitonic ground state and the dark-bright exciton splitting of about 30 meV has been experimentally estimated from temperature activa- tion type analysis [31–33]. Magnetic fields, applied in an adequate configuration with respect to a crystal axis, can mix electronic wave functions and thus the excitonic states which are built out of these wave functions. This effect triggered the spectroscopy of optically dark excitons in a large variety of condensed matter systems, ranging from bulk semi- conductors [34], semiconductor quantum dots [35, 36], to single wall carbon nanotubes [37, 38]. One expects that also in monolayers of S-TMDs, the in-plane mag- netic field acts as a perturbation to the system's Hamil- tonian, mixing the two lowest spin levels in the CB and hence, the bright excitons giving some optical activity to the initially dark excitonic states [39]. 2 FIG. 1. (a) Diagram of relevant subbands in the CB and VB at the K+ and K− points of the BZ in the bright and darkish monolayers of S-TMDs. The orange (green) curves indicate the spin-up (spin-down) subbands. The red and blue wavy lines show the A exciton transitions which are optically active. ∆so,cb denotes the spin-orbit splitting in the conduction band. (b) Schematic representation of the experimental configuration for magneto-PL measurements in Voigt configuration. In this paper, we provide a direct measurement of the dark exciton emission in darkish monolayers of S-TMDs by mixing the spin levels of bright and dark excitons by an in-plane magnetic field. Dark excitons appear in the low temperature magneto-photoluminescence (PL) spec- tra as clear features growing with the magnetic field at energies lower than that of the bright exciton. This observation gives a direct access to the values of dark- bright exciton splitting in WSe2 and WS2 monolayers. In the case of MoSe2, no significant change in the emission spectrum is observed when applying a magnetic field, in agreement with its bright exciton ground state. MoS2 is shown to belong to the family of darkish materials with a dark-bright splitting energy close to 100 meV. The emis- sion intensity of dark excitons increases as B2, in accor- dance with their perturbative activation by the in-plane magnetic field. II. THEORETICAL BACKGROUND To examine the band edge interband transitions in S- TMD monolayers we consider the top VB and two spin- orbit split CB subbands [see Fig. 1(a)]. Associated with these subbands and relevant for our considerations are intravalley interband transitions (intravalley A excitons) which involve the states from the the same K+ or K− valley. Four types of intravalley A excitons can be dis- tinguished and labelled according to their valley τ = ± and CB spin scb =↑,↓ indices (the VB spin index, svb is fixed to svb =↑ in K+ valley and svb =↓ in K− valley). The configurations with svb = scb =↑ from the K+ val- ley and svb = scb =↓ from the K− valley correspond to optically active, bright A excitons, referred to as τ, b(cid:105). The configurations with svb =↑, scb =↓ from the K+ val- ley and svb =↓, scb =↑ from the K− valley correspond to optically inactive, dark A excitons, refereed to as τ, d(cid:105). The ground state of bright (darkish) S-TMD monolay- ers is then formed from bright (dark) A excitons. Because these quasi-particles differ from each other by their spin configuration in the CB, spin-flip processes in the CB can make the dark states optically active and can allow for investigations of the ground state of darkish materials. A magnetic field B = (Bx, By), applied along the plane of a S-TMD monolayer, mixes the spin states in the CB and VB via the Zeeman interaction. Since ∆so,vb (cid:29) ∆so,cb, the spin-mixing in the VB can be ne- glected. The Zeeman term acting on the CB states can be expressed as: HZ = 1 2 gcbµB(σxBx + σyBy) (1) Here gcb is the in-plane gyromagnetic ratio for the CB, µB is the Bohr magneton and σx,y are the Pauli matrices in the CB spin subspace. The in-plane magnetic field results in the mixing of the dark and bright excitons. It can be described by an effective 2× 2 Hamiltonian in the basis of {τ, b(cid:105), τ, d(cid:105)}, obtained by the projection of spin states of the CB, mixed by the magnetic field, on the exciton states Ed H τ ex = 1 2 gcbµBBτ (2) Here we introduced B± = Bx±iBy. Eb and Ed are the energies of the bright and dark excitons in the absence of an external magnetic field, Ed − Eb = ∆so,cb. The application of the in-plane magnetic field does not lift the double degeneracy of each dark and bright exciton states as H τ ex does not depend on valley index τ. Assuming that the Zeeman term gives a small correc- tion to the basic exciton states, we obtain the mixed eigenstates up to second order in magnetic field: τ, b(cid:105)mix = τ, b(cid:105) 1 + w/2 − gcbµBBτ 2∆so,cb τ, d(cid:105), (3) (cid:20) Eb 1 2 gcbµBB−τ (cid:21) . τ, d(cid:105)mix = τ, d(cid:105) 1 + w/2 + gcbµBB−τ 2∆so,cb τ, b(cid:105). (4) III. EXPERIMENTAL RESULTS AND DISCUSSION 3 Here w = g2 so,cb) (cid:28) 1. Their eigenener- gies are very close to the energies of the dark and bright excitons (the correction is ∝ w∆so,cb). cbµ2 BB2/(4∆2 The admixture of bright states to the dark exciton state makes the latter resonance to be possibly observed in the PL spectra when the in-plane magnetic field is ap- plied to the layer. The intensity Id of such a PL line can be expected to be proportional to the fraction w of bright exciton in the corresponding mixed state and to the population nd of dark excitons: Id = ndIbw ∝ ndIbB2, (5) where Ib is the intensity of the pure bright exciton state emission in the absence of the magnetic field. With available magnetic fields, the factor w remains rather small and dark excitons can hardly be observed in ab- sorption experiments. We note two different situations. i) For bright materials, such as MoSe2 or MoTe2, the energy of dark excitons is larger than the energy of the bright ones. Therefore, at low temperatures, the popula- tion of dark excitons is suppressed by a Boltzmann factor exp(−∆so,cb/kBT ) and optical transitions are mainly due to low-lying bright exciton states. In this case the obser- vation of dark excitons at low temperature is extremely unlikely. ii) For darkish materials, such as WSe2 or WS2, the situation is opposite and the direct observation of dark excitons is possible. So far we have considered the A excitons formed by the direct electron-hole Coulomb interaction and have not included effects of the exchange part of the Coulomb interaction. The exchange interaction is expected to lift the double valley degeneracy of dark intravalley A exci- tons due to the presence of a transition dipole moment perpendicular to the monolayer plane, absent for bright excitons [39, 40]. This degeneracy lifting can be viewed as a local-field effect due to the out-of-plane transition dipole moment of spin-forbidden dark excitons. It is analogous to the exchange energy shift of the Z-excitons in semiconductor quantum wells [41, 42]. The result- ing energy splitting between the two spin-forbidden dark exciton components in S-TMD monolayers was roughly estimated in Ref. 39 to be about 10 meV; a more precise, microscopic calculation of this splitting is still lacking, to the best of our knowledge. The discussed above effects of the in-plane magnetic field are equally valid for each component of the expected doublet structure of dark ex- citons in S-TMD monolayers. Monolayers of S-TMDs have been prepared by me- chanical exfoliation of bulk crystals purchased from HQ Graphene. Initially, the flakes were exfoliated onto a polydimethylsiloxane (PDMS) stamp attached to a glass plate. MLs of S-TMDs were then identified by their opti- cal contrast and cross-checked by Raman scattering and PL measurements at room temperature. In order to de- posit them on target Si/SiO2(320 nm) substrates, an all- dry PDMS-based transfer method similar to the one de- scribed in Ref. 43 was employed. Low temperature magneto-PL experiments were per- formed in the Voigt configuration [see Fig. 1(b)] using an optical-fiber-based insert placed in a superconduct- ing magnet producing magnetic fields up to 14 T. The samples were placed on top of a x-y-z piezo-stage kept in gaseous helium at T = 4.2 K. The light from a semicon- ductor diode laser (λ=515 nm) was coupled to an opti- cal fiber with a core of 50 µm diameter and focused on the sample by an aspheric lens (spot diameter around 10 µm). PL signals were collected by the same lens, in- jected into a second optical fiber of the same diameter, and analyzed by a 0.5 m long monochromator equipped with a charge-couple-device (CCD) camera. To investigate the effect of an in-plane magnetic field on the PL signal of S-TMD monolayers, we measured the evolution of the low temperature (T = 4.2 K) PL spectra of the WSe2, WS2, MoSe2, and MoS2 MLs in the Voigt configuration as a function of an external magnetic field up to B = 14 T. The obtained spectra at B = 0 and at B = 14 T are presented in the upper panels of Fig. 2. The zero-field PL spectra of all our monolayers display two characteristic emission features, labelled A and T, which are associated with recombination of the neutral [an electron-hole (eh) pair] and charged [an eh pair + an extra carrier (electron or hole)] excitons formed at the K± points of the BZ [9, 30, 31, 44–50]. In the case of WSe2, WS2, and also of MoS2, additional features are apparent in the PL spectra in the form of a series of emission lines (WSe2 and WS2) or a broad band (MoS2), at energies below the A exciton energy and overlapping with the T peak. These additional lines have been attributed in the literature to the so-called localized/bound or defect- related excitons [9, 30, 31, 44–46]. We start with the analysis of the results obtained for the tungsten-based family, i.e. WSe2 and WS2 MLs, as both of them are rather firmly predicted to belong to the family of darkish monolayers [25–27]. The zero-field PL spectra, apart the A and T peaks, consist of several overlapping emission lines on the lower energy side of the spectrum [upper panels of Fig. 2(a) and (b)]. We show in Fig. 2(a) and (b) that the application of a magnetic field in the plane of these monolayers strongly affects their PL spectra at energies 50− 60 meV below the A exciton line. To better visualize the effects of magnetic fields and com- pare the results obtained for different materials, we define 4 FIG. 2. (upper panels) PL spectra of (a) WSe2, (b) WS2, (c) MoSe2, and (d) MoS2 monolayers at T = 4.2 K measured at zero field (red curves) and at B = 14 T (blue curves) applied in the plane of the crystal. The PL spectra were normalized to the intensity of the A exciton line. (lower panels) Corresponding relative intensities of the monolayers defined as (PLB=14 T - PLB=0 T)/PLB=0 T are represented by black dots. The orange and green curves indicate Gaussians fits of the data. a relative spectrum as (PLB(cid:54)=0 - PLB=0)/PLB=0. Such relative intensity spectra for B = 14 T are presented in the lower panels of Fig. 2(a) and (b). For WSe2 and WS2 MLs, these spectra are composed of two peaks, labelled D1 and D2, which appear on the lower energy side of the bright A exciton. In agreement with our theoretical arguments, these two peaks are assigned to the magnetic- field induced emission due to dark excitons. The higher energy peak, D1, emerges about 47 meV below the A ex- citon line for both members of the tungsten-based family. The energy separation between the D1 and D2 peaks is 14 meV for WSe2 monolayer and 23 meV for the WS2 monolayer. To analyze further the data, we fitted the D1 and D2 features using two Gaussian functions [see lower panels of Fig. 2 (a) and (b)]. In the whole range of investi- gated magnetic fields, the energy and the full width at half maximum (FWHM) of the two D1 and D2 peaks are constant. The brightening of these dark excitons is evidenced by the quadratic evolution of the integrated in- 1.601.651.701.750.00.51.01.52.02.5Relative int. (arb. u.)14 meV47 meV Energy (eV)D1D2A0481216(a)PL (arb. u.) B= 0 T 14 TTWSe21.952.002.052.100.00.51.01.52.02.5Relative int. (arb. u.) Energy (eV)T0481216202428(b)PL (arb. u.)WS2 B= 0 T 14 TA23 meV47 meVD1D21.551.601.651.700.00.51.01.52.02.5020406080Relative int. (arb. u.) Energy (eV)(c)PL (arb. u.)MoSe2x5 B= 0 T 14 TTA012345671.651.701.751.801.851.901.952.000.00.51.01.52.02.5 Relative int. (arb. u.)Energy (eV)98 meVD(d)PL (arb. u.)MoS2 B= 0 T 14 TAtensity of these peaks as a function of the magnetic field (∼ αB2, where α is a fitting parameter). This behavior is presented in Fig. 3 and is in agreement with the ar- guments presented in the preceding section (Eq. 5). Im- portant here is the observed B2 dependence and not the precise rates of increase of the two D lines, which appar- ent values are affected by the chosen normalization of the relative spectra. We consider that the energy difference between the bright A exciton peak and the dark D1 exci- ton peaks corresponds well to the theoretical predictions of the ∆so,cb magnitude [25–27]. Note that the values for ∆so,cb calculated in Ref. 25–27 do not include the electron-hole Coulomb effects, which obviously affect the interband transition energies but can also significantly influence the apparent bright-dark exciton splitting due to electron-hole exchange effects. An MoSe2 monolayer is predicted to belong to the fam- ily of bright S-TMDs. Its zero-field PL spectra is rather simple (and similar to that observed for MoTe2 monolay- ers [51]). It is composed of only two A and T features [see Fig. 2(c)][47–49]. When a magnetic field is applied in the direction along the plane of the layer, no significant changes of the PL spectra are observed. In particular, there are no additional growing structures on the high energy side of the A exciton line, where the dark exciton emission could be expected according to the band order- ing at the K± points [see Fig. 1(a)]. The dark exciton emission can not be detected with our experimental con- ditions as a result of the fast relaxation of carriers to the lowest energy state which is a bright exciton. The only field induced effect observed in the magneto-PL spectra of the MoSe2 monolayer is a small decrease in the in- tensity of the T-peak [see lower panel of Fig. 2(c)]. The origin of this field induced suppression of the trion emis- sion is not clear for us and calls for a possible theoretical explanation, thought one may speculate that it reflects an influence of the magnetic field on the formation of the charged excitons in a MoSe2 monolayer through a mixing of the spin split bands in both valleys. Existing models describing the band ordering for MoS2 monolayer largely predict a positive, thought small, ∼ 3 meV value for ∆so,cb [25–27], thus placing MoS2 in the family of bright materials. Recently, however, a nega- tive dark-bright exciton splitting has been predicted [28] with a value close to −40 meV. By comparing the PL spectra measured at zero magnetic field for the different materials presented in this study (Fig. 2), the low tem- perature PL spectrum of MoS2 resembles more the one observed for WS2 and WSe2 than the one of MoSe2 or of MoTe2 [51]. In similarity to the low temperature PL spectra of darkish monolayers, the spectrum of MoS2 also displays a significantly broad emission band at energies lower than that of the A exciton [see Fig. 2(d)]. The ob- servation of either a well defined two peaks PL spectrum arising from the A and T excitons or an additional broad band associated with localized/bound excitons [6, 50], 5 FIG. 3. Magnetic-field dependence of the intensities of dark exciton lines (D1, D2, and D) obtained on WSe2, WS2, and MoS2 monolayers. appears to be characteristic of the two families of bright or darkish S-TMD monolayers. The presence of emis- sion due to localized/bound excitons in the low temper- ature B = 0 PL of darkish monolayers and not in the bright ones could be due to the appearance of long-lived reservoir of dark excitons in the former systems, which then effectively diffuse and/or relax towards other possi- ble radiative centers. Similar relaxation processes can be largely suppressed in bright monolayers, as the ground state excitons in these systems already represent the ef- fective recombination channel. Following this logic, the ground state exciton should be dark in the MoS2 mono- layer. The darkish character of the MoS2 monolayer is con- firmed by our magneto-PL study as indeed the in-plane magnetic field has a dramatic effect on the PL spectrum of this monolayer. The relative intensity spectrum dis- played in the lower panel of Fig. 2(d) shows a rather single but broad peak, labelled D, which is centered at about 97 meV below the A exciton line of MoS2. Even though all observed PL peaks are much broader in our MoS2 monolayer than in other studied materials, we have performed the same analysis as for the other materials. Similarly to the case of WS2 and WSe2 monolayers, the 02468101214015304560750246810048121620MoS2 Magnetic field (T)DWS2D370x10-6/T2D239x10-6/T2D154x10-6/T2 Dark excitons intensity x10-3 (arb. units)D2D1WSe2D181x10-6/T2D2 D1D296x10-6/T2shape (width and center position) of the relative spec- trum of the MoS2 monolayer remain field independent but its amplitude increases quadratically, ∼ αB2, with the magnetic field [see Fig. 3]. This result confirm placing the MoS2 in the family of darkish S-TMD with a dark- bright exciton splitting twice bigger than that found in WS2 or in WSe2. IV. CONCLUSIONS To conclude, we have presented the experimental in- vestigations supported by the theoretical consideration of the effect of brightening of dark excitons in S-TMD monolayers induced by the application of a magnetic field in the direction along the plane of the layer. Field in- duced emission due to dark excitons can be observed at low temperatures in S-TMD monolayers for which the dark excitons are lower in energy than the bright exci- tons. Emission intensities of dark excitons grow quadrat- ically with the strength of the in-plane magnetic field. These results lead us to establish the WS2, WSe2 and MoS2 monolayer as darkish materials, i.e., the direct bandgap systems but with a dark excitonic ground state, 6 and monolayers of MoSe2 as a bright materials with a bright exciton ground state. The bright-dark exciton splitting is found to be of about 50 meV in WS2 and WSe2 monolayers in fair agreement with theoretical ex- pectations [25–27], but its value derived for the MoS2 monolayer is surprisingly large [28]. The characteristic doublet structure of dark excitons has been observed for WS2, WSe2 monolayers, along the lines of the recent the- oretical proposal [39]. Different ordering of the spin-orbit split subbands in the conduction band for two, bright and darkish TMD families, is also speculated to be reflected in the B = 0 low temperature PL spectra: bright mono- layers show a simple emission due to exciton and trions, the darkish ones display an additional broad/multipeak emission band due to localised/bound excitons. V. ACKNOWLEDGEMENTS The work has been supported by the European Re- search Council (MOMB project no. 320590), the EC Graphene Flagship project (no. 604391), the National Science Center (grant no. DEC-2013/10/M/ST3/00791) and the Nanofab facility of the Institut Néel, CNRS UGA. 1 K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). 2 M. M. Ugeda, A. J. Bradley, S.-F. Shi, F. H. da Jornada, Y. Zhang, D. Y. Qiu, W. Ruan, S.-K. Mo, Z. Hussain, Z.- X. Shen, F. Wang, S. G. Louie, and M. F. Crommie, Nat. Mater. 13, 1091 (2014). 3 Z. Ye, T. Cao, K. O'Brien, H. Zhu, X. Yin, Y. Wang, S. G. Louie, and X. Zhang, Nature 513, 214 (2014). 4 A. Chernikov, A. M. van der Zande, H. M. Hill, A. F. Rigosi, A. Velauthapillai, J. Hone, and T. F. Heinz, Phys. Rev. Lett. 115, 126802 (2015). 5 D. Xiao, G.-B. Liu, W. Feng, X. Xu, and W. Yao, Phys. Rev. Lett. 108, 196802 (2012). 6 K. F. Mak, K. He, J. Shan, and T. F. Heinz, Nat. Nan- otechnol. 7, 494 (2012). 7 T. Cao, G. Wang, W. Han, H. Ye, C. Zhu, J. Shi, Q. Niu, P. Tan, E. Wang, B. Liu, and J. Feng, Nat. Commun. 3, 887 (2012). 8 H. Zeng, J. Dai, W. Yao, D. Xiao, and X. Cui, Nat. Nan- otechnol. 7, 490 (2012). 9 A. M. Jones, H. Yu, N. J. Ghimire, S. Wu, G. Aivazian, J. S. Ross, B. Zhao, J. Yan, D. G. Mandrus, D. Xiao, W. Yao, and X. Xu, Nat. Nanotechnol. 8, 634 (2013). 10 G. Wang, X. Marie, B. L. Liu, T. Amand, C. Robert, F. Cadiz, P. Renucci, and B. Urbaszek, Phys. Rev. Lett. 117, 187401 (2016). 11 B. W. H. Baugher, H. O. H. Churchill, Y. Yang, P. Jarillo-Herrero, Nat. Nanotechnol. 9, 262 (2014). 12 Y. Zhang, T.-R. Chang, B. Zhou, Y.-T. Cui, H. Yan, Z. Liu, F. Schmitt, J. Lee, R. Moore, Y. Chen, H. Lin, H.-T. Jeng, S.-K. Mo, Z. Hussain, A. Bansil, and Z.-X. Shen, Nat. Nanotechnol. 9, 111 (2014). and 13 J. M. Riley, F. Mazzola, M. Dendzik, M. Michiardi, T. Takayama, L. Bawden, C. Granerød, M. Leandersson, T. Balasubramanian, M. Hoesch, T. K. Kim, H. Takagi, W. Meevasana, P. Hofmann, M. S. Bahramy, J. W. Wells, and P. D. C. King, Nat. Phys. 10, 835 (2014). 14 J. S. Ross, S. Wu, H. Yu, N. J. Ghimire, A. M. Jones, G. Aivazian, J. Yan, D. G. Mandrus, D. Xiao, W. Yao, and X. Xu, Nat. Commun. 4, 1474 (2013). 15 A. Chernikov, T. C. Berkelbach, H. M. Hill, A. Rigosi, Y. Li, O. B. Aslan, D. R. Reichman, M. S. Hybertsen, and T. F. Heinz, Phys. Rev. Lett. 113, 076802 (2014). 16 W. Zhao, Z. Ghorannevis, L. Chu, M. Toh, C. Kloc, P.-H. Tan, and G. Eda, ACS Nano 7, 791 (2012). 17 D. Kozawa, R. Kumar, A. Carvalho, K. K. Amara, W. Zhao, S. Wang, M. Toh, R. M. Ribeiro, A. H. C. Neto, K. Matsuda, and G. Eda, Nat. Commun. 5, 4543 (2014). 18 H. Zeng, G.-B. Liu, J. Dai, Y. Yan, B. Zhu, R. He, L. Xie, S. Xu, X. Chen, W. Yao, and X. Cui, Sci. Rep. 3, 1608 (2013). 19 W. Li, A. G. Birdwell, M. Amani, R. A. Burke, X. Ling, Y.-H. Lee, X. Liang, L. Peng, C. A. Richter, J. Kong, D. J. Gundlach, and N. V. Nguyen, Phys. Rev. B 90, 195434 (2014). 20 B. Zhu, X. Chen, and X. Cui, Sci. Rep. 5, 9218 (2015). 21 A. R. Klots, A. K. M. Newaz, B. Wang, D. Prasai, H. Krzyzanowska, J. Lin, D. Caudel, N. J. Ghimire, J. Yan, B. L. Ivanov, K. A. Velizhanin, A. Burger, D. G. Mandrus, N. H. Tolk, S. T. Pantelides, and K. I. Bolotin, Sci. Rep. 4, 6608 (2014). 22 A. Hanbicki, M. Currie, G. Kioseoglou, A. Friedman, and B. Jonker, Sol. State Commun. 203, 16 (2015). 7 075451 (2013). 035009 (2016). 23 K. Kośmider and J. Fernández-Rossier, Phys. Rev. B 87, 39 A. O. Slobodeniuk and D. M. Basko, 2D Materials 3, 24 K. Kośmider, J. W. González, and J. Fernández-Rossier, Phys. Rev. B 88, 245436 (2013). 40 H. Dery and Y. Song, Phys. Rev. B 92, 125431 (2015). 41 Y. Chen, B. Gil, P. Lefebvre, and H. Mathieu, Phys. Rev. 25 G.-B. Liu, W.-Y. Shan, Y. Yao, W. Yao, and D. Xiao, B 37, 6429 (1988). Phys. Rev. B 88, 085433 (2013). 26 A. Kormányos, G. Burkard, M. Gmitra, J. Fabian, V. Zó- lyomi, N. D. Drummond, and V. Fal'ko, 2D Materials 2, 022001 (2015). 27 J. P. Echeverry, B. Urbaszek, T. Amand, X. Marie, and I. C. Gerber, Phys. Rev. B 93, 121107 (2016). 28 D. Y. Qiu, T. Cao, and S. G. Louie, Phys. Rev. Lett. 115, 176801 (2015). 29 M. M. Glazov, T. Amand, X. Marie, D. Lagarde, L. Bouet, and B. Urbaszek, Phys. Rev. B 89, 201302 (2014). 30 T. Smoleński, M. Goryca, M. Koperski, C. Faugeras, T. Kazimierczuk, A. Bogucki, K. Nogajewski, P. Kossacki, and M. Potemski, Phys. Rev. X 6, 021024 (2016). 31 A. Arora, M. Koperski, K. Nogajewski, J. Marcus, C. Faugeras, and M. Potemski, Nanoscale 7, 10421 (2015). 32 X.-X. Zhang, Y. You, S. Y. F. Zhao, and T. F. Heinz, Phys. Rev. Lett. 115, 257403 (2015). 33 G. Wang, C. Robert, A. Suslu, B. Chen, S. Yang, S. Alam- dari, I. C. Gerber, T. Amand, X. Marie, S. Tongay, and B. Urbaszek, Nat. Commun. 6, 10110 (2015). 34 J. Brandt, D. Fröhlich, C. Sandfort, M. Bayer, H. Stolz, and N. Naka, Phys. Rev. Lett. 99, 217403 (2007). 35 M. Nirmal, D. J. Norris, M. Kuno, M. G. Bawendi, A. L. Efros, and M. Rosen, Phys. Rev. Lett. 75, 3728 (1995). 36 M. Bayer, O. Stern, A. Kuther, and A. Forchel, Phys. Rev. B 61, 7273 (2000). 37 S. Zaric, G. N. Ostojic, J. Kono, J. Shaver, V. C. Moore, M. S. Strano, R. H. Hauge, R. E. Smalley, and X. Wei, Science 304, 1129 (2004). 38 A. Srivastava, H. Htoon, V. I. Klimov, and J. Kono, Phys. Rev. Lett. 101, 087402 (2008). 42 L. C. Andreani and F. Bassani, Phys. Rev. B 41, 7536 (1990). 43 A. Castellanos-Gomez, M. Buscema, R. Molenaar, V. Singh, L. Janssen, H. S. J. van der Zant, and G. A. Steele, 2D Materials 1, 011002 (2014). 44 A. A. Mitioglu, P. Plochocka, A. Granados del Aguila, P. C. M. Christianen, G. Deligeorgis, S. Anghel, L. Ku- lyuk, and D. K. Maude, Nano Lett. 15, 4387 (2015). 45 G. Plechinger, P. Nagler, J. Kraus, N. Paradiso, C. Strunk, C. Schüller, and T. Korn, Phys. Stat. Sol. RRL 9, 457 (2015). 46 J. Shang, X. Shen, C. Cong, N. Peimyoo, B. Cao, M. Egin- ligil, and T. Yu, ACS Nano 9, 647 (2015). 47 Y. Li, J. Ludwig, T. Low, A. Chernikov, X. Cui, G. Arefe, Y. D. Kim, A. M. van der Zande, A. Rigosi, H. M. Hill, S. H. Kim, J. Hone, Z. Li, D. Smirnov, and T. F. Heinz, Phys. Rev. Lett. 113, 266804 (2014). 48 A. Arora, K. Nogajewski, M. Molas, M. Koperski, and M. Potemski, Nanoscale 7, 20769 (2015). 49 F. Cadiz, C. Robert, G. Wang, W. Kong, X. Fan, M. Blei, D. Lagarde, M. Gay, M. Manca, T. Taniguchi, K. Watan- abe, T. Amand, X. Marie, P. Renucci, S. Tongay, and B. Urbaszek, 2D Materials 3, 045008 (2016). 50 F. Cadiz, S. Tricard, M. Gay, D. Lagarde, G. Wang, C. Robert, P. Renucci, B. Urbaszek, and X. Marie, Appl. Phys. Lett. 108, 251106 (2016). 51 I. G. Lezama, A. Arora, A. Ubaldini, C. Barreteau, E. Gi- annini, M. Potemski, and A. F. Morpurgo, Nano Lett. 15, 2336 (2015). 

1710.00048

1

1710

Tailoring Magnetic Skyrmions by Geometric Confinement of Magnetic Structures

Nanoscale magnetic skyrmions have interesting static and transport properties that make them candidates for future spintronic devices. Control and manipulation of the size and behavior of skyrmions is thus of crucial importance. Using a Ginzburg-Landau approach, we show theoretically that skyrmions and skyrmion lattices can be stabilized by a spatial modulation of the uniaxial magnetic anisotropy in a thin film of centro-symmetric ferromagnet. Remarkably, the skyrmion size is determined by the ratio of the exchange length and the period of the spatial modulation of the anisotropy, at variance with conventional skyrmions stabilized by dipolar and Dzyaloshinskii--Moriya interactions (DMIs).

cond-mat.mes-hall

cond-mat

Tailoring Magnetic Skyrmions by Geometric Confinement of Magnetic Structures Material Science Division, Argonne National Laboratory, Lemont, Illinois 60439, USA and Department of Physics and Astronomy, University of Missouri, Columbia, Missouri 65211, USA Steven S.-L. Zhang∗ Material Science Division, Argonne National Laboratory, Lemont, Illinois 60439, USA C. Phatak Material Science Division, Argonne National Laboratory, Lemont, Illinois 60439, USA and Department of Materials Science and Engineering, Northwestern University, Evanston, Illinois 60208, USA A. K. Petford-Long Material Science Division, Argonne National Laboratory, Lemont, Illinois 60439, USA and O. G. Heinonen Northwestern-Argonne Institute of Science and Technology, 2145 Sheridan Road, Evanston, Illinois 60208, USA (Dated: September 29, 2018) Nanoscale magnetic skyrmions have interesting static and transport properties that make them candidates for future spintronic devices. Control and manipulation of the size and behavior of skyrmions is thus of crucial importance. Using a Ginzburg-Landau approach, we show theoretically that skyrmions and skyrmion lattices can be stabilized by a spatial modulation of the uniaxial magnetic anisotropy in a thin film of centro-symmetric ferromagnet. Remarkably, the skyrmion size is determined by the ratio of the exchange length and the period of the spatial modulation of the anisotropy, at variance with conventional skyrmions stabilized by dipolar and Dzyaloshinskii -- Moriya interactions (DMIs). I. INTRODUCTION Two dimensional magnetic skyrmions are nanoscale spin textures that are topologically protected: the spin structure of an individual skyrmion is associated with an integer winding number which cannot be continuously changed into another integer number without overcoming a finite energy barrier1 -- 5. The creation, annihilation and transport of magnetic skyrmions strongly rely on their topological properties6 -- 12, which make them promising candidates as spin information carriers in future spintronic devices. Multiple formation mechanisms of magnetic skyrmions have been identified2,6 -- 9,13 -- 15. Most commonly, stable skyrmions are found in bulk chiral magnets such as MnSi2,16,17 and other B20 transition metal alloys18 -- 22. In these systems, strong spin-orbit coupling conspires with broken bulk inversion symmetry to give rise to the Dzyaloshinskii -- Moriya interactions (DMIs)23,24 that favor canted spin structure and thus can stabilize skyrmions with definite chi- rality. DMI can also be induced in a nonchiral transition metal thin film in contact with a heavy metal layer25,26. This gives rise to structural inversion-symmetry breaking and strong interfacial spin orbit interaction, which can support the formation of skyrmions. Even in the absence of DMI, long range dipolar interactions alone may stabilize skyrmions, or magnetic bubbles, as well, but the size of this type of skyrmion (∼ 0.1 to 1 µm) 3,15, is usually larger than that stabilized by DMI as it scales with the ratio of the exchange coupling to the dipolar interaction. This type of skyrmion will not, in the absence of DMI, have a distinct chirality, and both chiralities are degenerate in energy. Crystallization of skyrmions occurs when the inter-skyrmion distance is sufficiently reduced so that the repulsive skyrmion-skyrmion interaction leads to a packing in a hexagonal lattice.27 -- 29 A skyrmion crystal phase (SkX) has been observed both in thin films of chiral magnets16,30 and magnetic multilayer with perpendicular anisotropy31. In chiral magnets with very low Curie temperatures, the SkX phase is stabilized at temperatures well below room temperature and requires an external magnetic field (of the order of 1 T 1,3) perpendicular to the film plane. In case of magnetic thin films as well, a magnetic field perpendicular to the film plane is required for the strip domain ground state to evolve into a (chiral) bubble lattice8,10 -- 12,32. Better control of the physical properties of magnetic skyrmions, such as their stability, size, chirality, etc., is not only of fundamental interest but also crucial for the application of skyrmions in spintronics. Some recent research has focused on this control. Small individual skyrmions (with diameters smaller than 100 nm) were stabilized at room temperature by additive interfacial DMIs23,24 in PtCoIr multilayers32. Nucleation of magnetic skyrmions with a wide range of sizes and ellipticities was recently observed in a wedge-shaped FeGe nanostripe33. Montoya et. al. experimentally demonstrated34 that the stability and size of skyrmions originating from dipolar interaction 7 1 0 2 p e S 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 4 0 0 0 . 0 1 7 1 : v i X r a 2 can be controlled by tuning the magnetic properties such as the magnitude of the perpendicular uniaxial and shape anisotropies of FeGd multilayers. In addition to chiral magnets and transition-metal multilayers, stable skyrmions may also be hosted in centrosym- metric systems including various multiferroic materials35 -- 39. An inherent advantage of multiferroic materials is that they are characterized by more than one order parameter, which may impart unique features to the skyrmions. For example, helicity reversal inside skyrmions was observed in Sc-doped hexagonal barium ferrite36, a multiferroic with tunable magnetic anisotropy. More recently, a room temperature SkX state was observed in thin films of the centro- symmetric material Ni2MnGa39 in the absence of DMI. Surprisingly, in that work all skyrmions in each realization of a SkX state had the same chirality, but different realizations may exhibit different chiralities, in contrast with the SkXs stabilized via DMI and dipolar interactions. The degeneracy of the different chiralities opens up the possibility of controlling and altering the chirality of such SkXs, which may enable new applications. Phatak and coworkers associated the formation of this type of SkX with the geometric confinement of magnetic structures by narrow twin variants39 with alternating in-plane and out-of-plane uniaxial magnetic anisotropy. Inspired by these experimental works, we investigate theoretically the energy landscape of various magnetic states in a ferromagnetic thin film that arises from the competition between exchange, shape, and modulated uniaxial anisotropy with a goal of understanding the phase diagram and under what conditions a Skx can be stabilized by these competing energy terms. We generalize the Ginzburg-Landau (GL) theory40 for an Ising ferromagnet to take into account the general three dimensional magnetization in a ferromagnetic thin film. In particular, we will consider a novel uniaxial anisotropy with periodic in-plane to out-of-plane spatial variation of the easy axis, which can be realized in a multiferroic material such as Ni2MnGa39. Intuitively, this form of anisotropy favors a helical ground state as well as a canted-spin state with an antiferromagnetic arrangement in consecutive in-plane anisotropy twins, as shown schematically in Fig. 1; we shall show that the SkX can form during the transition between these two dominant magnetic phases. We will discuss the stability of the SkX phase as a function of temperature and external magnetic field. Furthermore, we derive an explicit expression for the anisotropy energy density of the SkX. This allows us to determine the size-dependence of the skyrmions on the ratio of the exchange length and the period of the spatial variation of the anisotropy, which is a hallmark of this unconventional SkX. FIG. 1: Schematics of three typical magnetic states in a thin film where the anisotropy exhibits periodic spatial variation of easy axis (EA) from in-plane (denoted by ↔) to out-of-plane (denoted by (cid:12)⊗). II. GINZBURG-LANDAU MODEL We assume that the film lies in the x − y plane, and that the film thickness is sufficiently thin so that the magneti- zation density M is uniform along the z-direction and hence is a function only of x and y, i.e., M = M (x, y). In the GL theory, the spatial average of the total magnetic free energy Ftot of a ferromagnetic thin film of area S may be written as (1) where m = M/M0 with M = M (T ) the local magnetization density at temperature T and M0 = M(T → 0), t and u are GL parameters that are in general functions of temperature and external magnetic field, µ0 is the magnetic (cid:90) (cid:104) Ftot = S−1 Aex (∇m)2 + tm2 + u(cid:0)m2(cid:1)2 z + fa (m)(cid:3) , d2x −µ0m · HM0 + Kdm2 permeability, and Kdm2 z denotes the demagnetizing energy density in the thin film approximation41. We shall draw particular attention to the anisotropy energy density fa (m). The uniaxial magnetic anisotropy with spatially varying easy axis may be modeled as 2 µ0M 2 0 m2 z = 1 fa [m (x)] = −Ku z + κ− (x) m2 y (cid:2)κ+ (x) m2 (cid:3) , 3 (2) (cid:104)(cid:12)(cid:12)(cid:12)cos (cid:16) πx wt (cid:17)(cid:12)(cid:12)(cid:12) ± cos where κ±(x) ≡ 1 the anisotropy. 2 (cid:17)(cid:105) (cid:16) πx wt with wt is the twin width, and Ku (> 0) characterizes the magnitude of A generalized spatial profile of the magnetization of a skyrmion lattice can be approximated as a superposition of three spin helices2,3 m (r) = [m0 + mi,⊥ cos (ki · r)] z (cid:16) (cid:17) + mi,(cid:107) z × ki sin (ki · r) , (3) where m0 is the uniform magnetization induced by the external magnetic field perpendicular to the film plane, mi,⊥ and mi,(cid:107) are the out-of-plane and in-plane components of the magnetization, and we use Einstein's summation convention over repeated indices. The wave vectors of the three helices are all in the plane of the layer and form an angle of 120◦ with each other; explicitly, we choose k1 = qx, k2 = q, and ki = ki/ki is the unit vector of ki. The general spatial profile given by Eq. (3) can be reduced to multiple magnetic states including: (i) Helix (single-q state) when m1,(cid:107) ≈ m1,⊥ (cid:54)= 0 and mi,(cid:107) = mi,⊥ = 0 (i = 2, 3); (ii) Stripe domain when only m1,⊥ is nonzero; (iii) (nonchiral) bubble lattice when mi,⊥ (cid:54)= 0 and mi,(cid:107) = 0 (i = 1, 2, 3); (iv) SkX (triple-q state) when mi,(cid:107) ≈ mi,⊥ (cid:54)= 0 and mi,⊥mi,(cid:107) (i = 1, 2, 3) have the same sign so that the three superimposed helices exhibit the same chirality. Other magnetic states are possible as we will mention below. (cid:16)− 1 2 x − √ 3 2 y (cid:16)− 1 q and k3 = √ 3 2 y 2 x + (cid:17) (cid:17) To simplify the problem without loss of generality, we shall assume that the magnitude of the magnetization has mirror symmetry about the y = 0 plane, and let m2,⊥ = m3,⊥, and m2,(cid:107) = m3,(cid:107) . By placing Eq. (3) in Eq. (1) and carrying out the integration in Eq. (1), one can obtain the spatially averaged free energy density. The expression for this is complicated and not very instructive, and is given in the supplementary material. 42,43, i.e., δf = χ(cid:0)m2 − m2 The global minimum of the magnetic free energy density can be computed with the following six variational parameters: m0, m1,(cid:107), m2,(cid:107), m1,⊥, m2,⊥ and q. To specify the two GL parameters t and u, we introduce an extra positive definite free energy density term that arises when the magnetization density deviates from its uniform bulk . Here, χ is a positive definite parameter which in principle relies on saturation value Ms the magnetic property of the material, and the reduced saturation magnetization ms(≡ Ms/M0) can be determined by self-consistently solving the equation Ms = M0BS (T, Hz) in the mean field approximation with BS (T, Hz) the s and u = χ. Just below Tc, the mean field equation gives ms ∼ Brillouin function. We thus identify t = −2χm2 (1 − T )1/2, by which one recovers t = kB (T − Tc) /2v0 in the case of a 1-D Ising ferromagnet40 when χ = kBTc/4v0. We also impose the constraint that the spatially averaged magnetization modulus be no greater than the saturation magnetization, i.e., S−1(cid:82) d2rm2 ≤ m2 (cid:1)2 Tc s s. III. RESULTS AND DISCUSSION In Fig. 2, we show the phase diagram in the temperature-field plane, where the field is applied out-of-plane, for several different magnitude of the demagnetizing and anisotropy energy density coefficients, i.e., Kd and Ku. We note that in the absence of uniaxial anisotropy, only the uninteresting uniform magnetized state is observed (not shown). Once the anisotropy is turned on, the helix state prevails at low magnetic fields because of the alternating in-plane and out-of-plane easy axis variants. When the out-of-plane magnetic field Hz increases, the negative z-component of the magnetization diminishes in a manner such that magnetizations with an out-of-plane component start to swirl around in order to lower the exchange energy, similar to the case in chiral magnets3. This gives birth to the SkX phase as seen in Figs 2 (a) and (b). We also note that, for a given magnitude of the anisotropy, the SkX phase region is more extended for larger Kd since then a greater magnetic field is needed to overcome the demagnetizing field. Further increasing the magnetic field leads to the canted-spin state with a large out-of-plane magnetization component parallel to the magnetic field together with small in-plane magnetization components forming an antiferromagnetic arrangement along the x-axis. Also, for a given magnitude of the magnetistatic energy Kd, a larger anisotropy lowers the energy barrier between the helix and canted-spin states and hence makes the SkX phase unstable, as indicated in Figs. 2 (a) and (c). 4 (a) Ku = Ku0 & Kd = Kd0 (b) Ku = Ku0 & Kd = 1.5 Kd0 (c) Ku = 1.5 Ku0 & Kd = Kd0 FIG. 2: Phase diagram in the plane of the out-of-plane magnetic field hz(≡ M0Hz) and temperature T . Material parameters used in the calculation (corresponding to Ni2MnGa44 -- 47): Aex = 1.0× 10−11 J/m, M0 = 6.0× 105 A/m, Ku0 = 2.5× 105 J/m3 and wt = 50 nm. Small variations in the parameters may change the phase boundary, but the topology of the phase diagrams remains the same. FIG. 3: Skyrmion diameter as a function of exchange length for a thin film with total of 50 twin variants, i.e., Nt = 50 and fixed width of twin width of wt = 50 nm. The inset shows the coefficient of anisotropy energy density fu as a function of the diameter of a skyrmion Dsk(= 2π q ). Next, we show that the the size of the skyrmions depends strongly on the magnitude of the anisotropy as well as on the width of the twin variant. In order to see this, let us focus on the free energy density of a standard SkX state given by ¯fsk (q) =(cid:0)t + Kd − π−1Ku (cid:1) m2 (cid:2)fex (q) + fu (q) + 48um2 0 + um4 0 − hzm0 0 + 6πKd (cid:3) m2 sk + 1 4π + 9um0m3 sk + 51 4 um4 sk , (4) where we have set mi,(cid:107) = mi,⊥ = msk (i = 1, 2, 3) in Eq. (3). The q-dependence of the free energy enters through the coefficients of the exchange and anisotropy energy densities given by fex (q) = 12πAexq2 and fu (q) = −Ku [9 − 2ζNt (2wtq) + 2ηNt (2wtq) −ζNt (wtq) + 4ηNt (wtq)] , (5) respectively, where ηNt (x) ≡ sin[(Nt−1)x]+sin(x) π )2] sin( x 2 ) variants which is taken to be an even number without loss of generality [see the supplementary material for a detailed derivation of fu(q)]. with Nt the total number of twin and ζNt (x) ≡ 4Nt[1−( x sin[Ntx] 4Nt[1−( x π )2] sin( x 2 ) The diameter of a skyrmion can thus be determined via Dsk = 2π qm , where qm is the wave vector obtained by 5 (cid:113) Aex minimizing fq = fex + fu. In Fig. 3, we show Dsk as a function of the exchange length lex = 2π which is the length scale of a 360◦ Bloch domain. When lex is greater than the width of the twin wt (corresponding to small anisotropy Ku for fixed exchange stiffness), the size of the skyrmion is comparable to and eventually approaches the lateral size of the film (i.e., L = Ntwt = 2.5 µm for Nt = 50 and wt = 50 nm); in other words, the system is essentially in a uniformly magnetized state in the small anisotropy regime. In the intermediate anisotropy regime where lex (cid:46) wt, a plateau of Dsk = 4wt appears, which agrees with the experimental observation of the the close-packed hexagonal skyrmion lattice in the narrow twinned region of Ni2MnGa. Finally, in the large anisotropy limit where the exchange length is much smaller than the twin width, another plateau of Dsk = 2wt appears. These two plateaus result from the two local minimum in the anisotropy density characterized by the function fu given by Eq. (5) at Dsk = 2wt and Dsk = 4wt, as shown by the inset of the Fig. 3. Ku Before we close this section, we briefly discuss the chirality and magnetic field dependence of the SkX. It was observed experimentally that the Ni2MnGa crystal with inversion symmetry (and thus no DMI) still may host a SkX with a single chirality, in contrast to the SkX stabilized by long range dipolar interaction for which the chirality of each individual skyrmion could in principle be completely random. By adopting our Ansatz magnetization profile, we presumed all skyrmions have the same chirality, but our analysis provides several hints about the origin of the fixed chirality of the SkX in this system. First, as shown in the phase diagram (Fig. 2), the helix state is favored at low magnetic fields. Continuous transition from the helix state to SkX state necessitates a single chirality of the SkX even in the absence of the DMI, since the in-plane magnetization orientation must conform to that of the helix in order to lower the exchange energy cost during the transition. Second, as the separation distance between two neighboring skyrmions becomes shorter, same chirality becomes energetically preferable considering that the in-plane magnetization component along a line connecting the centers of the two skyrmions with opposite chiralities would carry higher order harmonics and thus results in higher exchange energy. We finally comment on the dependence of the stability of the SkX hosted in the Ni2MnGa system on the exter- nal magnetic field. Based on our theoretical model, a small but finite magnetic field is required to overcome the demagnetizing field and stabilizing the SkX phase; the magnetic field is of the order of 50 Oe as estimated for an Ni2MnGa ultrathin film of 10 monolayers at room temperature, where hzv0/kBTc (cid:39) 0.05, v0 = a2 md with the mag- netic spacing am ∼ 10 A47,48 and d the thickness. Experimentally, SkX was observed even in the absence of external magnetic field39. The observed (metastable) zero-field SkX may arise from the history-dependence of the system, e.g., a quenched SkX. Furthermore, we note that Ni2MnGa is a ferromagnetic shape memory alloy, and there is a strain energy and an associated internal magnetic field induced by the magneto-elastic coupling involved in the martensite transformation; this is not considered explicitly in our present model, but would be interesting for future studies, as stabilization of skyrmions with zero magnetic field will be beneficial for the application of skyrmions in future electronic devices. IV. SUMMARY AND OUTLOOK In this work, we generalized the Ginzburg-Landau theory for Ising ferromagnets to include a general continuum magnetization profile that encapsulates various magnetic configurations with three dimensional magnetization direc- tions. We demonstrated that stabilization of room temperature SkX can be facilitated by geometric modulation of the uniaxial anisotropy easy axis in nonchiral materials with inversion symmetry. Remarkably, the size of the skyrmions can be tailored by the period of the spatial modulation of the anisotropy, in contrast with skyrmions originating from dipolar interaction and DMI. Such novel uniaxial anisotropy was realized experimentally in a multiferroic material Ni2MnGa with narrow twin variants for which the anisotropy easy axis is rotated by 90◦ across the twin boundary, and our work explains the underlying physics that gives rise to the observed SkX in Ni2MnGa. Our model can be further generalized to take into account other magnetic interactions such as the DMI as well as various forms of geometric confinement on magnetic structures. It will also be very intriguing to investigate the transport and dynamic behaviors of skyrmions in the presence of geometric modulation of the anisotropy. For example, with the nonuniform anisotropy that we considered here, one would expect the skyrmions to respond rather differently when they are driven along and perpendicular to the twin boundaries. It may also be possible to control and alter the chirality of the SkX using, e.g., strain or electrical currents. These centrosymmetric SkXs may therefore enable interesting applications in spintronics. Work by S.S.-L.Z, C.P, A.P.-L, O.H was supported by Department of Energy, Office of Science, Materials Sciences and Engineering Division. Initial work by S.S.-L.Z was also partly supported by NSF Grants DMR-1406568. Acknowledgements 6 Appendix A: General expression of the spatially averaged free energy density By placing the general magnetization profile (Eq. (3) in the main text) in the magnetic free energy expression given by Eq. (1) in the main text, we obtain the following spatially average free energy density 0 + um4 ¯f =(cid:0)t + Kd − π−1Ku 0 (cid:1) m2 (cid:8)π(cid:0)Aexq2 + t + 2um2 (cid:8)π(cid:0)Aexq2 + t + 6um2 (cid:8)4π(cid:0)Aexq2 + t + 2um2 (cid:8)π(cid:0)Aexq2 + t + 6um2 (cid:104) (cid:104) (cid:16) 1,⊥ + 2m2 1,(cid:107) + 3m4 3m4 1,(cid:107)m2 2,(cid:107) + 2m2 1,⊥m2 m2 + + + + 1 2π 1 2π 1 4π 1 π + u 8 + 12 + um0 2m1,(cid:107)m2,(cid:107)m2,⊥ + m1,⊥ 0 + Kd 1,(cid:107) 0 − hzm0 (cid:1) − Ku [1 − ζNt (2wtq)](cid:9) m2 (cid:1) − Ku [1 + ηNt (2wtq)](cid:9) m2 (cid:1) − Ku [1 − ζNt (wtq)](cid:9) m2 (cid:1) − Ku [1 + ηNt (wtq)](cid:9) m2 (cid:16) (cid:16) (cid:17)(cid:105) 1,⊥ + 12m4 2,(cid:107) + 18m4 2,(cid:107) + 6m2 2,⊥ + 12m2 (cid:17)(cid:105) (cid:17) 2,⊥ 2,⊥ m2 2,(cid:107) 0 m2 2,(cid:107)m2 1,⊥ + m2 2,⊥m2 1,(cid:107) + 8 1,(cid:107)m2 2,⊥ 0 + Kd 1,⊥ 2,(cid:107)m2 2,⊥ . (A1) where hz = M0Hz, and we have assumed, for simplicity without losing generality, that the magnitude of the mag- netization has mirror symmetry about the y = 0 plane, and let m2,⊥ = m3,⊥, and m2,(cid:107) = m3,(cid:107); the two functions ηNt (wtq) and ζNt (wtq), with Nt the total number of twin layers, are derived from the spatial integral of the anisotropy energy density. We will present the detailed derivation of the anisotropy free energy density term in the next section. Appendix B: Derivation of the anisotropy free energy density term Let us consider the uniaxial magnetic anisotropy of the form with and K⊥ (x) = 1 2 Ku K(cid:107) (x) = 1 2 Ku fa [m (x)] = −K⊥ (x) m2 x + z − K(cid:107) (x) m2 (cid:12)(cid:12)(cid:12)(cid:12)cos (cid:19) (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) − cos (cid:18) πx (cid:18) πx (cid:19)(cid:12)(cid:12)(cid:12)(cid:12)(cid:21) (cid:19)(cid:21) wt wt , cos (cid:20) (cid:20)(cid:12)(cid:12)(cid:12)(cid:12)cos (cid:18) πx (cid:18) πx wt wt (A2) (A3) (A4) where the prefactor Ku measures the magnitude of the anisotropy energy density, wt is the period of the spatial variation of the anisotropy (or the width of the twin). 7 For the interesting case of q > π/L with L the side length of the rectangular film, the spatially averaged anisotropy energy can be calculated by the following piece-wise integration ¯fa = − √ 3q 4π ¯fa = − Ku 4πNt (cid:90) 2π√ 3q − 2π√ 3q Nt/2−1(cid:88) n=0 where an (x) = cos (2nx) cos(cid:0) x ing series are given as follows and l(cid:88) n=0 bn (x) = n=0 2 dy n=0 2 )wt 2 )wt 4m2 dx cos (cid:40) (2n− 1 (cid:34)(cid:90) (2n+ 1 0 + 2m2 Ku Ntwt 1,⊥ + 4m2 2,⊥ + 2m2 Nt/2−1(cid:88) (cid:16) (cid:16)  (cid:1) and bn (x) = cos [(2n + 1) x] cos(cid:0) x l(cid:88) l(cid:88) (cid:17) (cid:1)2 − 1 (cid:17) (cid:16) x 4m0m1,⊥ + 2m2 (cid:0) wtq an (x) = cos (cid:16) x cos [(2n + 1) x] cos (x/2) = cos an (wtq) −2 2,⊥ n=0 2 π + 2 2 (cid:17) 2 (cid:18) πx (cid:19) wt (cid:90) (2n+ 3 2 )wt (cid:18) πx (cid:19) (cid:35) , wt m2 y 2 )wt 2m2 (A5) (2n+ 1 dx cos 1,(cid:107)bn (2wtq) z − m2 (cid:17) (cid:0) 2wtq (cid:1), and the results of the summation of correspond- (cid:0) 2wtq (cid:1)2 − 1  , (cid:1)2 − 1 1 −(cid:0) wtq 1,⊥an (2wtq) 2,(cid:107)bn (wtq) − m2 (cid:1)2 (A6) m2 + π π π 1,(cid:107) + m2 2,(cid:107) csc (wtq) cos (lwtq) sin [(l + 1) wtq] (A7) csc (x) cos [(l + 1) x] sin [(l + 1) x] . (A8) where we have assumed Nt being an large even integer number. By carrying out the integration, we obtain Placing Eqs. (A7) and (A8) in Eq. (A6), we obtain (cid:8)4m2 ¯fa = − Ku 4π 1,(cid:107) [1 − ζNt (2wtq)] + m2 0 + 2m2 +2m2 1,⊥ [1 + ηNt (2wtq)] + 4m2 (cid:111) 2,(cid:107) [1 − ζNt (wtq)] − 8m0m1,⊥ζNt (wtq) 2,⊥ [1 + ηNt (wtq)] where and ζNt (wtq) = π2 sin (Ntwtq) 4Nt (π2 − q2w2 t ) sin(cid:0) wtq (cid:1) t ) sin(cid:0) wtq (cid:1) . ηNt (wtq) = π2 sin [(Nt − 1) wtq] + sin (wtq) 4Nt (π2 − q2w2 2 2 , (A9) (A10) (A11) 2 and ηNt (x) ≤ 1 Note that ζNt (x) ≤ 1 2 , and the cross-term of m0m1,⊥ in Eq. (A9) vanishes when one carries out the spatial integration with an overall sinusoidal envelop function with the period of the film length on top of the magnetization profile [corresponding to several replica of the entire thin film]. Taking this into account, we arrive at the final expression of the anisotropy energy density in the presence of the geometric confinement with only quadratic terms, i.e., ¯fa = − Ku 4π 1,(cid:107) [1 − ζNt (2wtq)] + m2 (cid:111) 1,⊥ [1 + ηNt (2wtq)] + 4m2 2,(cid:107) [1 − ζNt (wtq)] 0 + 2m2 +2m2 2,⊥ [1 + ηNt (wtq)] . (A12) (cid:8)4m2 The above expression is valid when q > π/Li, where Li (i = x or y) are side lengths of the rectangular thin film. ∗ Electronic address: [email protected] 8 1 A. Fert, N. Reyren, and V. Cros, Nat. Rev. Mater. 2, 17031 (2017). 2 S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Boni, Science 323, 915 (2009). 3 N. Nagaosa and Y. Tokura, Nat Nano 8, 899 (2013). 4 F. Hellman, A. Hoffmann, Y. Tserkovnyak, G. S. D. Beach, E. E. Fullerton, C. Leighton, A. H. MacDonald, D. C. Ralph, D. A. Arena, H. A. Durr, P. Fischer, J. Grollier, J. P. Heremans, T. Jungwirth, A. V. Kimel, B. Koopmans, I. N. Krivorotov, S. J. May, A. K. Petford-Long, J. M. Rondinelli, N. Samarth, I. K. Schuller, A. N. Slavin, M. D. Stiles, O. Tchernyshyov, A. Thiaville, and B. L. Zink, Rev. Mod. Phys. 89, 025006 (2017). 5 S. Rohart, J. Miltat, and A. Thiaville, Phys. Rev. B 93, 214412 (2016). 6 T. Okubo, S. Chung, and H. Kawamura, Phys. Rev. Lett. 108, 017206 (2012). 7 S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer, and S. Blugel, Nat Phys 7, 713 (2011). 8 W. Jiang, P. Upadhyaya, W. Zhang, G. Yu, M. B. Jungfleisch, F. Y. Fradin, J. E. Pearson, Y. Tserkovnyak, K. L. Wang, O. Heinonen, S. G. E. te Velthuis, and A. Hoffmann, Science 349, 283 (2015). 9 O. Heinonen, W. Jiang, H. Somaily, S. G. E. te Velthuis, and A. Hoffmann, Phys. Rev. B 93, 094407 (2016). 10 S. Woo, K. Litzius, B. Kruger, M.-Y. Im, L. Caretta, K. Richter, M. Mann, A. Krone, R. M. Reeve, M. Weigand, P. Agrawal, I. Lemesh, M.-A. Mawass, P. Fischer, M. Klaui, and G. S. D. Beach, Nat Mater 15, 501 (2016). 11 O. Boulle, J. Vogel, H. Yang, S. Pizzini, D. de Souza Chaves, A. Locatelli, T. O. Mentes, A. Sala, L. D. Buda-Prejbeanu, O. Klein, M. Belmeguenai, Y. Roussign´e, A. Stashkevich, S. M. Ch´erif, L. Aballe, M. Foerster, M. Chshiev, S. Auffret, I. M. Miron, and G. Gaudin, Nat Nano 11, 449 (2016). 12 S. Zhang, A. K. Petford-Long, and C. Phatak, Sci. Rep. 6, 31248 (2016). 13 Y. S. Lin, P. J. Grundy, and E. A. Giess, Appl. Phys. Lett. 23, 485 (1973), http://dx.doi.org/10.1063/1.1654968 . 14 S. Takao, J. Magn. Magn. Mater. 31, 1009 (1983). 15 A. P. Malozemoff and J. C. Slonczewski, Magnetic Domain Walls in Bubble Materials, edited by R. Wolfe (Academic Press, 1979). 16 B. Lebech, P. Harris, J. S. Pedersen, K. Mortensen, C. Gregory, N. Bernhoeft, M. Jermy, and S. Brown, J. Magn. Magn. Mater. 140, 119 (1995). 17 A. Bauer and C. Pfleiderer, Phys. Rev. B 85, 214418 (2012). 18 S. V. Grigoriev, D. Chernyshov, V. A. Dyadkin, V. Dmitriev, S. V. Maleyev, E. V. Moskvin, D. Menzel, J. Schoenes, and H. Eckerlebe, Phys. Rev. Lett. 102, 037204 (2009). 19 M. Uchida, N. Nagaosa, J. P. He, Y. Kaneko, S. Iguchi, Y. Matsui, and Y. Tokura, Phys. Rev. B 77, 184402 (2008). 20 X. Z. Yu, N. Kanazawa, Y. Onose, K. Kimoto, W. Z. Zhang, S. Ishiwata, Y. Matsui, and Y. Tokura, Nat Mater 10, 106 (2011). 21 H. Wilhelm, M. Baenitz, M. Schmidt, U. K. Rossler, A. A. Leonov, and A. N. Bogdanov, Phys. Rev. Lett. 107, 127203 (2011). 22 K. Shibata, Z. X. Yu, T. Hara, D. Morikawa, N. Kanazawa, K. Kimoto, S. Ishiwata, Y. Matsui, and Y. Tokura, Nat Nano 8, 723 (2013). 23 I. Dzyaloshinsky, J Phys Chem Solids 4, 241 (1958). 24 T. Moriya, Phys. Rev. 120, 91 (1960). 25 S. Emori, U. Bauer, S.-M. Ahn, E. Martinez, and G. S. D. Beach, Nat Mater 12, 611 (2013). 26 K.-S. Ryu, L. Thomas, S.-H. Yang, and S. Parkin, Nat Nano 8, 527 (2013). 27 W. L. McMillan, Phys. Rev. B 12, 1187 (1975). 28 A. Bocdanov and A. Hubert, physica status solidi (b) 186, 527 (1994). 29 A. O. Leonov, T. L. Monchesky, N. Romming, A. Kubetzka, A. N. Bogdanov, and R. Wiesendanger, New J. Phys. 18, 065003 (2016). 30 A. Bauer and C. Pfleiderer, Phys. Rev. B 85, 214418 (2012). 31 S. A. Montoya, S. Couture, J. J. Chess, J. C. T. Lee, N. Kent, D. Henze, S. K. Sinha, M.-Y. Im, S. D. Kevan, P. Fischer, B. J. McMorran, V. Lomakin, S. Roy, and E. E. Fullerton, Phys. Rev. B 95, 024415 (2017). 32 C. Moreau-Luchaire, C. Moutas, N. Reyren, J. Sampaio, C. A. F. Vaz, N. Van Horne, K. Bouzehouane, K. Garcia, C. Der- anlot, P. Warnicke, P. Wohlhuter, J.-M. George, M. Weigand, J. Raabe, V. Cros, and A. Fert, Nat Nano 11, 444 (2016). 33 C. Jin, Z.-A. Li, A. Kov´acs, J. Caron, F. Zheng, F. N. Rybakov, N. S. Kiselev, H. Du, S. Blugel, M. Tian, Y. Zhang, M. Farle, and R. E. Dunin-Borkowski, Nat Commun 8, 15569 (2017). 34 S. A. Montoya, S. Couture, J. J. Chess, J. C. T. Lee, N. Kent, D. Henze, S. K. Sinha, M.-Y. Im, S. D. Kevan, P. Fischer, B. J. McMorran, V. Lomakin, S. Roy, and E. E. Fullerton, Phys. Rev. B 95, 024415 (2017). 35 S. Seki, X. Z. Yu, S. Ishiwata, and Y. Tokura, Science 336, 198 (2012). 36 X. Yu, M. Mostovoy, Y. Tokunaga, W. Zhang, K. Kimoto, Y. Matsui, Y. Kaneko, N. Nagaosa, and Y. Tokura, PNAS 109, 8856 (2012). 37 M. C. Langner, S. Roy, S. K. Mishra, J. C. T. Lee, X. W. Shi, M. A. Hossain, Y.-D. Chuang, S. Seki, Y. Tokura, S. D. Kevan, and R. W. Schoenlein, Phys. Rev. Lett. 112, 167202 (2014). 38 J. S. White, K. Prsa, P. Huang, A. A. Omrani, I. Zivkovi´c, M. Bartkowiak, H. Berger, A. Magrez, J. L. Gavilano, G. Nagy, J. Zang, and H. M. Rønnow, Phys. Rev. Lett. 113, 107203 (2014). 39 C. Phatak, O. Heinonen, M. D. Graef, and A. Petford-Long, Nano Lett. 16, 4141 (2016). 40 T. Garel and S. Doniach, Phys. Rev. B 26, 325 (1982). 41 K. J. Harte, J. Appl. Phys. 39, 1503 (1968). 42 J. Wang and J. Zhang, Int. J. Solids Struct. 50, 3597 (2013). 43 C. M. Landis, J. Mech. Phys. Solids 56, 3059 (2008). 44 R. Tickle and R. James, J. Magn. Magn. Mater. 195, 627 (1999). 45 O. Heczko, K. Jurek, and K. Ullakko, J. Magn. Magn. Mater. 226, 996 (2001). 46 P. Wu, X. Ma, J. Zhang, and L. Chen, Philos. Mag. 91, 2102 (2011). 47 T. Sakon, Y. Adachi, and T. Kanomata, Metals 3, 202 (2013). 48 L. Righi, F. Albertini, L. Pareti, A. Paoluzi, and G. Calestani, Acta Mater. 55, 5237 (2007). 9 10 11 

1508.05770

2

1508

Broadband architecture for galvanically accessible superconducting microwave resonators

In many hybrid quantum systems, a superconducting circuit is required that combines DC-control with a coplanar waveguide (CPW) microwave resonator. The strategy thus far for applying a DC voltage or current bias to microwave resonators has been to apply the bias through a symmetry point in such a way that it appears as an open circuit for certain frequencies. Here, we introduce a microwave coupler for superconducting CPW cavities in the form of a large shunt capacitance to ground. Such a coupler acts as a broadband mirror for microwaves while providing galvanic connection to the center conductor of the resonator. We demonstrate this approach with a two-port $\lambda/4$-transmission resonator with linewidths in the MHz regime ($Q\sim10^3$) that shows no spurious resonances and apply a voltage bias up to $80$ V without affecting the quality factor of the resonator. This resonator coupling architecture, which is simple to engineer, fabricate and analyse, could have many potential applications in experiments involving superconducting hybrid circuits.

cond-mat.mes-hall

cond-mat

a Broadband architecture for galvanically accessible superconducting microwave resonators Sal J. Bosman,1 Vibhor Singh,1 Alessandro Bruno,1, 2 and Gary A. Steele1 1)Kavli Institute of NanoScience, Delft University of Technology, PO Box 5046, 2600 GA, Delft, The Netherlands. 2)Qutech Advanced Research Center, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands. (Dated: 24 November 2015) In many hybrid quantum systems, a superconducting circuit is required that combines DC-control with a coplanar waveguide (CPW) microwave resonator. The strategy thus far for applying a DC voltage or current bias to microwave resonators has been to apply the bias through a symmetry point in such a way that it appears as an open circuit for certain frequencies. Here, we introduce a microwave coupler for superconducting CPW cavities in the form of a large shunt capacitance to ground. Such a coupler acts as a broadband mirror for microwaves while providing galvanic connection to the center conductor of the resonator. We demonstrate this approach with a two-port λ/4-transmission resonator with linewidths in the MHz regime (Q ∼ 103) that shows no spurious resonances and apply a voltage bias up to 80 V without affecting the quality factor of the resonator. This resonator coupling architecture, which is simple to engineer, fabricate and analyse, could have many potential applications in experiments involving superconducting hybrid circuits. Embedding a quantum system, such as a qubit, in a microwave resonator is an attractive and commonly used approach. Resonators provide isolation that shields the system from environmental noise and can controllably inhibit spontaneous decay. At millikelvin temperatures, microwave resonators are in their ground state, free of en- tropy, while still providing microwave frequency access to read-out and manipulate quantum states. There is a wide range of hybrid systems exploring new quantum phenom- ena and technologies that require a microwave resonator that also offers DC-access to the device, including cou- pling mechanical resonators to qubits1,2, microwave stor- age and conversion circuits3, Josephson and quantum dot radiation4–7, spin qubits8, circuits coupled to ultra cold atoms9, and more10. A cavity with galvanic access al- lows the possibility to measure simultaneously a device’s DC response, such as a current-voltage curve, and its microwave response, like emitted radiation or scattering characteristics. Such a setup could form a bridge between DC quantum transport measurements and all-microwave setups employed in circuit QED, with a wide range of ap- plications from the study of topological and other exotic junctions11, to superconducting molecular junctions12, to carbon nanotubes13,14. While attractive for many applications, applying a DC current or voltage bias to a superconducting res- onator without sacrificing its quality factor is a non- trivial challenge15–18. The first approach to incorporate DC-control into a superconducting microwave resonator was to access the resonator galvanically at a voltage node with a bias line made from a λ/2-section of transmission line15. If the frequencies of resonator and the bias line are perfectly matched, then the bias line loads the circuit with an infinite impedance (an effective open), resulting in no leakage of the microwave field on resonance. How- ever, off-resonant circuit excitations, such as a detuned qubit, are free to decay into the bias-line. This issue FIG. 1. (a) Schematic of a generic two-port cavity for con- fining electromagnetic fields using impedance mismatches as mirrors. The fields inside the cavity are isolated from the input-output fields by local impedance mismatches Z1 and Z2. (b) For a microwave CPW, mirrors can be implemented by incorporating capacitor or inductor elements either in se- ries with the center conductor or from the center conductor to ground. This defines either an quasi-open or -short boundary condition for microwaves and a conducting or non-conducting path for DC signals. (c) Schematic of a cavity design imple- mented here providing DC access to the center conductor of the cavity using a shunt-capacitor as a mirror at port 2. was addressed recently18 using a reflective T-filter, mak- ing the suppression band a few GHz wide and reaching quality factors ∼ 105, at the cost of increased complex- ity. The insertion of the bias into a symmetry point, typically breaking ground plane symmetry, makes these designs susceptible to slot-line modes and spurious res- onances. Particularly for more complex circuits, these resonances can complicate design, operation and analy- sis of the device. In a third approach17, a lumped element resonator circuit was split symmetrically in two such that a DC voltage can be applied to the two halves using iso- lated ground planes without any radiation losses, leading to high quality factors for perfectly balanced designs. Here, we explore a different approach in which we re- 2 FIG. 2. Device: (a) Optical microscope image of the complete λ/4-transmission cavity. Port 1, on the left, is coupled with a gap capacitor, and on the right, port 2, is coupled through a shunt capacitor. (b) Shows a zoom in of the gap capacitor, (c) a zoom in on the shunt capacitor, with black (sapphire), blue (Si3N4), beige (MoRe). (d) Shows an overview of the shunt capacitor, (e) schematic top-view (above) and cross section (below) of the shunt capacitor. The center conductor is depicted in blue, the ground plane in red and the top-plate of the capacitive coupler in green. (f) Shows the equivalent circuit of the shunt capacitor. place the typical gap capacitor used as an input coupler in CPW cavities with a large shunt capacitance to ground. Doing so, we achieve a highly reflective microwave mir- ror in which the center conductor of the waveguide is still galvanically connected to the input line, allowing the ap- plication of a DC current or voltage. Such a design has several advantages: in contrast to resonant filter designs, the reflectivity is broadband up to the self-resonance fre- quency of the shunt capacitor. As no extra port is re- quired for the DC-signal, any energy that leaks through the bias line can contribute to the measurement signal. Finally, this approach does not rely on any symmetry considerations of the cavity, simplifying design and anal- ysis. Fig. 1 shows a schematic of a generic transmission line cavity, consisting of a waveguide that is isolated from the input and output ports by impedance mismatches. At the impedance mismatch points, the propagating waves in the transmission line are reflected and a standing wave forms at resonance. Depending on the choice of impedance mismatch, one creates a boundary condition for the microwave field corresponding to either a voltage node (short), or a current node (open). To couple the cavity to external circuitry one, or both, of the boundary conditions are relaxed, which causes part of the power to be transmitted. In microwave cavities, a voltage node can be implemented by a short to ground, while a cur- rent node by a gap capacitor. These also have analogues in optical cavities: for optics, a short circuit boundary can be implemented by a semi-silvered mirror, while an open-circuit boundary can be implemented by a magnetic mirror19. In CPW microwave resonators, partially transmitting impedance mismatches are typically implemented using lumped-element components such as inductors or capac- itors. In general, there are four types of lossless couplers possible, which are depicted in Fig. 1b. Depending on the choice and configuration of the inductor or capaci- tor one obtains unity reflection (Sij = δij) in the limit ω (cid:55)→ 0 (DC-block) or ω (cid:55)→ ∞ (AC-block). Including that reflection on a short causes a π-phase shift, each quad- rant of Fig.1b can be classified according to its scattering behaviour as Sij(ω) = ±δij, in the appropriate limit20. For a λ/4 CPW resonator made from a transmission line coupled to a single external port, both of impedance Z0, the coupling quality factor Qc for all four types of couplers can be written in the following unified form20: Qc = π 4 1 b±2 c , (1) Γ Qc Cseries [fF] Lseries [nH] Cshunt [pF] Lshunt [pH] .5 3.14 318 .9 254 35.4 .99 31 · 103 3.22 .999 3.1 · 106 .319 1.27 11.5 126 1272 3.18 28.6 315 3180 796 88.4 8.03 0.79 TABLE I. Typical values for different couplers; reflection co- efficients, Γ, coupling Qc’s, and the required value of capaci- tance or inductance, for a single port cavity of 5 GHz of 50 Ω coupled with the coupler to a feedline of 50 Ω. where bc is the normalized susceptance of the coupler with bc = Z0ωC for capacitive couplers and bc = Z0/Lω for inductive couplers, with +/− for a series / shunt con- figuration respectively. In Table 1 we tabulated the val- ues of the inductor and capacitor components required to achieve typical values Qc and reflectivities Γ. In choosing which microwave coupler is most suitable for galvanic access, either the series inductor or the shunt capacitor, consideration of stray reactance is important. For example, a resonator with Qc ∼ 103 coupled with a series inductor would require an inductance of 100 nH with a self-resonance frequency above the cavity reso- nance. Realizing such a high on-chip inductance with a self-resonance frequency above ∼ 8−10 GHz is not prac- tical using geometric inductance, and requires high ki- netic inductance nano-inductors21 or Josephson junction arrays22,23, which suffer from limiting dynamic range and can complicate operation due to their non-linear nature. In contrast, using an on-chip parallel plate capacitor, as we will show here, it is possible to engineer a large ca- pacitance with small stray inductance using conventional thin-film technologies. As a proof of principle, we demonstrate the concept of a shunt capacitor microwave mirror using a λ/4- transmission cavity that incorporates one shunt capacitor and one series capacitor, shown schematically in Fig.1c. Fig. 2 shows an optical microscope image of the device us- ing a CPW cavity and an on-chip parallel plate capacitor. The devices are fabricated in a three step e-beam lithog- raphy process. First we pattern the resonator on sapphire using a layer of ∼ 45 nm of sputtered Molybdenum- Rhenium alloy with reactive-ion etching24. Subsequently we deposit ∼ 100 nm of PECVD Si3N4, which is wet- etched using HF. The upper plate of the capacitor is patterned in a lift-off process using PMMA and a MoRe film of ∼ 90 nm. The resulting top plate acts as a ca- pacitive coupler between the center conductor and the ground plane, see Fig. 1e-f. We estimate the parallel plate contribution to the shunt capacitance to be 27 pF and a stray capacitance from the lower capacitor elec- trode to the ground plane of 3 − 6 pF. From formula 0 ω2C 2/4 ∼ 2000. Using finite 1, we estimate Qc = πZ 2 element analysis simulations20, we estimate the first self- resonance frequency of the capacitor to be ∼ 22 GHz, significantly above the cavity frequency. From equation 1, we need a gap capacitance at port 1 of 10 fF to obtain a symmetrically coupled cavity. For the microwave characterization, we cool the device to T ∼ 15 mK in a radiation-tight microwave box. Using two identical cryogenic microwave reflectometry configu- rations connected to both ports20, we measure the full S- matrix of our device. We apply a DC voltage to the center conductor of our cavity using a bias tee attached to port 2. Fig. 3a shows a measurement of the reflection from port 1 (S11) with a resonance at ω = 2π × 5.3889 GHz, and a linewidth of κtot = 2π × 3.49 MHz. This demon- strates we are able to make superconducting microwave resonators with loaded quality factors in the range of 3 FIG. 3. Microwave network measurement characterizing the scattering matrix of the cavity measured with −135 dBm at the sample. (a) Reflection from port 1, measuring ω = 2π × 5.3889 GHz, κtot = 2π× 3.49 MHz, and κ1 = 2π× 0.51 MHz. (b) Reflection from port 2, measuring the same ω, and κtot, with κ2 = 2π × 2.75 MHz. (c) Transmission from port 2 to 1, we measure a single resonance with the same frequency and linewidth. Q ∼ 103 that are galvanically accessible through a fre- quency non-selective (non-resonant) connection. In addition to measuring the cavity through the gap capacitance, we can also use the shunt capacitor as a partially transmitting mirror. Fig. 3b shows a measure- ment of the cavity resonance through the shunt capac- itor. We observe a resonance at the same frequency ω and linewidth, demonstrating that the coupler works as a mirror. From the fit20, we find a coupling Qc2 = 1960, giving an effective capacitance of 29.5 pF. With the fits of both reflection measurements we can extract the cou- pling rates κ1,2 for each port20. By combining these re- sults we can extract the magnitude of the internal losses using κtot = κ1 + κ2 + κint, and find an internal loss rate of about κint ∼ 2π × 230 kHz. We estimate20 a contribution of 8-80 kHz to the total internal loss rate of the cavity from the dielectric losses of the shunt capac- itor, assuming a dielectric loss tangent δ ∼ 10−3 − 104. Reducing the dielectric thickness to 10 nm, we predict20 that this loss rate could be reduced to 0.8 kHz, limiting the internal Q of the cavity to 6.3 × 105. In transmission, Fig.3c, we observe a single resonance, with a clean spectrum over the full measurement band- 4 large DC voltage to the center conductor of the resonator. By reducing the dielectric layer thickness to 10 nm, we predict that coupling quality factors of up to 3 × 105 should be possible while the first the self-resonance of the capacitor appears at ∼ 8 GHz.20 For larger capacitances the dielectric losses of the deposited dielectric contribute less than a kHz to the internal loss rate of the cavity20, at the expensive of a reduced dielectric breakdown voltage. The simplicity of this broadband technique together with the possibility of large quality factors suggests that this new design could be very attractive for many applica- tions and experiments involving superconducting hybrid circuits. The authors would like to thank Daniel Bothner, Joshua Island, Nodar Samkharadze, David van Wo- erkom, and Martijn Cohen for useful discussions. 1M. LaHaye, J. Suh, P. Echternach, K. Schwab, and M. Roukes, “Nanomechanical measurements of a superconducting qubit,” Nature 459, 960–964 (2009). 2J.-M. Pirkkalainen, S. Cho, J. Li, G. Paraoanu, P. Hakonen, and M. Sillanpaa, “Hybrid circuit cavity quantum electrodynamics with a micromechanical resonator,” Nature 494, 211–215 (2013). 3R. Andrews, A. Reed, K. Cicak, J. Teufel, and K. Lehnert, “Quantum-enabled temporal and spectral mode conversion of mi- crowave signals,” arXiv preprint arXiv:1506.02296 (2015). 4M. Hofheinz, F. Portier, Q. Baudouin, P. Joyez, D. Vion, P. Bertet, P. Roche, and D. Est`eve, “Bright side of the coulomb blockade,” Physical review letters 106, 217005 (2011). 5L. P. Rokhinson, X. Liu, and J. K. Furdyna, “The fractional ac josephson effect in a semiconductor-superconductor nanowire as a signature of majorana particles,” Nature Physics 8, 795–799 (2012). 6F. Chen, J. Li, A. Armour, E. Brahimi, J. Stettenheim, A. Sirois, R. Simmonds, M. Blencowe, and A. Rimberg, “Realization of a single-cooper-pair josephson laser,” Physical Review B 90, 020506 (2014). 7Y.-Y. Liu, J. Stehlik, C. Eichler, M. Gullans, J. Taylor, and J. Petta, “Semiconductor double quantum dot micromaser,” Sci- ence 347, 285–287 (2015). 8K. Petersson, L. McFaul, M. Schroer, M. Jung, J. Taylor, A. Houck, and J. Petta, “Circuit quantum electrodynamics with a spin qubit,” Nature 490, 380–383 (2012). 9D. Bothner, M. Knufinke, H. Hattermann, R. Wolbing, B. Ferdi- nand, P. Weiss, S. Bernon, J. Fort´agh, D. Koelle, and R. Kleiner, “Inductively coupled superconducting half wavelength resonators as persistent current traps for ultracold atoms,” New Journal of Physics 15, 093024 (2013). 10Z.-L. Xiang, S. Ashhab, J. You, and F. Nori, “Hybrid quantum circuits: Superconducting circuits interacting with other quan- tum systems,” Reviews of Modern Physics 85, 623 (2013). 11J. E. Moore, “Viewpoint: An extraordinary josephson junction,” Physics 5, 84 (2012). 12J. Skoldberg, T. Lofwander, V. S. Shumeiko, and M. Fogelstrom, “Spectrum of andreev bound states in a molecule embedded in- side a microwave-excited superconducting junction,” Physical re- view letters 101, 087002 (2008). 13V. Ranjan, G. Puebla-Hellmann, M. Jung, T. Hasler, A. Nunnenkamp, M. Muoth, C. Hierold, A. Wallraff, and C. Schonenberger, “Clean carbon nanotubes coupled to supercon- ducting impedance-matching circuits,” Nature communications 6, doi:10.1038/ncomms8165. 14B. Schneider, S. Etaki, H. van der Zant, and G. Steele, “Cou- pling carbon nanotube mechanics to a superconducting circuit,” Scientific reports 2 (2012). 15F. Chen, A. Sirois, R. Simmonds, and A. Rimberg, “Introduction FIG. 4. Voltage bias characterization: (a) Coupling quality factor to port 2, Qc2 and effective internal quality factor seen by port 2, Qi2, using a reflection measurement through the shunt capacitance (S22), as a function of bias voltage mea- sured at a power of −135 dBm, corresponding to 100 pho- tons. (b) Leakage current as a function of bias voltage. The dashed line indicates a resistance of 56 GΩ. width of 4− 8 GHz, showing that this approach does not suffer from spurious resonances. In figure 4, we characterize the performance of our cavity with a DC voltage bias applied to port 2 using a bias tee. As shown in Fig. 4a, we observe no notable effect on the microwave response up to ∼ 80 V. The same behaviour is also observed at single photon level (input power of −155 dBm). At 80 V, we observe a leakage cur- rent of 1.4 nA, giving a resistance of ∼ 56 GΩ. Above 80 V, we see the onset of dielectric break down with the leakage current rising sharply to 25 nA, leading to an estimated breakdown field of 4 MV/cm. Beyond 80 V, we see a degradation of the internal quality factor, corre- sponding to an increase of ∼ 320 kHz to the internal loss rate. Also we observe a slight increase in the coupling quality factor to port 2, Qc2, implying an increase in the shunt capacitance of about ∼ 5 pF. This could be caused by an inversion layer induced in the Si3N4 dielectric, ef- fectively decreasing the plate separation. The increase of internal losses could be related to Ohmic losses in the inversion layer. To conclude, we have introduced a new type of mi- crowave coupler that allows galvanic access to the center conductor of a CPW microwave resonator. We demon- strated this concept by engineering a λ/4-transmission cavity with a high quality factor in which we can apply a of a dc bias into a high-q superconducting microwave cavity,” Applied Physics Letters 98, 132509 (2011). 16S.-X. Li and J. Kycia, “Applying a direct current bias to su- perconducting microwave resonators by using superconducting quarter wavelength band stop filters,” Applied Physics Letters 102, 242601 (2013). 17S. E. de Graaf, D. Davidovikj, A. Adamyan, S. Kubatkin, and A. Danilov, “Galvanically split superconducting microwave res- onators for introducing internal voltage bias,” Applied Physics Letters 104, 052601 (2014). 18Y. Hao, F. Rouxinol, and M. LaHaye, “Development of a broad- band reflective t-filter for voltage biasing high-q superconducting microwave cavities,” Applied Physics Letters 105, 222603 (2014). 19M. Esfandyarpour, E. C. Garnett, Y. Cui, M. D. McGehee, and M. L. Brongersma, “Metamaterial mirrors in optoelectronic de- vices,” Nature nanotechnology 9, 542–547 (2014). 20See supplementary material. 5 21A. J. Annunziata, D. F. Santavicca, L. Frunzio, G. Catelani, M. J. Rooks, A. Frydman, and D. E. Prober, “Tunable super- conducting nanoinductors,” Nanotechnology 21, 445202 (2010). and K. Lehnert, “Amplification and squeezing of quantum noise with a tunable josephson metamaterial,” Nature Physics 4, 929–931 (2008). 22M. Castellanos-Beltran, K. Irwin, G. Hilton, L. Vale, 23V. E. Manucharyan, J. Koch, L. I. Glazman, and M. H. Devoret, “Fluxonium: Single cooper-pair circuit free of charge offsets,” Science 326, 113–116 (2009). 24V. Singh, B. H. Schneider, S. J. Bosman, E. P. Merkx, and G. A. Steele, “Molybdenum-rhenium alloy based high-q superconduct- ing microwave resonators,” Applied Physics Letters 105, 222601 (2014). 25D. M. Pozar, Microwave engineering (John Wiley & Sons, 2009). Supplementary Material: Broadband architecture for galvanically accessible superconducting microwave resonators 6 Sal J. Bosman1, Vibhor Singh1, Alessandro Bruno1,2, Gary A. Steele1 1Kavli Institute of NanoScience, Delft University of Technology, PO Box 5046, 2600 GA, Delft, The Netherlands. 2Qutech Advanced Research Center, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands. S1. MEASUREMENT SETUP The complete measurement setup used for the device characterization is shown schematically in Fig. S1. It consists of two identical microwave reflectometry measurement chains, such that we have full access to the scattering matrix of the device (S11, S12, S22, S21). The vector network analyser (VNA) outputs a signal that is fed through a variable attenuator at room temperature into the fridge, where it is first heavily attenuated before reaching the sample through a directional coupler. The reflected signal from the device is sent back to the VNA using two isolators and amplifiers. From each reflectometry setup we can measure the reflection of that port and by combining them we can measure transmission in both directions. Additionally we can apply a DC voltage to port 2 of the device through an unfiltered DC-line using a bias-tee just before the device. FIG. S5. Measurement setup: In the dilution refrigerator we have two identical microwave reflectometry setups. The directional coupler near the sample separates the incident and reflected power from the sample. Additionally, the measurement chain coupled to port 2 of the device has a bias-tee to inject the DC voltage bias onto the chip. VNA (Keysight PNA N5222A)SMU (Keysight B2961A)300K4K15mKsource port 1source port 2receiverport 2receiverport 1VPasternackPE2204-30Aero(cid:31)ex8800 SMF2-12PasternackPE8302LNF LNC4-8A+40 dBAgilent11713B(0-120 dB)MiteqAFS3-0400800-07-10P-4+35 dB-9 dB-29 dB-30 dBS2. MICROWAVE CIRCUIT ANALYSIS 7 For two-port components, as depicted in Fig.1b of the main text, the transmission matrices, or ABCD-matrices are of the form [S1]: Tseries = (cid:19) (cid:18) 1 Z 0 1 (cid:18) 1 0 (cid:19) , 1 Z 1 (cid:19) (cid:18) V1 I1 (cid:19) (cid:18) V2 I2 , Tshunt = = T . (S2) These matrices relate incoming voltages and currents of port 1 (V1, I1) to outgoing (V2, I2) from port 2, hence characterize transmission of the two-port network. From these matrices we can already understand the limiting behaviour Sij(ω) = ±δij stated in the main text. If one takes the limit wherein the off-diagonal element goes to zero, the component shows perfect transmission. And in the limit where the off-diagonal element diverges, all power is reflected. The minus sign for reflection of a shunt component is due to the reversed sign of I2 in T -matrices compared to scattering matrices (i.e. the current flows out of port 2). Using standard formulas we can rewrite T into explicit scattering parameters for each component as: Sseries ii = z z + 2 , Sseries ij = 2 2 + z , Sshunt ii = −1 2z + 1 , Sshunt ij = 2z 2z + 1 . (S3) Using these expressions the reflection coefficient of port i, Γi = Sii, can easily be calculated, where z is the normalized impedance z = Z/Z0 and z = 1/jZ0ωC (z = jωL/Z0) for the capacitor (inductor), with j the imaginary It is perhaps interesting to note that the 50 − 50 beam splitter condition (Γ = 0.5), for such a two-port unit. component embedded in a 50 Ω transmission line (TL) needs a 25 Ω impedance for a shunt and 100 Ω for a series component To derive formula 1 in the main text we need to determine the coupling quality factor for each of the four coupler types for a single-port λ/4-microwave resonator. A λ/4 resonator requires opposite boundary conditions on each end of the TL, hence the open type of couplers are connected to a shorted TL and vice versa for the short type of couplers. Following reference [S1], here we present the generalized case. First we consider the open types of terminations; the gap capacitor and series inductor. In this configuration we have the coupler in series with a shorted TL, which has an impedance of zT L(ω) = j tan(βl), with l the length of the TL and β = ω/vp, where vp is the phase velocity and j the imaginary unit. At resonance we have βl = π/2, and the input impedance diverges. Now the impedance of the coupled resonator seen from the feedline is zin = zc + zT L, with zc the impedance of the coupler, either 1/jZ0ωC or jωL/Z0. To perform a Taylor expansion around the resonance, where zT L diverges it is convenient to rewrite it in terms of the susceptance of the coupler bc = Im(1/zc) and the cotangent. Note that for capacitors zc = −j/bc and for inductors zc = j/bc, which we abbreviate as ± (upper sign corresponds to an inductive element and the lower sign for a capacitive element). Now we can write the input impedance as follows: (cid:18) (cid:18) bc ∓ cot(βl) ± 1 bc + 1 (cid:19) cot(βl) (cid:19) , , zin(ω) = j = j bccot(βl) 1 + cot2(βl) (ω − ω0) ω0 , cot2(βl) 1 + b2 c (ω − ω0) b2 ω0 c (ω − ω0) , ω0 1 b2 c (cid:39) ±j π 2 (cid:39) ±j (cid:39) ±j π 2 π 2 at resonance: bc ± cot(βl) = 0, (I) , where we Taylor expanded around βl = (cid:39) π/2, ωl vp where we used (I), if bc (cid:28) 1. The last approximation breaks down for very large coupling susceptances. Remark that we used the trigonometric identity 1 + cot2 x = 1/ sin2 x to rewrite dz/d(βl) = ±j/ cos2(βl) = ±j/(cot2(βl) sin2(βl)). In this way we could invoke the resonance condition. For a low loss resonator we can include internal losses by substitution of ω0 → ω0(1 ∓ j/(2Qint)) in the numerator and obtain: zin(ω) = π 4Qintb2 c ± j π 2 ω − ω0 ω0b2 c . (S4) This expression of the input impedance is equivalent to that of a series LC-resonator, with a resistance on resonance of R = Z0π/(4Qintb2 c). At critical coupling the resonator on resonance is impedance matched to the feedline, and 8 Qint equals Qc, which is captured with the coupling factor g = Z0/R = Qint/Qc. Now using Qc = QintR/Z0 we obtain the result: Qc = π 4 1 b2 c . (S5) For the shunt couplers the derivation is very similar. Now we consider a shunt inductor (Im(zc) = 1/bc = ωL/Z0) or a shunt capacitor (Im(zc) = 1/bc = 1/(Z0ωC) coupled to an open ended λ/4 TL, with zT L = −j cot(βl). Now we have a parallel circuit so we obtain: 1 zin(ω) = j(∓bc + 1 cot(βl) ). (S6) We see that if we replace bc → 1/bc and reverse the sign we obtain the same expression, though in admittance yin = 1/zin. This shows that such coupled cavities behave as parallel LC-resonators. Following the same logic we obtain: yin(ω) = πb2 c 4Qint ∓ j π 2 ω − ω0 ω0b2 c , (S7) with (+) for capacitors and (-) for inductors respectively. So we conclude that for the shunt couplers we obtain: (S8) In the derivation for the series couplers we used bc (cid:28) 1, since for example a 10 fF gap capacitor in a 5 GHz cavity of 50 Ohms is bc ∼ 0.02, making the approximation valid. In the case of the shunt capacitor the reasoning goes exactly c (cid:28) 1, in the case of 30 pF for the same cavity we obtain bc ∼ 50, justifying this in opposite direction, where b−1 approximation. Qc = π 4 b2 c. S3. MICROWAVE RESPONSE With the expression for the input impedance at hand, it is easy to derive the microwave response of the cavity in reflection. Each coupling port contributes to the total loss rate κtot and causes a small shift of the resonance frequency, which is here not of our interest. In our implemented microwave cavity, as depicted in Fig. 1c and Fig. 2 of the main text, we can distinguish three separate loss channels that contribute to the total linewidth of the cavity: κtot = κ1 + κ2 + κint. A reflection measurement on port i is sensitive to the ratio between the total linewidth and coupling rate to port i, captured with the coupling coefficient: ηi = κi/κtot. Therefore if we absorb the coupling rate of the other port into κint we can suffice using the previously derived expressions. The reflection of the effective single-port cavity can then be written as: S11(ω) = zin − 1 zin + 1 = 1 − 2 1 + zin . Using that ω0/κ1 = Qc1 and similar for κint we can rewrite z(ω) (formula S3) as follows: zin(ω) = − 2j(ω − ω0) κ1 . κint κ1 Combining both expressions and rearranging the κ’s we obtain: S11(ω) = 1 − 2η1 1 − 2j(ω − ω0)/κtot (S9) (S10) (S11) The reflection response on port 2 (S22) is very similar, with the only difference that it was written in admittance. Here we will use zin as the input impedance for port 2: in − 1 y−1 y−1 in + 1 zin − 1 zin + 1 S22(ω) = = Following the same logic as before we obtain: 1 − yin 1 + yin = = −1 + 2 1 + yin . (S12) S22(ω) = −1 + 2η2 1 + 2j(ω − ω0)/κtot . 9 (S13) For deriving the transmission response S21 = S12 one cannot use this trick and the cavity should be analysed using transfer matrices. By multiplying the T -matrices of the gap-coupler, the TL and shunt coupler, one arrives at the T -matrix for the full cavity, from which an expression for S12 can be determined. If we abbreviate b1 = Z0ωCgap and b2 = Z0ωCshunt, and follow a similar procedure as before we obtain: S12(ω) = This expression can then be simplified to: 2 b1 b2 j(b2 1 + b−2 2 ) + 2(ω − ω0)/(4ω0/π) S12(ω) = S21(ω) = j + 2(ω − ω0)/κtot , √ 2 η1η2 . (S14) (S15) and we have obtained all response functions. For reflection of the gap capacitor (S11, formula S10) with a large detuning from the resonance frequency, the response is anchored at the point (1, 0) in the complex plane, which shows that the reflection on a (quasi)-open does not cause a phase shift of the signal. The response of the shunt capacitor, described by formula S12, is anchored at (−1, 0), showing the π phase shift on a (quasi)-short boundary condition. Transmission, S21 = S12, for a large detuning is anchored at the origin, indicating there is no transmission of very off-resonant signals. Close to resonance the term 2j(ω − ω0) becomes important and the response will trace a (resonance) circle in the complex plane, where the radius depends on the coupling coefficient ηi. For an over-coupled port (ηi > 0.5) the resonance circle will enclose the origin and the phase angle will change by 2π, whereas the phase angle of an under- coupled port will remain on the same branch. We would like to make a cautionary note on interpretting phase data without inquiring the resonance circle, as the ‘unwrapping’ of different phase branches is arbitrary and can lead to erroneous conclusions. At resonance the magnitude of the reflection response will show a minimum (dip) of magnitude Sii(ω0) = 1 − 2ηi. After determining whether the port is under- or over-coupled, ηi can be found by measuring the depth of the resonance dip. To illustrate the differences in the microwave response we show the results of a simulation in QUCs (an open-source circuit simulator [S2]) in Fig. S2. 10 FIG. S6. Numeric S-parameter simulation in QUCs of a cavity with a gap capacitance of 5 fF and a shunt capacitance of 30 pF (not the same values as in the experiment) illustrating the differences in microwave response; (left S11 under coupled and √ anchored at (1,0), middle S22 over coupled and anchored at (-1,0), and right shows S12 anchored in (0,0) and its magnitude η1η2 and hence is sensitive to the coupling imbalance κ1/κ2 as well as to the magnitude of internal losses κint is equal to (which is sometimes omitted). S4. DATA ANALYSIS Here we demonstrate the data analysis with the reflection data of port 2. Experimental realities like cable resonances and other uncontrolled factors in the circuitry can change the response from a Lorentzian line shape to a skewed- Lorentzian or Fano-line shape. This can be captured by the following adaptation of formula S12: (cid:18) (cid:19) S22(ω) = αeiφ + (1 − α) − 1 + 2η2 1 + 2j(ω − ω0)/κtot) , (S16) where A, α and φ account for an offset of the anchor of the resonance circle. Taking these effects correctly into P1Num=1Z=50 OhmP2Num=2Z=50 OhmCL1Subst=Subst1W=10 umS=10 umL=5 mmC1C=30 pFSubst1er=13h=0.5mmt=0.1 umtand=5e-6rho=0.0D=0S parametersimulationSP1Type=linStart=5.5GStop=6GPoints=5000C2C=5 fF5.5e95.6e95.7e95.8e95.9e96e900.51frequencyS[2,1]5.5e95.6e95.7e95.8e95.9e96e90.60.81frequencyS[1,1]5.5e95.6e95.7e95.8e95.9e96e9-20020frequencyphaseS115.5e95.6e95.7e95.8e95.9e96e90.60.81frequencyrealS115.5e95.6e95.7e95.8e95.9e96e9-0.200.2frequencyimagS115.5e95.6e95.7e95.8e95.9e96e90.60.81frequencyS[2,2]5.5e95.6e95.7e95.8e95.9e96e9-2000200frequencyphaseS225.5e95.6e95.7e95.8e95.9e96e9-10frequencyrealS225.5e95.6e95.7e95.8e95.9e96e9-101frequencyimagS225.5e95.6e95.7e95.8e95.9e96e9-1-0.50frequencyimagS215.5e95.6e95.7e95.8e95.9e96e9-0.500.5frequencyrealS215.5e95.6e95.7e95.8e95.9e96e9-200-1000frequencyphaseS210.51frequencyS[1,1]0.51frequencyS[2,2]0.51frequencyS[1,2]11 FIG. S7. Data analysis (a) Schematic of the loss and coupling rates. (b) Reflection of port 2 including the experimental data and fit. (c) Resonance circle plot of the S22 response showing data and fit, demonstrating the response is anchored at (-1,0) in the IQ-plane. The fitted resonance frequency and FWHM points are indicated with enlarged blue points. account is non-trivial and described for example in [S3,S4]. As the transmission response is insensitive to such effects, we first determine the total linewidth by fitting S12 to formula S14 and obtain κtot = 3.49 MHz. Subsequently we can fit the reflection data as shown in Fig. S3 for S22 and determine the coupling rates. From these fits we obtain κ1 = 0.51 MHz (η1 = 0.147) and κ2 = 2.75 MHz (η2 = 0.79). Using κtot = κ1 + κ2 + κint, we extract the internal loss rate as κint = 230 kHz, corresponding to a Qint ∼ 23.4× 103. S5. SHUNT CAPACITOR SIMULATIONS In this section we explore the implications of using the shunt-capacitor coupling architecture on the (potential) performance of a microwave resonator. First we discuss the simulation setup used, and follow with discussing the self-resonances of the shunt capacitor, losses incurred by the dielectric of the capacitor and maximum obtainable Q-factors. 12 S5.1. Sonnet simulation setup Sonnet is simulation software specialized in planar microwave circuits. Both the full cavity and the shunt capacitor are simulated separately, as depicted in Fig. S4. To allow simulations on a 10× 2.5 µm grid we implemented the input coupler as an ideal capacitor of 10 fF. We used a kinetic (sheet) inductance of Lk = 0.78 pH/(cid:3) for the Molybdenum- Rhenium layers based on earlier measurements for similar films [S5]. For the SiN brick we used
r = 7.4, and defined 3 degrees of freedom per lattice point in the z-direction. FIG. S8. Sonnet simulation setup: (a) 3D view of the shunt capacitor showing the substrate, metallization and the SiN dielectric brick defined inside a vacuum layer. (b) Top view of the shunt capacitor simulation. (c) Top view of the simulation configuration where the full bias cavity is simulated. The inset on the left hand side shows the implementation of an ideal input capacitor of 10 fF, such that a much lower resolution could be chosen. S5.2. Low frequency response Here we simulate the low frequency response of the shunt capacitor as a function of thickness. In this configuration it can be considered as a single-pole low-pass filter, which enables us to determine the capacitance from its cut- off frequency. For applications it can be used to determine the required DC-filtering to protect potential circuits against low frequency noise. Fig. S5a shows that by decreasing the thickness t of the dielectric the transmission for high frequencies is more suppressed, which is expected as Cshunt ∝ t−1 in the parallel plate approximation. More instructive is to consider the transmission on a logarithmic frequency scale, as depicted in Fig. S5b, known as a Bode plot. We observe the slope to be close to 20 dB/octave, characteristic of a single-pole low pass filter. Using a RC-circuit model we can determine the capacitance from the cut-off frequency using C = 1/(R 2πfc), where R = 25 Ω because both 50 Ω ports are in parallel. For t = 100 nm we determine a cut-off frequency of 255 MHz, corresponding to 25 pF of capacitance, close to the 27 pF we expect from a parallel plate approximation. 13 FIG. S9. Simulated low frequency response of the shunt capacitor, while varying the Si3N4 dielectric thickness. (a) Transmis- sion response (S21) as a function of frequency. (b) Transmission (S21) on a logarithmic frequency scale. We determine the slope as 19.6 dB/octave. For a t = 10 nm we see that the stray inductance causes deviation from this ideal filter model. The inset shows the cut-off frequency (−3 dB-point) as a function of thickness t. S5.3. Self-resonance frequency In the previous section we observed that for a 10 nm dielectric a self-resonance of the shunt capacitor due to stray inductance appeared around ∼ 8 GHz. These self-resonances place an upper bound to achievable Qc’s, because the coupling quality-factor increases as the total capacitance increases. To determine the maximum obtainable coupling Q for a 50 Ω cavity and a shunt capacitor of this geometry, we study these self-resonances as shown in Fig. S6. The first resonance drops from ∼ 25 GHz for a 100 nm thickness, to ∼ 8 GHz for 10 nm thickness. Thus for circuits requiring coupling Q’s in access of 3 · 105, corresponding t = 10 nm, one needs to take measures to prevent these resonances to interfere with the circuit, like increasing the impedance of the resonator or introducing structures in the shunt capacitor that prevent box modes (like via’s in a pcb). FIG. S10. Simulated mi- crowave transmission, while varying the Si3N4 dielec- tric thickness from 100 nm to 10 nm in steps of 10 nm. (a) Transmission re- sponse (S21) of just the shunt capacitor showing self- resonances of the shunt ca- pacitor (dashed grey lines). Each subsequent curve has an offset of 50 dB. (b) Trans- mission response for the com- plete bias cavity showing the first 5 cavity resonances and the self-resonances of the shunt capacitor. Clearly changing the dielectric thick- ness does not have a no- ticeable effect on the cav- ity resonances, but does pull down the self-resonances of the shunt capacitor. 14 S5.4. Dielectric losses Deposited dielectrics are well known to be lossy with loss tangents varying from tan δ ∼ 10−3 − 10−5. Here we consider this contribution to the losses of a cavity using a shunt capacitor as coupler. The dielectric loss tangent is specified as tan δ =
2/
1, with
=
1 + j
2, where
2 denotes the dissipative part of the dielectric constant. For a lossy parallel plate capacitor, that we showed in previous sections to be a valid approximation, we can use C =
A/d = C(1 + j tan δ). As a circuit this can be modelled by including an equivalent series resistor (Resr), capturing the losses in the dielectric, where Resr = tan δ/(ωC), as depicted in Fig. S7a. Now we consider a single-port λ/4 cavity, as displayed in Fig. S7b, and simulate it in QUCs [S2] for various coupling capacitances and loss tangents. From the simulated scattering data we can extract the coupling rate to port 2, κ2 as a function of dielectric thickness, showed by the dashed line in Fig. S7. As expected κ2 decreases from about 2 MHz to 20 kHz by decreasing the dielectric thickness. (a) in the FIG. S11. Lumped element simulation of losses shunt capacitor. In- corporating dielectric losses in a capacitor. (b) Lumped element model of single port cavity showing the loss processes. (c) Loss rates as a function of thickness of the capac- itor for various dielec- tric loss tangents. The coupling rate κ2 is indi- cated with the dashed line, the internal losses due to the dielectric losses in the shunt ca- pacitor are shown by the solid lines. loss int in the Dielectric FIG. S12. shunt losses capacitor. (a) Fig- ure showing the dielec- tric losses as a func- tion of tangent for various thicknesses. In the inset we see that the losses show a linear relation on a log-log plot indicating a monomial relation- ∝ t · ship as κshunt δ. (b) Figure showing the maximum obtain- able loaded quality fac- tor for different δ and thicknesses. In this plot other loss factors are ignored and just the coupling rate and dielectric losses in the capacitor are included. The losses incurred by the lossy dielectric capacitor can be isolated by comparing κint for different δ’s to the lossless case, which is shown in Fig. S7c as the solid lines. These incurred losses decrease as the coupling capacitance increases, 15 because the voltage across the capacitor plates decreases and hence the dielectric losses decrease as well. Assuming a int ∼ 8−80 loss tangent of around δ ∼ 10−3−10−4 we estimate these losses in our experiment to be on the order of κshunt kHz. Finally from this figure we see that for thin dielectrics and realistic δ’s the incurred losses can be minimized to sub-kHz level. Finally we can use this data to determine what the maximum possible loaded quality factor for this resonator architecture is, shown in Fig. S8. By including both the coupling rate and dielectric losses of the capacitor, we see that the loaded quality factor is limited to Ql (cid:46) 3 × 105 for a dielectric thickness of 10 nm. Such a loaded quality factor corresponds to linewidth’s in the range of κtot (cid:38) 20 kHz, sufficiently narrow for most applications. Beyond this regime a different geometry or a higher impedance resonator is required. [S1] D. M. Pozar, Microwave engineering (John Wiley & Sons, 2009) [S2] M. Brinson and S. Jahn, International Journal of Numerical Modelling: Electronic Networks, Device and Fields 22, 297 (2009) [S3] M. Khalil, M. Stoutimore, F. Wellstood, and K. Osborn, Journal of Applied Physics 111, 054510 (2012). [S4] S. Probst, F. Song, P. Bushev, A. Ustinov, and M. Weides, Review of Scientific Instruments 86, 024706 (2015). [S5] V. Singh, B. H. Schneider, S. J. Bosman, E. P. Merkx, and G. A. Steele, Applied Physics Letterrs 105, 222601 (2014). 

1305.0942

2

1305

Effect of perpendicular uniaxial anisotropy on the annihilation fields of magnetic vortices

The magnetic vortex structure, that is present in several nanoscopic systems, is stable and can be manipulated through the application of a magnetic field or a spin polarized current. The size and shape of the core are strongly affected by the anisotropy, however its role on the core behavior has not yet been clarified. In the present work we investigate the influence of a perpendicular anisotropy on the annihilation and shape of magnetic vortex cores in permalloy disks. We have used both micromagnetic simulations with the OOMMF code, and an analytical model that assumes that the shape of the core does not change during the hysteresis cycle, known as the rigid core model, to calculate the annihilation fields. In both cases we found that the annihilation fields decrease with increasing perpendicular anisotropy for almost all the structures investigated. The simulations show that for increasing anisotropy or dot thickness, or both, the vortex core profile changes its shape, becoming elongated. For every dot thickness, this change does not depend on the dot radius, but on the relative distance of the core from the center of the dot.

cond-mat.mes-hall

cond-mat

Effect of perpendicular uniaxial anisotropy on the annihilation fields of magnetic vortices E.R.P. Novais,1 S. Allende,2, 3 D. Altbir,3 P. Landeros,4 F. Garcia,5 and A.P. Guimaraes 1)Centro Brasileiro de Pesquisas F´ısicas, 22290-180, Rio de Janeiro, RJ, Brazil 2)Departamento de Ciencias F´ısicas, Universidad Andr´es Bello, Avenida Rep´ublica 220, 837-0134, Santiago, Chile 3)Departamento de F´ısica, Universidad de Santiago de Chile and CEDENNA, Avda. Ecuador 3493, Santiago, Chile 4)Departamento de F´ısica, Universidad T´ecnica Federico Santa Mar´ıa, Avenida Espana 1680, Valpara´ıso, Chile 5)Laborat´orio Nacional de Luz S´ıncrotron, 13083-970, Campinas, SP, Brazila) (Dated: 31 August 2018) The magnetic vortex structure, that is present in several nanoscopic systems, is stable and can be manipulated through the application of a magnetic field or a spin polarized current. The size and shape of the core are strongly affected by the anisotropy, however, its role on the core behavior has not yet been clarified. In the present work we investigate the influence of a perpendicular anisotropy on the annihilation and shape of magnetic vortex cores in permalloy disks. We have used both micromagnetic simulations with the OOMMF code, and an analytical model that assumes that the shape of the core does not change during the hysteresis cycle, known as the rigid core model, to calculate the annihilation fields. In both cases we found that the annihilation fields decrease with increasing perpendicular anisotropy for almost all the structures investigated. The simulations show that for increasing anisotropy or dot thickness, or both, the vortex core profile changes its shape, becoming elongated. For every dot thickness, this change does not depend on the dot radius, but on the relative distance of the core from the center of the dot. I. INTRODUCTION Among the nano- and mesoscopic magnetic structures that have attracted the attention of researchers in re- cent years stand out those that exhibit a vortex, since this state presents both interesting physical properties and a high potential for applications.1 -- 6 Magnetic vortex states in nanodots are characterized by in-plane magnetic moments curling around a core which has magnetization pointing out-of-plane. Two main features are defined in a vortex, the circulation, i.e., the sense of the magne- tization curling, being −1 (+1) for clockwise (counter- clockwise) rotation direction, and the polarity defined by the direction of the core magnetization denoted by p = +1 (−1) for upward (downward) direction. The core profile mz(r) (the z component of the unit magnetiza- tion) of a vortex in equilibrium is cylindrically symmetric, usually approximated by a Gaussian curve surrounded by a small dip (see Fig. 1).1,7 A vortex configuration is the ground state of different nanodots with regular shape such as ellipses, squares, spheres, caps and disks, with lateral dimensions rang- ing from one hundred nanometers to a few microns, with some tens of nanometers thickness.8 -- 14 While an external in-plane magnetic field that increases continuously from zero is applied to a disk exhibiting a magnetic vortex, its core will be displaced perpendicularly to the field di- rection, until its center reaches the disk edge. The field a)Present address: Centro Brasileiro de Pesquisas F´ısicas, 22290- 180, Rio de Janeiro, RJ, Brazil corresponding to this limiting situation, i.e., a field that expels the vortex core, is known as the annihilation field. A further field increase will expel the vortex from the disk, and the saturated state will eventually be reached. On the other hand, when starting from a fully saturated state, by decreasing the field to a certain critical value (commonly referred in the literature as the nucleation field) the vortex will again be formed. The knowledge and control of the magnitude of these fields is a key issue for several applications considering the manipulation of magnetic vortices, such as non-volatile magnetic memory devices, or high-resolution magnetic field sensors.15 -- 17 The vortex-core nucleation and annihilation processes have been discussed by several authors18 -- 23 and, in par- ticular, the influence of extrinsic properties on the anni- hilation field has been taken into account. Wu et al.24 investigated the role of geometrical asymmetries, find- ing that the annihilation of the vortex depends strongly on the asymmetry. The effect of the shape asymmetry has also been studied by Dumas et al.25, by measuring the angular dependence of the annihilation field. Miha- jlovi´c et al.26 have shown that temperature also affects the reversal mechanism and the vortex annihilation field, while experiments by Davis et al.27 suggest that the nu- cleation and annihilation fields depend on the magnetic field sweep rate. This problem was also examined from the theoretical point of view, within the framework proposed by Gus- lienko et al.20 This model approximates the core as a magnetization distribution whose profile does not change during the reversal process. Some important properties of the vortices, such as the core size and some dynamic features, can be tailored in- 3 1 0 2 y a M 7 ] l l a h - s e m . t a m - d n o c [ 2 v 2 4 9 0 . 5 0 3 1 : v i X r a 2 troducing a uniaxial perpendicular magnetic anisotropy, as has been recently shown17,28 In this case, as the perpendicular anisotropy increases, important deviations from the vortex core profile and from the canonical mag- netic vortex configuration result. Beyond a critical value of the anisotropy (K crit it is no longer observed a vortex, with the formation of a skyrmion (e.g., Fert et al.29), a structure that was found in experiments with BFeCoSi30 and, more relevant to the present study, was also apparent in experiments with Co/Pt disks17 and simulations.14,17 ), z Vortex core deformations, even under the action of a magnetic field, have not been so far systematically ana- lyzed. The aim of this paper is to get a better understand- ing of the vortex annihilation process in magnetic dots. For this, we have compared the description using the rigid vortex model with results obtained by micromag- netic simulations. In order to explore the effect of the perpendicular anisotropy on the vortex core properties, and concomitantly, verify the limits of validity of the rigid vortex model, we have introduced an anisotropic term in both the theory and simulations. We also characterized the vortex core deformations that are present in some simulations. The paper is organized as follows: after the Introduc- tion we describe how we perform our micromagnetic sim- ulations that lead us to study the annihilation fields ex- tracted from the hysteresis curves of disks with various sizes (Sec. II). Analytical calculations are presented in Sec. III, with the inclusion of anisotropy terms into the rigid vortex model. The results are contained in Sec. IV, and finally, in Sec. V we summarize and draw conclu- sions. II. NUMERICAL SIMULATIONS We investigated the hysteresis loops of individual mag- netic nanodots defined by their thickness L varying be- tween 10 and 30 nm, and diameters D from 100 to 1000 nm, while the uniaxial perpendicular anisotropy Kz ranges from 0 to 300 kJ/m3. This study was conducted through micromagnetic simulations14,31 us- ing the OOMMF code.32 We used a stiffness constant A = 13 × 10−12 J/m and a saturation magnetization Ms = 860 × 103 A/m, the standard values used for bulk permalloy, taking a cell size of 5 × 5 × 5 nm3. The maxi- mum anisotropies used in this work that keep the vortex structure are K max = 300, 225 and 165 kJ/m3 for the thicknesses L = 10, 20 and 30 nm, respectively. z For larger anisotropies, a skyrmion structure is ob- served, and perpendicular magnetization appears on the rim of the disk. For this reason, in all our calculations the anisotropy constant value was chosen such that the magnetic configuration at zero external applied field is a vortex configuration, as shown in Fig. 1. In our simulations we developed a systematic study of FIG. 1. (Color Online) a) profile of the vortex core corre- sponding to disks with D = 500 nm and L = 10 nm for Kz = 0 and Kz = 300 kJ/m3; b) and c) depict the magne- tization for Kz = 0 and Kz = 300 kJ/m3 with L = 10 nm. d) profile of the vortex core with D = 500 nm and L = 30 nm for Kz = 0 and Kz = 165 kJ/m3. e) and f ) represent the magnetization for Kz = 0 and Kz = 165 kJ/m3 for L = 30 nm. Note that from a) to d) the depth of the negative part of the magnetization (the dip) increases. the annihilation field that is determined from the maxi- mum of the derivative dM/dB in the increasing magne- tization branch of the hysteresis loop, which corresponds to the expulsion of the vortex core. All hysteresis curves were obtained starting from the unperturbed configura- tion of the disks (with the vortex core at the center), in- creasing the field from B = 0, in steps of ∆B = 0.1 mT, leading us to obtain the annihilation field, and finally reaching the magnetic saturation. In some simulations we observed a deformation of the vortex core. In order to characterize it we define δ = ry − rx rx , (1) where rx and ry are the sizes of the vortex core along the x and y axes, respectively. As shown in Fig. 1, two orthogonal sections (x and y directions) of the profiles of the vortex core passing through the core center (max- imum of mz) were made. The dimensions of the core along the x and y directions were obtained by the full widths at half maximum of the repective profile fit, us- ing a pseudo-Voigt function. III. ANALYTICAL MODEL To obtain analytical expressions for the annihilation field in the magnetic nanodots we started with a model proposed by Guslienko et al.20,21 to investigate the vortex behavior in submicron dots. These authors considered a ferromagnetic dot with a height L and a radius R that and ρm = x cos φ +pR2 − x2 + x2 cos2 φ . Using these expressions we can write Eq. 5 as 3 (7) (8) (9) 0 2bx m2 m2 0 0 zρdρ# dφ zρdρ(cid:21) dφ WK = −KzLb2 sec−1(cid:20) WK = −2KzL φm −2KzL π " b φm(cid:20) ρm b2 − R2 + x2(cid:21) (3 − 2 ln 4) − G . z = (cid:16)1 − 4b2ρ2/(cid:0)b2 + ρ2(cid:1)2(cid:17) and (b2 + ρ2)2! ρ dρ# dφ . (10) 0 1 − When φm (x → R) ≈ π/2 or c = b/R ≪ 1, G can be approximated to zero at first order of (R − x). How- ever, in our calculations we considered it explicitly. If the anisotropy energy is normalized to M 2 s V , that is, wK = WK/(M 2 s V ), and using s = x/R, c = b/R and V = πR2L, we obtain In this expression m2 G represents the contributions to the anisotropy energy shown in the dark regions in Fig. 2a G = 2KzL π φm" ρm 4b2ρ2 wK (s) = −Kzc2 πM 2 s sec−1(cid:20) 2cs c2 − 1 + s2(cid:21) (3 − 2 ln 4) −g (s) , (11) where g (s) = G/M 2 s V . We proceed by minimizing the magnetic anisotropy energy with respect to s and eval- uating in the equilibrium displacement where the vortex center reaches the dot perimeter. In other words, differ- entiating Eq. 11 with respect to s and taking the limit s → 1, we obtain the value of the contribution of the anisotropy to the annihilation field = − Kz M 2 s c(cid:0)c2 − 2(cid:1) (ln 16 − 3) π√4 − c2 − lim s→1 ∂wK (s) hK = lim s→1 ∂s ∂g (s) ∂s . (12) In this way, and adding this expression to the annihi- lation field given by Eq. 3, we obtain the annihilation field for a nanodot with perpendicular anisotropy FIG. 2. Geometrical relation between the vortex core, defined by the dotted line, and the full dot. a) Illustration of the angle φm that depends on the radius of the dot, R, radius of the core, b, and separation between the centers of the dot and core, x. b) Representation of ρm, that depends on R, x, and the angle φ between x and ρm. presents a vortex state with a distribution of the unit magnetization in cylindrical coordinates ρ, ϕ, z given by ~m = sin θ (ρ) φ + cos θ (ρ) z, where21 mφ = sin θ (ρ) =(cid:26) (2bρ/(cid:0)b2 + ρ2(cid:1)) ρ ≤ b ρ ≥ b 1 . (2) Here b is the radius of the core. If we consider mag- netostatic, exchange, and Zeeman contributions to the energy, the normalized dimensionless vortex annihilation field in the rigid core model proposed by Guslienko et al.21 is written as R (cid:19)2 han (β, R) = 4πF1 (β) −(cid:18) R0 , (3) where β = L/R, R0 is the exchange length and F1 (β) is given by F1 (β, R) = ∞ 0 (cid:18)1 − 1 − e−βt βt (cid:19) J 2 1 (t) dt t . (4) A. Introducing a Perpendicular Uniaxial Anisotropy While the model proposed by Guslienko et al.20,21 con- tains no anisotropy, in our calculations we include a uni- axial anisotropy along the z axis and focus on its effect on the annihilation field. We start calculating the anisotropy energy contribution of the system that is given by WK = −LKz ( ~m · z)2 ρdφdρ , (5) where Kz > 0 is the anisotropy constant and z is the easy axis. From this expression, the contribution to the energy due to the anisotropy comes only from the core region inside the dot. From Fig. 2 we obtain φm = arccos(cid:18) x2 + b2 − R2 2xb (cid:19) , (6) Kz M 2 s − R (cid:19)2 han (β, R) = 4πF1 (β) −(cid:18) R0 π√4 − c2 c(cid:0)c2 − 2(cid:1) (−3 + ln 16) − lim ∂g (s) ∂s s→1 . (13) IV. RESULTS AND DISCUSSION A. Annihilation fields 4 Our analyses for the annihilation field are based on a theoretical approach and on micromagnetic simula- tions. From both approaches we have obtained the de- pendence with the perpendicular anisotropy of the anni- hilation fields; from the simulations the core diameters were also obtained. We present below the results for the two cases: IV A annihilation field, and IV B evolution of the magnetic core shape. We have observed that the analytical calculations result in larger annihilation fields as compared with the numerical simulations, however, there is a qualitative agreement between the results from both models. A qualitative agreement between theretical results and measured annihilation fields was previously reported.33 a) L= 10 nm 50 ) T m l i ( d e F n o i t a l i i h n n A 40 30 Micromagnetic simulations b) L= 10 nm 20 60 50 40 30 ) T m l i ( d e F n o i t a l i i h n n A Kz= 0 kJ/m3 Kz= 100 kJ/m3 Kz= 200 kJ/m3 Kz= 250 kJ/m3 Kz= 300 kJ/m3 Kz= 0 kJ/m3 Kz= 100 kJ/m3 Kz= 200 kJ/m3 Kz= 250 kJ/m3 Kz= 300 kJ/m3 20 Analytical method 200 400 600 Diameter (nm) 800 1000 FIG. 3. (Color Online) Annihilation fields versus diameter for dots a) obtained by micromagnetic simulation, and b) obtained by analytical method. These graphs show the in- fluence of the value of the perpendicular anisotropy on the annihilation field for different diameters. The values of the annihilation fields Ban as a function of disk diameter, for different anisotropies Kz, obtained from both methods are given in Figs. 3, 4 and 5 for L = 10, 20 and 30 nm, respectively. The two sets of results agree in the fact that, as the diameters of the disks increase, the values become less dependent on the anisotropy (Figs. 3, 4 and 5). a) L= 20 nm 90 80 70 60 50 ) T m l i ( d e F n o i t a l i i h n n A 40 110 100 Micromagnetic simulations b) L= 20 nm Kz= 0 kJ/m3 Kz= 100 kJ/m3 Kz= 150 kJ/m3 Kz= 200 kJ/m3 Kz= 225 kJ/m3 Kz= 0 kJ/m3 Kz= 100 kJ/m3 Kz= 150 kJ/m3 Kz= 200 kJ/m3 Kz= 225 kJ/m3 ) T m l i ( d e F n o i t a l i i h n n A 90 80 70 60 50 40 30 Analytical method 200 600 400 Diameter (nm) 800 1000 FIG. 4. (Color Online) Annihilation fields versus diameter for dots a) obtained by numerical calculation and b) obtained by the analytical method. Increasing the anisotropy, which results in larger core sizes, one is led to lower annihilation fields. For small diameters, the effect of the anisotropy is more noticeable than for the large disks, both in the theory and simulations. In the simulations, the sensitivity of the annihilation field to changes in anisotropy increases with the value of Kz. The disagreement between the two methods can be related to a deformation of the core observed in the mi- cromagnetic simulations.  a) L= 30 nm 120 110 Micromagnetic simulations b) L= 30 nm ) T m l i ( d e F n o i t a l i i h n n A 100 90 80 70 60 50 140 120 100 80 60 40 ) T m l i ( d e F n o i t a l i i h n n A Kz= 0 kJ/m3 Kz= 50 kJ/m3 Kz= 100 kJ/m3 Kz= 150 kJ/m3 Kz= 165 kJ/m3 Kz= 0 kJ/m3 Kz= 50 kJ/m3 Kz= 100 kJ/m3 Kz= 150 kJ/m3 Kz= 165 kJ/m3 5 shape of the vortex core is very clear. Comparing Fig. 1 and Figs. 6 and 7, we observe that the vortex core deformation increases as the core moves away from the center. The deformation is zero (i.e., the core is circular) for B = 0 and maximum when the vortex core is on the edge of the disk, in the instant immediately before the vortex annihilation. To clarify this point we have drawn the magnetization profile of the core for fields close to the annihilation field for dots of L = 10 and 30 nm, as shown in Figs. 6a and 7a, respectively. The dark region (blue online) in these figures represents the core region. The profile of the vortex core at the center of the disk is shown in Fig. 1. The Figs. 1a, 1b and 1c show the effect of the anisotropy on a dot of diameter 500 nm with L = 10 nm, and Figs. 1d, 1e and 1f for diameter 500 nm, with L = 30 nm. For L = 10 nm the core keeps a nearly circular shape, the core deformation reaches about 10 % for zero anisotropy (Kz = 0), and for Kz = 300 kJ/m3 the defor- mation is around 30%. Comparing Figs. 1a, 1b and 1c with Fig. 6, it is evident the vortex core deformation in the latter, due to an increase of its size along the y axis and the depth of the magnetization dip. However, for L = 30 nm, the smallest core deformation is 30%, that corresponds to anisotropy Kz = 0, and a larger core deformation is observed for Kz = 165 kJ/m3, of about 100%. Therefore, the core loses its circular shape for large anisotropy constant values Kz, leading to a core of roughly elliptical section that is not well de- scribed by the rigid vortex model. This is shown in Figs. 6 and 7, where it is more evident the deformation of the vortex core, as well as the variation in the magnetiza- tion dip. It is important to note that a higher anisotropy constant for L = 30 nm will result in a magnetic configu- ration that is not a vortex, as observed above. Therefore Kz = 165 kJ/m3 is the largest value that can be consid- ered in order to obtain a vortex state. We observe that the core deformation is not present for B = 0, as shown in Fig. 1. However, as the field begins to increase, the process of deformation of the core sets in, as can be seen in Figs. 8 and 9, plotted from values obtained using Eq. (1); they are shown as a function of relative core positions (pcore/R), defined as the ratio of the distance of the core center from the center of the disk (pcore) divided by the disk radius (R). Figures 8 and 9 show different stages of deformation. In the case where the vortex core is located below 25% of the radius of the disk, the deformation can be neglected. From this point onwards the deformation begins to in- crease, and the maximum is reached when the core ap- proaches the edge of the disk. Note in Figs. 8 and 9 that for each thickness (10, 20 and 30 nm) four curves were plotted for different diameters (400, 500, 750 and 1000 nm) and that these curves overlap; it appears to exist a scaling law for the deformation of the core, i.e., for each disk thickness, the deformation does not depend on the disk diameter, it depends only on the relative position of the vortex core. Analytical method 200 400 600 Diameter (nm) 800 1000 FIG. 5. (Color Online) Annihilation fields versus diameter for dots with a) and obtained by numerical calculation and b) obtained by the analytical method. B. Evolution of the magnetic core shape In the search of the effect that gives rise to the dis- agreement between the values of Ban derived by ana- lytical method and by numerical calculation, as we men- tioned above, we considered the possibility that this effect would also cause a deformation of the vortex cores. Our micromagnetic simulations show that this deformation is in fact present, as shown in Figs. 6 and 7. Of course, the rigid vortex model breaks down in the cases where important deformations of the vortex core are observed. We have therefore analyzed the evolution of the shape of the magnetic vortex core as it moves towards the edge of the disks, under the influence of an applied magnetic field; the effect of varying the value of a perpendicular anisotropy Kz was also investigated. The simulation re- sults show a gradual deformation of the vortex cores, that can be quantified; the vortex cores change from a cir- cular shape to a nearly elliptical ("banana-like") shape. Figs. 6 and 7 show images of the vortex core and the core profile for L = 10 nm and L = 30 nm; the change in 6 FIG. 6. (Color Online) Change of the shape of the vortex core for L = 10 nm, D = 500 nm and Kz = 300 kJ/m3 immediately before the annihilation. a) profile of the core along the x axis (red dotted line) and along the y axis (blue continuous line). b) representation of the disk. c) a detail of the disk section close to the vortex core. FIG. 7. (Color online) Change of the shape of the vortex core for L = 30 nm, D = 500 nm and Kz = 165 kJ/m3 at positions immediately before annihilation. In a) profile of the core along the x axis (red dotted line) and along the y axis (blue continuous line), in b) image of the disk and c) a detail of the disk. V. CONCLUSIONS In summary, by means of an analytical model and nu- merical simulations we have obtained the annihiliation fields for dots of different heights and anisotropy con- stants. In all cases, the annihilation fields decrease with increasing anisotropy constant Kz and with increasing disk diameter. The values of Kz and disk height L have a stronger effect on the annihilation fields of the smaller disks. However, the anisotropy and the thickness operate in an inverse way on the annihilation field; whereas the increase in height increases the annihilation field, from the analytical results using the rigid vortex model, the increase in anisotropy decreases this field. Finally, we have shown the variation in the deformation of the vor- tex core δ as a function of the perpendicular anisotropy and disk height. The occurrence of this deformation ev- idenced in the micromagnetic simulations suggests that it has to be taken into account in the description of the dynamics of the magnetic vortices. The deformation of the core does not scale with the radius of the disks, it is only related to the relative position of the core. ACKNOWLEDGMENTS In Brazil we acknowledge the support of the agencies CAPES, CNPq, FAPERJ and FAPESP. In Chile we ac- knowledge the partial support from FONDECYT under grants 11121214, 1120356 and 1120618, from the Center for the Development of Nanoscience and Nanotechnology, and from ICM P10-06-F funded by Fondo de Innovaci´on para la Competitividad, from the MINECON. L. and J.-G. Zhu, Chien, 1A. P. Guimaraes, Principles of Nanomagnetism (Springer, Berlin, 2009). 2C. Phys. Today 60, 40 (2007). 3K. Y. Guslienko, J. Nanoscience Nanotechnol. 8, 2745 (2008). 4S. Bohlens, B. Kruger, A. Drews, M. Bolte, G. Meier, D. Pfannkuche, Appl. Phys. Lett. 93, 142508 (2008). F. Q. Zhu, and  D=400, 500, 750 and 1000 nm L = 10 nm L = 20 nm L = 30 nm Kz = 0 D = 400 nm D = 500 nm D = 750 nm D = 1000 nm L=30 0.4 0.2 0.8 Relative core position 0.6 1.0 L=20 L=10 2 - 0 1 L / n o i t a m r o f e D 1.0 0.5 0.0 0.0 n o i t a m r o f e D 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 0.2 0.4 0.6 0.8 1.0 Relative core position FIG. 8. (Color online) Deformation δ = (ry − rx)/rx for Kz = 0, L = 10, 20, and 30 nm for diameters D = 400, 500, 750 and 1000 nm versus normalized core po- sition (pcore/R). Note that a scaling law is apparent. 2 - 0 1 X L / n o i t a m r o f e D 2.0 1.5 1.0 0.5 0.0 D = 400, 500, 750 and 1000 nm L= 10 nm L= 20 nm L= 30 nm L=30 Kz = 100 kJ/m3 D = 400 nm D = 500 nm D = 750 nm D = 1000 nm 0.0 0.4 0.2 0.8 Relative core position 0.6 1.0 L=20 L=10 n o i t a m r o f e D 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0.2 0.4 Relative core position 0.6 0.8 1.0 FIG. 9. (Color online) Deformation δ = (ry − rx)/rx for Kz = 100 kJ/m3, L = 10, 20, and 30 nm for diameters D = 400, 500, 750 and 1000 nm versus normalized core posi- tion (pcore/R). Note that a scaling law is apparent. 5A. Ruotolo, V. Cros, B. Georges, A. Dussaux, J. Grollier, C. De- ranlot, R. Guillemet, K. Bouzehouane, S. Fusil, and A. Fert, Nat. Nanotechnol. 4, 528 (2009). 6H. Jung, Y.-S. Choi, K.-S. Lee, D.-S. Han, Y.-S. Yu, M.- Y. Im, P. Fischer, and S.-K. Kim, ACS Nano 6, 3712 (2012), http://pubs.acs.org/doi/pdf/10.1021/nn3000143 . 7M. Bode, O. Pietzsch, A. Kubetzka, W. Wulfhekel, D. McGrouther, and J. N. Chapman, S. McVitie, 7 Phys. Rev. Lett. 100, 029703 (2008). 8P. Landeros, J. Escrig, D. Altbir, D. Laroze, J. d'Albuquerque e Castro, and P. Vargas, Phys. Rev. B 71, 094435 (2005). 9D. Altbir, J. Escrig, P. Landeros, F. S. Amaral, and M. Bahiana, Nanotechnology 18, 485707 (2007). 10W. Zhang, R. Singh, N. Bray-Ali, and S. Haas, Phys. Rev. B 77, 144428 (2008). 11K. L. Metlov and Y. Lee, Appl. Phys. Lett. 92, 112506 (2008). 12M. M. Soares, E. de Biasi, L. N. Coelho, M. C. dos Santos, F. S. de Menezes, M. Knobel, L. C. Sampaio, and F. Garcia, Phys. Rev. B 77, 224405 (2008). 13S.-H. Chung, R. D. McMichael, D. T. Pierce, and J. Unguris, Phys. Rev. B 81, 024410 (2010). 14E. R. P. Novais, P. Landeros, A. G. S. Barbosa, and A. P. Guimaraes, M. D. Martins, F. Garcia, J. Appl. Phys 110, 053917 (2011). 15M. Rahm, M. Schneider, J. Biberger, R. Pulwey, J. Zweck, D. Weiss, and V. Umansky, Appl. Phys. Lett. 82, 4110 (2003). and Y.-S. Choi, 16S.-K. Kim, K.-S. Lee, Y.-S. Yu, Appl. Phys. Lett. 92, 022509 (2008). 17F. Garcia, H. Westfahl, J. Schoenmaker, E. J. Carvalho, A. D. Santos, M. Pojar, A. C. Seabra, R. Belkhou, A. Bendounan, E. R. P. Novais, and A. P. Guimaraes, Appl. Phys. Lett. 97, 022501 (2010). Hoffmann, Schneider, Zweck, 18M. and H. J. Appl. Phys. Lett. 77, 2909 (2000). 19A. Fernandez and C. J. Cerjan, J. Appl. Phys 87, 1395 (2000). 20K. Metlov, Guslienko and K. Y. L. Phys. Rev. B 63, 100403 (2001). 21K. Y. Guslienko, V. Novosad, Y. Otani, H. Shima, and K. Fukamichi, Phys. Rev. B 65, 024414 (2001). 22J. Mejia-Lopez, D. Altbir, A. H. Romero, X. Batlle, Schuller, I. K. and I. V. Roshchin, J. Appl. Phys 100, 104319 (2006). C.-P. Li, 23J. Mej´ıa-L´opez, D. Altbir, P. Landeros, J. Escrig, A. H. Romero, I. V. Roshchin, C.-P. Li, M. R. Fitzsimmons, X. Batlle, and I. K. Schuller, Phys. Rev. B 81, 184417 (2010). 24K.-M. Wu, L. Horng, J.-F. Wang, J.-C. Wu, Y.-H. Wu, and C.-M. Lee, Appl. Phys. Lett. 92, 262507 (2008). 25R. K. Dumas, T. Gredig, C.-P. Li, I. K. Schuller, and K. Liu, Phys. Rev. B 80, 014416 (2009). 26G. Mihajlovi´c, M. S. Patrick, J. E. Pearson, V. Novosad, and A. Hoffmann, S. D. Bader, M. Field, G. J. Sullivan, Appl. Phys. Lett. 96, 112501 (2010). 27J. P. Davis, D. Vick, J. A. J. Burgess, D. C. Fortin, and M. R. Freeman, P. Li, V. Sauer, W. K. Hiebert, New J. Phys. 12, 093033 (2010). 28T. S. Machado, T. G. Rappoport, and L. C. Sampaio, Appl. Phys. Lett. 93, 112507 (2008). 29A. Fert, V. Cros, and J. Sampaio, Nat Nano 8, 152 (2013). 30X. Z. Yu, Y. Onose, N. Kanazawa, J. H. Park, and Y. Tokura, J. H. Han, Y. Matsui, N. Nagaosa, Nature 465 (2010), 10.1038/nature09124. 31S.-K. Kim, J. Phys. D: Appl. Phys 43, 264004 (2010). 32M. Donahue and D. Porter, "Object oriented micromagnetic framework," (OOMMF) (1999). 33V. Novosad, K. Y. Guslienko, H. Shima, Y. Otani, and Y. Shimada, K. Fukamichi, N. Kikuchi, O. Kitakami, Magnetics, IEEE Transactions on 37, 2088 (2001). 

1106.0407

1

1106

Transmission through a quantum dot molecule embedded in an Aharonov-Bohm interferometer

We study theoretically the transmission through a quantum dot molecule embedded in the arms of an Aharonov-Bohm four quantum dot ring threaded by a magnetic flux. The tunable molecular coupling provides a transmission pathway between the interferometer arms in addition to those along the arms. From a decomposition of the transmission in terms of contributions from paths, we show that antiresonances in the transmission arise from the interference of the self-energy along different paths and that application of a magnetic flux can produce the suppression of such antiresonances. The occurrence of a period of twice the quantum of flux arises to the opening of transmission pathway through the dot molecule. Two different connections of the device to the leads are considered and their spectra of conductance are compared as a function of the tunable parameters of the model.

cond-mat.mes-hall

cond-mat

Transmission through a quantum dot molecule embedded in an Aharonov-Bohm interferometer Daniel A. Lovey, Sergio S. Gomez, and Rodolfo H. Romero∗ Instituto de Modelado e Innovaci´on Tecnol´ogica, CONICET, and Facultad de Ciencias Exactas y Naturales y Agrimensura, Universidad Nacional del Nordeste, Avenida Libertad 5500 (3400) Corrientes, Argentina. (Dated: June 6, 2018) We study theoretically the transmission through a quantum dot molecule embedded in the arms of an Aharonov-Bohm four quantum dot ring threaded by a magnetic flux. The tunable molecular coupling provides a transmission pathway between the interferometer arms in addition to those along the arms. From a decomposition of the transmission in terms of contributions from paths, we show that antiresonances in the transmission arise from the interference of the self-energy along different paths and that application of a magnetic flux can produce the suppression of such antiresonances. The occurrence of a period of twice the quantum of flux arises to the opening of transmission pathway through the dot molecule. Two different connections of the device to the leads are considered and their spectra of conductance are compared as a function of the tunable parameters of the model. I. INTRODUCTION The reduction of size in the electronic devices to nanometer scale has highlighted the importance of the effects of quantum coherence and interference. The con- trol of such effects is important to provide both a bet- ter understanding of the quantum realm as well as new functionalities to the circuits [1, 2]. Quantum interfer- ence allows to enhance or to cancel, total or partially, the response of the system beyond the simple classical additive behaviour. Such an effect can pose a problem to be avoided but also could provide new capabilities to the device with respect to its classical counterpart [3, 4]. There is currently an increasing interest in being able to tune the parameters of mesoscopic systems, what would enable to manipulate their quantum behaviour for the advantageous design and applications of future electronic devices [5]. The characteristic quantum phenomenon of the change of phase of the wave function along two paths enclosing a magnetic flux, i.e. the Aharonov-Bohm (AB) effect [6], has been envisioned for the feasible exploitation of the quantum phase in electronic devices [4, 7–9]. Phase co- herent effects in AB rings have been treated theoretically in the literature [10–15]. Closely related to the concept of coherence is the quantum interference between a discrete state with a continuum of states. This phenomenon was firstly studied by Fano in the spectrum of photoionization of atoms and termed Fano effect after him [16], but was found ubiquitous in a wide range of physical phenomena. Interestingly, it has been observed in properly tailored nanoscale systems [17–20]. The first tunable experiment showing the characteristic asymmetric Fano profile in the electron transmission was reported in an AB ring with a quantum dot embedded in one of its arms [17]. Various parameters, such as the gate voltage and the magnetic ∗Electronic address: [email protected] flux through the ring among others, allowed to tune the peak and dip of the profile. More recently, a quantum dot molecule, i.e. two coupled quantum dots, has been embedded between the arms of an AB interferometer, with even a larger number of tunable parameters [19]; the transmission thus exhibits a large variety of behaviours as a function of them. In [19] it is stated that the sim- plest fully coherent single-mode picture is of limited use for the interpretation of their experiments. In such a pic- ture, the transmission is assumed to arise from a direct reflection and another one after travelling once around the AB ring. It is the purpose of this paper to improve the theoretical description of the experimental results. Here we present theoretical calculations of a model in- spired in the system of Ref. [19] within a phase coherent non-interacting formalism. On one hand, we assessed the sensitivity of the conductance upon variation of the vari- ous model parameters. On the other hand, we show that the self-energy matrix elements, obtained from applying a partitioning technique to the Hamiltonian, can give in- sight on the quantum interference and its dependence on the magnetic flux through the ring. This interpretation in terms of contributions through pathways, allows one to predict the onset or cancellation of the antiresonances as a function of the parameters of the model. This paper is organized as follows: Section II describes the model of the device and its connections to the leads. Section III introduce the spacial contributions to the con- ductance by means of a partitioning technique and dis- cusses the conditions for the onset of peaks and cancel- lation of conductance in the transmission in terms of the Green functions of the isolated device. In Section IV we present the results obtained varying the various model parameters; finally, in Section V we summarize our con- clusions. 1 1 0 2 n u J 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 0 4 0 . 6 0 1 1 : v i X r a II. MODEL We consider four quantum dots forming a ring and coupled to two leads L and R. Sites 2 and 4 of the ring are connected to each other forming an artificial molecule, as shown in Fig. 1. We consider only one energy level in FIG. 1: Scheme of the device: (a) in the (1,2)-connection and (b) (1,3)-connection to the leads. each dot and both the intradot and interdot electron- electron interactions are neglected. The system shall be described by a Hamiltonian H = Hr + Hl + Ht, (1) where Hr is the Hamiltonian of the isolated bicyclic ring, 4(cid:88) 4(cid:88) Hr = † εid i di + i=1 i=1 + V (d ti,i+1(d † 2d4 + d † † i+1die−iϕ) i di+1eiϕ + d † 4d2) (2) Hl is the Hamiltonian of the leads † εkαc kαckα, Hl = (cid:88) k,α∈L,R (3) (cid:88) and Ht is the Hamiltonian describing the tunneling be- tween the leads and the ring Ht = † 1 + VRckRd† (VLckLd n) + H.c., (4) k where εi are the on-site energy at the dots, ti,i+1 are the nearest-neighbour hopping parameters (where t45 = t41 should be understood), V is the interdot hopping that couples the upper and lower arms of the interferometer, ϕ = 2πΦ/4Φ0 = πΦ/2Φ0 is the phase ϕ = πφ/2 acquired due to interdot hopping in a magnetic field threading the ring with a reduced flux φ = Φ/Φ0 (i.e., in units of the quantum Φ0 = h/e), and n is the site of contact to the right lead. The left lead is always attached to the dot 1, as shown in Fig. 1; for brevity we refer to them as connections (1,2) and (1,3). The current through the device can be calculated with the Landauer equation (cid:90) I = 2e h dE T (E) [fL(E) − fR(E)] , (5) 2 where fL and fR are the Fermi distributions at the L and R leads. At low temperatures, the transmission function represents the dimensionless conductance (in units of the quantum e2/2h) and is calculated as T (E) = 4Tr(ΓLGr(E)ΓRGa(E)), (6) where Ga and Gr are the matrix representation of the advanced and retarded Green functions, and ΓL and ΓR are the spectral densities of the leads. In the wide band approximation, the Green function of the connected system is given by Gr 1n = 1 − Γ2(g11gnn − g1n2) − iΓ(g11 + gnn) g1n . (7) where gij is the retarded Green function of the isolated system, and ΣL = ΣR = iΓ are the self-energies of the leads, considered to be energy independent. III. TRANSMISSION PATHWAYS Partitioning of the basis space is usually employed for isolating the effects on the part of interest from the rest of the system [21]. Here, we apply a partitioning technique to recover the notion of spatial transmission pathways [22]. The 4 × 4 Hamiltonian can be partitioned in terms of 2 × 2 matrices as follows: (cid:18) H P U U† H Q (cid:19) H = , (8) where H P = P HP is the part of the Hamiltonian pro- jected on the subspace of orbitals centered on the sites of connection 1 and n, where P = 1(cid:105)(cid:104)1 + n(cid:105)(cid:104)n, whilst H Q is the projection of H on the complementary sub- space Q = 1 − P . Matrices U and U† contain matrix elements connecting states belonging to P and Q. The Green function can be obtained by block matrix decom- position (cid:18) E − H P −U −U† E − H Q (cid:19)−1 . (9) g = (E − H)−1 = We are interested here in the Green function projected on the subspace of the connection sites, i.e., its P -block. Hence, gP can be obtained from the inverse of the Schur complement of the Q-block, E − H Q, gP = (E−H P−U (E−H Q)−1U†)−1 = (E−H P−U gQU†)−1, (10) from which an effective Hamiltonian can be defined as Heff = E − (gP )−1 = H P + U gQU† = H P + Σ, (11) where Σ = U gQU† is the self-energy that contains the interactions involving the orbitals not connected to the leads. An approach similar to the one outlined above has been used in an analytical treatment of quantum inter- ference in a benzene ring [22]. The Green function written in terms of the self-energy, (cid:18) E − εn − Σnn tn1 + Σn1 gP (E) = 1 ∆ t1n + Σ1n E − ε1 − Σ11 , (12) (cid:19) with ∆ = det(E − Heff ) = (E − ε1 − Σ11)(E − εn − Σnn) − Σ1n2, contains all the matrix elements needed for the calculation of the conductance between the sites 1 and n. Their poles are given by the zeroes of the secular determinant ∆ = 0, while the antiresonances comes from 3 ij + ΣC zeroes in g1n = (t1n + Σ1n)/∆. When t1n = 0, as in the (1,3) connection, the antiresonances arise from the zeroes of Σ1n. Interestingly, the self-energy Σij = ΣA ij + ΣB ij becomes a sum of contributions throughout paths from above (A), from below (B) and through the interarm coupling (C). The vanishing of the transmission occurs when the contributions from those paths interfere destructively thus cancelling the element t1n + Σ1n of the effective Hamiltonian. For the (1,3) connection, the contributions become ΣA ΣA ΣA 12g22, 11 = t2 13 = t12g22t23e2iϕ, 33 = t2 23g22, 41g44, ΣB 11 = t2 13 = t41g44t34e−2iϕ, ΣB ΣB 33 = t2 34g44, 11 = t12g24t41(e2iϕ + e−2iϕ), ΣC ΣC 13 = g24(t41t23 + t12t34), 33 = t23g24t34(e2iϕ + e−2iϕ), ΣC (13) (14) (15) g44 = (E − ε2)/D, where g22 = (E − ε4)/D, g24 = V /D, (16) and D = (E − ε2)(E − ε4) − V 2 = (E − Ea)(E − Eb), with Ea and Eb being the energies of the bonding (Eb) and antibonding (Ea) orbitals of the molecule formed between the sites 2 and 4 described by H Q, namely, Ea,b = (ε2 + ε4)/2 ±(cid:112)V 2 + (ε2 4 − 2ε2ε4)/4. 2 + ε2 On the other hand, the conditions for the onset of peaks of resonance in the linear conductance of the ring connected to the leads, can also be analyzed from the poles of the Green functions of the disconnected device. We state that the poles of g1n will show finite (or even perfect) transmission, while those poles of g11 or gnn (which are not poles of g1n) will show transmission zeros (antiresonances). Firstly, consider a non degenerate energy eigenvalue Ek of the disconnected device with eigenfunction ψk(cid:105) which is a linear combination of the site orbitals i(cid:105), If the state ψk(cid:105) has a non vanishing weight at the connection sites 1 and i ckii(cid:105), with cki = (cid:104)iψk(cid:105). ψk(cid:105) = (cid:80) n simultaneously, i.e. ck1 (cid:54)= 0 (cid:54)= ckn, then the spectral representation g1n(E) = (cid:104)1ψk(cid:105)(cid:104)ψkn(cid:105) E − Ek , = ck1c∗ E − Ek kn (17) (cid:88) k (cid:88) k shows that Ek is a pole of g1n because the term kn/(E − Ek) is present in the expansion (17). Fur- ck1c∗ thermore also the terms ck1c∗ E − Ek k1 , and cknc∗ E − Ek kn (18) will be present in the spectral representation of g11 and gnn, respectively, and Ek will also be a pole of them. That is, the poles of g1n also become poles of g11 and gnn and all three Green functions gij diverges as gij(E) ≈ Rij/(E − Ek), where Rij = ckic∗ kn (i, j = 1, n) is the residue of gij at the simple pole Ek. Therefore, the Green function of the connected ring, Eq.(7), can be approxi- mated as G1n(E) ≈ = 1 − Γ2(R11Rnn − R2 (E − Ek) − Γ2(R11Rnn − R2 R1n(E − Ek)−1 1n)(E − Ek)−2 − iΓ(R11 + Rnn)(E − Ek)−1 R1n 1n)(E − Ek)−1 − iΓ(R11 + Rnn) . (19) Taking (cid:104)1ψk(cid:105)2(cid:104)nψk(cid:105)2 − (cid:104)1ψk(cid:105)(cid:104)ψkn(cid:105)2 = 0, account into 1n = it reduces that R11Rnn − R2 to G1n ≈ R1n (E − Ek) − iΓ(R11 + Rnn) E→Ek−→ iR1n Γ(R11 + Rnn) (20) 4 which shows that the transmission has a pole at E = Ek + iΓ(R11 + Rnn) and a finite transmission T1n = 4R2 1n/(R11 + Rnn)2. The pole Ek acquires a finite width proportional to the coupling to the leads Γ. In the particular case where the sites 1 and n are topologi- cally equivalent because of the symmetry of the system, R11 = Rnn = R1n so that perfect transmission occurs. On the other hand, if E = Ek is a pole of g11 or gnn, but not of g1n, the numerator g1n(Ek) of Eq. (7) is finite whilst its denominator diverges; therefore T1n will show an antiresonance at E = Ek. In other words, a finite transmission occurs when the eigenstate ψk of the isolated system have non-vanishing projection on the orbitals 1(cid:105) and n(cid:105): (cid:104)1ψk(cid:105) and (cid:104)nψk(cid:105). Reciprocally, if one of them equals zero, the electron of energy E = Ek has a vanishing probability of being at both sites, and therefore no transmission can occur. The case when the eigenvalue Ek is degenerate requires a modification of the above argument. In such a case, there are more than one states ψ(1) . . . ψ(p) k , ψ(2) k having the same energy Ek. For the sake of simplicity, consider just two degenerate eigenfunctions ψ(p) ki i(cid:105), i c(p) (p = 1, 2). The spectral representation of the Green func- tion now reads k (cid:105) = (cid:80) k (cid:88) k gij(E) = Rij E − Ek , Rij = c(1) ik c(1)∗ jk +c(2) ik c(2)∗ jk . (21) 1n = 0, valid for the nondegen- The property R11Rnn − R2 erate case, does no longer hold here. Instead R11Rnn − R2 1n = (c(1) 1k )2(c(2) −c(1) 1k c(1) nk )2 + (c(2) 1k c(2)∗ nk c(2)∗ 1k )2(c(1) nk )2 nk c(1)∗ nk − c(2) 1k c(2) 1k c(1)∗ nk , (22) which can vanish or not depending on the c(p) If R11Rnn − R2 the real part of the denominator in Eq. near E = Ek as Γ(R11Rnn − R2 transmission becomes suppressed (T = 0). 1k and c(p) nk . 1n = 0, all above discussion holds; if not, (19) diverges 1n)/(E − Ek) and the IV. RESULTS AND DISCUSSION Firstly, consider a ring threaded by a magnetic flux with all interdot hopping parameters equal to each other (t12 = t23 = t34 = t41 = t) and interarm coupling V . The on-site energies are taken as ε1 = ε3 = 0, ε2 = 2 and ε4 = 4. All the results are obtained with a coupling to the leads Γ = 0.05. Figure 2 shows the dimensionless conductance T (E) for the ring connected to the leads in the configuration (1,3) (solid line), and for the three-sites chains forming the upper (dashed line) and lower (dotted line) arms with the side-dot 4 and 2, respectively, coupled by V . The transmission with an applied magnetic flux Φ = 0.1Φ0 is shown in dot-dashed line. The transmission show four peaks at the energy eigenvalues of the systems. FIG. 2: Transmission function of a ring with interarm cou- pling in the connection (1,3) (solid line), and three-sites chains with a laterally coupled side dot representing the up- per (dashed line) and lower (dotted line) arms as a function of the energy E. All dots are connected by the same interdot hopping t12 = t23 = t34 = t41 = 1 and interarm couplings V = −1, 0, 1, without magnetic flux (solid line) and with a flux Φ = 0.1Φ0 (dot-dashed line). (a) Ring with coupling V = −1, (b) ring with disconnected arms (V = 0), and (c) ring with coupling V = 1. The V couplings were chosen to show the tuning of the antiresonance with the one for the single chain with a lateral dot. The insets shows, in more de- tail, the suppression of the antiresonance due to the applied magnetic flux. The ring with disconnected arms Figure 2(b) shows also an antiresonance, not present in the chains because for V = 0 there is no side-coupled dot. This suppression of the transmission in the ring is due to the cancellation of the contributions to the self-energy throughout the upper and lower paths (Σ13 = ΣA 13 = 0). In general, eqs. (14) show that the self-energy vanishes if 13 + ΣB Σ13 = t2(g22e2iϕ + g44e−2iϕ + 2g24) (cid:2)(E − ε4)e2iϕ + (E − ε2)e−2iϕ + 2V(cid:3) = 0(23) = t2 D In absence of magnetic flux (ϕ = 0), Σ13 vanishes at the energy E = ¯ε − V = (ε2 + ε4)/2 − V . Figures 2(a)-2(c) depicts the tuning of the antiresonance with V , for V = (ε2−ε4)/2, V = 0 and V = (ε4−ε2)/2, respectively, such that the antiresonance of the ring can be made to coincide with that from the upper and lower arms with a lateral dot, respectively. When there is a finite magnetic flux, Σ13 is complex and its cancellation requires vanishing its real and imaginary parts, i.e, (2E − ε2 − ε4) cos 2ϕ + 2V = 0, and (ε2 − ε4) sin 2ϕ = 0. Both equations cannot be satisfied simultaneously, except when ε2 = ε4, which presents an antiresonance at ε2 − V / cos 2ϕ. Therefore, the magnetic field eliminates the antiresonance for a ring with different site energies ε2 (cid:54)= ε4 for arbitrary V . This suppression of the antiresonance is shown in the insets of Figures 2(a)-2(c) for a flux Φ = 0.1Φ0. Application of gate potentials at dots 2 and 4 allows to tune their on-site energies. Figure 3 shows the effect on the transmission of varying these parameters and the electron energy E with and without magnetic flux for the connection (1,2). Bright lines and dark regions rep- resent zones of high and low transmission, respectively. Variation of ε2 (left panels) was done at ε4 = 4 while variation of ε4 (right panels) was done at ε2 = 2. In absence of magnetic field (upper panels) three peaks of conductance are visible depicted by the bright curves. The central peak has a linear dependence on the energies ε2 and ε4 while the external ones are weakly dependent on them, as seen from the slope of the curves. Two an- tiresonances are also visible, namely, a faint vertical dark line at E = 0 independent on εi, and a vertical dark thicker line at E = ε4 − V (upper left panel) and the straight line E = ε4 − V (upper right panel). Hence, both ε2 and ε4 are equally suitable for tuning the max- ima of transmission, but only the site 4, which is not connected to the leads is efficient for tuning the antires- onances. The lower panels of Figure 3 show the effect of switching on a magnetic flux. As discussed for the connection (1,3), the antiresonances are cancelled out. In particular, it should be noted that the antiresonance at E = 0 turned into a peak of transmission. Also the second antiresonance at E = ε4 − V becomes weakened and the dark sharp straight line turns into a diffuse dark region of low transmission. Figure 4 shows in solid lines the transmission through an asymmetric ring where the upper arm has hoppings much smaller than those of the lower arm (t12 = t23 (cid:28) t34 = t41). The dotted blue and dashed red lines are the transmissions of the up- per and lower arms separately, calculated as three-site chains with a central site having energy ε2 and ε4, re- spectively, and lateral sites having on site energies ε = 0. In Figure 4, the four peaks of conductance, correspond- ing to connection (1,3), can be recognized as those from the lower arm along with the on-site energy of site 2. The transmission of the ring nearly coincides in almost the whole range with that of the lower arm, thus show- ing that the conduction is throughout such a pathway at almost every energy, except for E ≈ ε2. When the 5 FIG. 3: The logarithm of the transmission function of a ring with coupling between the arms in the connection (1,2) as a function of the energy E and the dots energies ε2 and ε4 with and without magnetic flux. All dots are connected by the same interdot hopping parameters t12 = t23 = t34 = t41 = V = 1, and the magnetic flux was taken as Φ = 0.1Φ0. Bright lines and dark regions represent zones of high and low transmission, respectively. Variation of ε2 (left panels) was done at ε4 = 4 while variation of ε4 (right panels) was done at ε2 = 2. up/(E − ε2) (cid:28) ΣB 13 = t2 incident electron is resonant with the site 2, the trans- mission is well described by the resonant peak of the up- per pathway. Nevertheless, none of the paths by them- selves can provide the onset of the antiresonance close to E ≈ ε2. The self energy Σ13 = ΣA 13 + ΣB 13 because down/(E − ε4), except for 13 = t2 ΣA E ≈ ε2 when they can become comparable. Hence, in a neighbourhood of E ≈ ε2, the self energy can be approx- down/(ε2 − ε4) which imated as Σ13 = t2 vanishes for E = ε2 + t2 down, that is, slightly to the right of the peak E = ε2, thus giving the Fano-like profile. up/(E − ε2) + t2 up(ε4 − ε2)/t2 13 ≈ ΣB The Fano-like peak shows the signature of the interfer- ence between both paths, typical when a localized state interferes with a continuum. Figures 4(a)-4(c) show the dependence of the Fano profile with the magnetic flux. With no magnetic flux, there is the above discussed Fano resonance due to the path interference; the application of a flux 0 < Φ < Φ0/4 suppress the antiresonance leaving only a dip in the transmission which also dis- appears at Φ = Φ0/4 leaving only the resonant peak, as seen in 4(b). Further increase of the flux in the range Φ0/4 < Φ < Φ0/2 produce a new dip at an energy slightly 6 field, such that Σ13(ϕ)2 = (ΣA 13)2 + (ΣB 13 + ΣB 13)2 + (ΣC 13)ΣC 13 cos 2ϕ, +2(ΣA 13)2 + 2ΣA 13ΣB 13 cos 4ϕ (24) where the first three terms represent the non-interfering transmission along the paths A, B and C. The last two terms contain the effect of the interference due to the quantum and magnetic phases. It is clearly noted that even for ϕ = 0 there is an interference between the path contributions to the self-energy. Interestingly, there are two periods in the magnetic phase; a period Φ = Φ0 (as- sociated to cos 4ϕ) and a period Φ = 2Φ0 (associated to cos 2ϕ). When there is no interarm coupling, ΣC 13 = 0, the latter is not present. On the other hand, as soon as a finite V exists, the self-energy acquires the longer period modulated by the shorter one. Such a behaviour has been observed in experiments [19] and were termed as Fano resonances of the big and small orbits. Figure 5 shows the transmission (in a log scale) T (E, ϕ) as a function of the energy and the magnetic flux, for vari- ous values of the interarm coupling V = 0, 0.5 and 1, and for the two ways of connecting to the leads. The top (left and right) panels, corresponding to V = 0, are the only ones showing a period Φ0 in the flux. As V increases, the period 2Φ0 becomes apparent. Finally, the bottom (left and right) panels show the transmis- sion for a single subring obtained from decoupling the sites 2 and 3 (i.e, t23 = 0), as also done in the experi- ments [19], where the period of the small orbit is clearly apparent. In the configuration (1,2) there are three in- terfering paths, namely, the direct path through sites 1 12 (1 → 4 → 2) and the path and 2 (t12), the path ΣC 12 (1 → 4 → 3 → 2), contributing to the self-energy ΣB Σ12(ϕ) = t12eiϕ + V t14g44e−iϕ + t14g43t31e−3iϕ. The ab- solute square of the self-energy turns out Σ12 = t2 12 + (ΣB +2t12ΣB 12)2 + (ΣC 12 cos 4ϕ, 12)2 + 2(t12 + ΣB 12)ΣC 12 cos 2ϕ (25) where the phase of the big orbit (4ϕ) characterize the interference between the pathways along the arms of the ring, whilst the phase of the small orbit (2ϕ) corresponds to the interference between them and the molecular bond. The bottom panels of Figure 5, shows the conductance when the hopping t23 has been set equal to zero, such that there is a single small orbit. Nevertheless, it should be noted a few differences visible as dark spots; they correspond to antiresonances determined by the different topology of the connections. V. CONCLUSIONS We have shown the conditions for the onset of reso- nances and antiresonances in an artificial quantum dot molecule embedded in the arms of an Aharonov-Bohm interferometer, and its dependence on the tunable param- eters. In general, the peaks of conductance are located FIG. 4: Transmission for the connection (1,3) as a function of the Fermi energy for an asymmetric ring with hopping param- eters t12 = t23 = 0.2 (upper arm), t34 = t41 = 1 (lower arm) and on-site energies ε2 = 2 and ε2 = 4 (solid line), as com- pared to the transmissions through the upper arm only (dot- ted blue line), and throughout the lower arm only (dashed red line). Figures (a)-(c) correspond to a decoupled ring (V = 0) (a) at zero magnetic field, (b) at Φ = Φ0/4, and (c) magnetic flux Φ = Φ0/2. Figure (d) has an interarm coupling V = 0.5 and magnetic flux Φ = Φ0/2. The insets show in more detail the behaviour of the curves around the resonance. smaller than ε2 while moves the peak to energies slightly higher. At Φ = Φ0/2 the dip in the transmission of the ring becomes an antiresonance, with the resonant peak tuned with that of the upper chain as seen in the inset of figure 4(c). Between Φ0/2 and Φ0, the behaviour of T (E) is reversed, such that a cycle is completed in a pe- riod of Φ0. Figure 4(d) depicts the transmission through the ring with an interarm coupling V = 0.5. Now the transmission through the lower (dashed line) and upper (dotted line) arms, including the site laterally coupled by V , shows Fano-like resonances at ε = ε2 and at ε = ε4, respectively. The transmission through the ring (solid line) still remains close to that of the lower arm with a lateral connection to site 2. Then, the overall picture for the transmission through a ring with different connec- tion strengths along each arm, is that of the transmis- sion throughout the stronger pathway (i.e., the one with larger hoppings) at almost every energy, except at the one resonant with the energy of the site connecting the arms where a Fano interference occurs. In the presence of a magnetic flux, the self energy is a complex quantity and its modulus should be considered. Eqs. (14) shows that the flux Φ introduces a phase ±2ϕ in the paths throughout the arms while no phase change occurs in the self-energy corresponding to the interarm coupling. Let us call Σ13(ϕ) the self energy with mag- 13e−2iϕ + ΣC netic field. Then, Σ13(ϕ) = ΣA 13, where ΣA,B,C are the real self-energies at zero magnetic 13e2iϕ + ΣB 13 7 at the energy eigenvalue of the isolated device. A parti- tioning technique enables to decompose the transmission as a sum along interfering pathways. The tunability of the coupling between the arms of the interferometer, al- lows one to weaken or enhance the contribution through the bond of the artificial molecule. The experimentally observed change in the period of the Aharonov-Bohm phase from one to twice the quantum of flux is inter- preted as due to the opening of one transmission pathway. Our results suggest that a modified coherent single-mode picture including the electron reflection in the subrings forming the small orbits, could also be of help in inter- preting the experiments. Application of a magnetic flux leads to a suppression of the antiresonances due to the partial cancellation of the destructive quantum interfer- ence. We have also discussed the differences with a con- nection not realized experimentally in which one of the dots of the molecule is attached to one lead. Such a con- figuration could provide other alternatives for tuning the conductance. Acknowledgements FIG. 5: The logarithm of the transmission coefficient log(T ) as a function of the electron energy E (horizontal axis) and the magnetic flux Φ (vertical axis) given in units of the quantum of flux Φ0, for V = 0, 0.5, and 1. All hopping parameters are ti,i+1 = 1 and all site energies ε = 0. The case V = 0 corresponds to that of a single loop (big orbit) enclosing the magnetic flux. The bottom pictures correspond to a single subring with V = t12 = t34 = t41 = 1 and t23 = 0 (small orbit) This work was partly supported by SGCyT (Universi- dad Nacional del Nordeste), National Agency ANPCYT and CONICET (Argentina) under grants PI 112/07, PICTO-UNNE 204/07 and PIP 11220090100654/2010. [1] Y. Imry, Introduction to mesoscopic physics, 2nd Ed. Ox- [11] M. L. Ladr´on de Guevara and P. A. Orellana, Phys. Rev. ford University Press, Oxford (2002). 73, 205303 (2006). [2] M. Di Ventra, Electrical transport in nanoscale systems, [12] I. G´omez, F. Dom´ınguez-Adame and P. Orellana, J. Cambridge University Press, Cambridge (2008). Phys.: Condens. Matter 16, 1613 (2004). [3] A. G. Aronov and Yu. V. Shavin Rev. Mod. Phys. 59, [13] E. R. Hedin, Y. S. Joe and A. M. Satanin, J. Phys.: 755 (1987). Condens. Matter 21 015303 (2009). [4] O. Hod, R. Baer and E. Rabani, J. Phys.: Condens. Mat- [14] X.-T. An and J.-J. Liu, Appl. Phys. Lett. 96, 223508 ter 20, 383201 (2008). (2010). [5] K. Barnham and D. Vvedensky (Eds.), Low-dimensional semiconductor structures. Fundamentals and device ap- plications, Cambridge University Press, Cambridge (2001). [6] Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). [7] G. Hackenbroich, Phys. Rep. 343, 463 (2001). [8] O. Hod, R. Baer and E. Rabani, J. Phys. Chem. B 108, 14807 (2004). [15] M. L. Ladron de Guevara, G. A. Lara, P. A. Orellana, Physica E 42, 1637 (2010). [16] U. Fano, Phys. Rev. bf 124, 1866 (1961). [17] K. Kobayashi, H. Aikawa, S. Katsumoto, and Y. Iye, Phys. Rev. Lett. 88, 256806 (2002). [18] A. Fuhrer, P. Brusheim, T. Ihn, M. Sigrist, K. En- sslin, W. Wegscheider, and M. Bichler, Phys. Rev. B 73, 205326 (2006). [9] O. Hod, R. Baer and E. Rabani, Phys. Rev. Lett. 97, [19] T. Ihn, M. Sigrist, K. Ensslin, W. Wegscheider and 266803 (2006). M Reinwald, New J. Phys. 9, 111 (2007). [10] M. L. Ladr´on de Guevara, F. Claro, and P. A. Orellana, [20] A. E. Miroshnichenko, S. Flach and Y. S. Kivshar, Rev. Phys. Rev. 67, 195335 (2003). Mod. Phys. 82, 2257 (2010) and references therein. [21] P.-O. Lowdin, J. Math. Phys. 3, 969 (1962). [22] T. Hansen, G. C. Solomon, D. Q. Andrews and M. A. Ratner, J. Chem. Phys. 131, 194704 (2009). 8 

1906.08695

1

1906

Classification of crystalline insulators without symmetry indicators: atomic and fragile topological phases in twofold rotation symmetric systems

Topological crystalline phases in electronic structures can be generally classified using the spatial symmetry characters of the valence bands and mapping them onto appropriate symmetry indicators. These mappings have been recently applied to identify thousands of topological electronic materials. There can exist, however, topological crystalline non-trivial phases that go beyond this paradigm: they cannot be identified using spatial symmetry labels and consequently lack any classification. In this work, we achieve the first of such classifications showcasing the paradigmatic example of two-dimensional crystals with twofold rotation symmetry. We classify the gapped phases in time-reversal invariant systems with strong spin-orbit coupling identifying a set of three $\mathbb{Z}_2$ topological invariants, which correspond to nested quantized partial Berry phases. By further isolating the set of atomic insulators representable in terms of exponentially localized symmetric Wannier functions, we infer the existence of topological crystalline phases of the fragile type that would be diagnosed as topologically trivial using symmetry indicators, and construct a number of microscopic models exhibiting this phase. Our work is expected to have important consequences given the central role fragile topological phases are expected to play in novel two-dimensional materials such as twisted bilayer graphene.

cond-mat.mes-hall

cond-mat

Classification of crystalline insulators without symmetry indicators: atomic and fragile topological phases in twofold rotation symmetric systems Sander H. Kooi,1 Guido van Miert,1 and Carmine Ortix1, 2 1Institute for Theoretical Physics, Center for Extreme Matter and Emergent Phenomena, Utrecht University, Princetonplein 5, 3584 CC Utrecht, the Netherlands 2Dipartimento di Fisica "E. R. Caianiello", Universit`a di Salerno I-84084 Fisciano (Salerno), Italy (Dated: June 21, 2019) Topological crystalline phases in electronic structures can be generally classified using the spatial symmetry characters of the valence bands and mapping them onto appropriate symmetry indicators. These mappings have been recently applied to identify thousands of topological electronic materials. There can exist, however, topological crystalline non-trivial phases that go beyond this paradigm: they cannot be identified using spatial symmetry labels and consequently lack any classification. In this work, we achieve the first of such classifications showcasing the paradigmatic example of two-dimensional crystals with twofold rotation symmetry. We classify the gapped phases in time- reversal invariant systems with strong spin-orbit coupling identifying a set of three Z2 topological invariants, which correspond to nested quantized partial Berry phases. By further isolating the set of atomic insulators representable in terms of exponentially localized symmetric Wannier functions, we infer the existence of topological crystalline phases of the fragile type that would be diagnosed as topologically trivial using symmetry indicators, and construct a number of microscopic models exhibiting this phase. Our work is expected to have important consequences given the central role fragile topological phases are expected to play in novel two-dimensional materials such as twisted bilayer graphene. I. INTRODUCTION Since the discovery of the quantum Hall effect [1], and its theoretical explanation in terms of the topological properties of the Landau levels [2 -- 4], topological phases of matter have become a rich playground for the theoreti- cal prediction and experimental verification of new quan- tum phenomena. From the birth of topological insulators [5 -- 12], to topological superconductors supporting Majo- rana zero modes [13 -- 17], to topological semimetals[18 -- 29], new types of topological phases keep arising. It is fair to say that the major theoretical effort in the field has been to classify, using appropriate mathemat- ical schemes, all possible topologically distinct gapped phases and subsequently relate them to topological in- dices. In the presence of internal symmetries -- time- reversal, particle-hole and chiral symmetry -- alone, the classification of free-fermion gapped phases has been ob- tained in all ten symmetry classes and arbitrary number of dimensions [30 -- 32]. The corresponding phases with non-trivial topology feature, by the bulk-boundary cor- respondence, protected gapless modes that are anoma- lous [33, 34]. The chiral (helical) edge states of quantum (spin) Hall insulators, as well as the single surface Dirac cones of strong three-dimensional topological insulators violating the fermion doubling theorem, are prime real- izations of such anomalies. In crystalline systems characterized by an additional set of spatial symmetries, new topologically distinct phases emerge [35 -- 38]. The non-trivial topology of a sys- tem is then manifested in the appearance of anomalous gapless surface modes, which are present only on surfaces that are left invariant under the protecting spatial sym- metry and violate stronger versions of the fermion dou- bling theorem [39, 40]. Furthermore, crystalline symme- tries can also yield non-trivial topological phases, dubbed higher-order topological states [41 -- 58], characterized by conventional gapped surfaces but with gapless anomalous one-dimensional modes at the hinges connecting two sur- faces related by the protecting spatial symmetry. As long as insulating systems are concerned, the ex- istence of anomalous surface or hinge boundary modes is deeply connected to the fact that non-trivial topolog- ical phases cannot be adiabatically connected to atomic insulators, whose insulating nature can be understood considering electrons as trapped classical point particles. In other words, a topological non-trivial insulator only arises when there is an obstruction in describing the sys- tem using an atomic picture. Therefore, the ground state of a topological non-trivial insulator cannot be repre- sented using exponentially localized Wannier functions respecting the internal and/or the set of spatial symme- tries of the system [59]. This obstruction to a "Wannier- representability", the classification in terms of topolog- ical invariants and the existence of gapless anomalous boundary modes can be formulated in a unique consis- tent framework for systems equipped only with internal symmetries [59, 60]. When adding spatial symmetries, however, different complications arise. First, distinct atomic insulators, which are by defini- tion topologically trivial, generally possess different crys- talline topological invariants. This, in turn, requires a careful inspection of such topological indices to identify the criteria dictating the appearance of topologically non- trivial crystalline phases. Second, there can exist "non- Wannierazible" topological phases in crystals which do not possess boundaries that are left invariant under the protecting spatial symmetry. As a result, the surfaces of 9 1 0 2 n u J 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 5 9 6 8 0 . 6 0 9 1 : v i X r a these systems are fully gapped even if the bulk is topolog- ical. Notwithstanding these complications, substantial progresses has been made with the theory of topologi- cal quantum chemistry [61] and that of symmetry-based indicators [62 -- 64], which allows one to discriminate all different atomic insulators from genuine topological non- trivial phases using the spatial symmetry character of the valence bands and their connectivity throughout the Brillouin zone. Combining these theories with density- functional-theory calculations has very recently led to catalogues containing a huge number of topological ma- terials [65 -- 67]. Nonetheless, there exist topological phases that are not detectable using the symmetry labels of the valence bands. An extreme case is a system with only translation symmetry: it can be in a topological "tenfold-way" phase due to its internal symmetries, but it is signaled as being topologically trivial using spatial symmetry indicators. More importantly, there can exist topological crystalline phases in low-symmetric crystals that are neither charac- terizable by the symmetry content of the valence bands nor by the tenfold-way [68]. To date, these phases lack any classification and consequently any material realiza- tion. In this work, we achieve the first of such classifica- tions. Specifically, we consider the paradigmatic example of two-dimensional crystals with twofold rotation sym- metry, i.e. in the wallpaper group p2, where the gapped phases of time-reversal symmetric (non-magnetic) sys- tems with sizable spin-orbit coupling cannot be classi- fied with the symmetry data of the valence bands. In- stead, we construct Berry phase related Z2 invariants to first isolate and remove topologically non-trivial quantum spin-Hall phases from the set of distinct gapped phases. Thereafter, we enumerate all distinct atomic insulating phases and classify them using a trio of Z2 topological invariants. Using our Berry phase based classification, we are able to determine: i) in systems with two oc- cupied valence bands, the existence of topological non- trivial crystalline phases similar in nature to the fragile phases detected by symmetry eigenvalues in other wall- paper groups [69, 70]. ii) with four occupied valence bands, the emergence of an additional fragile topologi- cal crystalline phase, whose possible existence has been overlooked so far. To underline the importance of these findings, we point out that topological crystalline phases of the fragile type have been predicted to occur in magic- angle twisted bilayer graphene [71 -- 74]. This paper is organized as follows. In Sec. II we first present the example of a time-reversal symmetric one- dimensional atomic chain where the symmetry character of the bands is not able to classify the distinct gapped phases, and show that such a classification becomes in- stead possible introducing a "partial" Berry phase Z2 invariant. We then show in Sec. III that these Z2 in- variants can be also defined on high-symmetry lines in the Brillouin zone of a two-dimensional crystal in the p2 wallpaper group, and can be used to first remove topo- 2 logical phases protected by time-reversal symmetry, and then classify atomic and fragile topological phases when two valence bands are occupied. In Sec. IV we introduce a new Z2 invariant corresponding to a "nested" quan- tized partial Berry phase, thanks to which we are able to diagnose the atomic insulating phases realized with four occupied valence bands and establish the existence of our novel NF = 4 fragile topological insulator. The trio of Z2 invariants is then used to classify all atomic insulating phases for a generic number of occupied Kramers pairs of bands in Sec. IV. Finally, we present our conclusions and comment on extensions of our work in Sec. V. II. MOTIVATION AND WARMUP IN 1D: MIRROR-SYMMETRIC CHAINS We start out by considering an atomic chain of spin one-half electrons with time-reversal symmetry and an additional mirror symmetry with respect to a one- dimensional (1D) mirror point. Moreover, we will assume inversion symmetry to be explicitly broken. The space group G for this atomic chain is generated by G = (cid:104){Et} ,{M0}(cid:105), where E is the identity, t the lattice translation vector, and M the mirror symmetry with respect to the 1D mir- ror point. In the unit cell of this 1D crystal, there are two distinct maximal Wyckoff positions whose site sym- metry group, or stabilizer group, is isomorphic to the point group Cs. The first, labelled 1a, has coordinate x = 0 and corresponds to the origin of the unit cell. Its stabilizer group is simply generated by {M0}. Simi- larly, the second maximal Wyckoff position, labelled 1b, corresponds to the edge of the unit cell with coordinate x = 1/2 in units of the lattice constant, and its sta- bilizer group is generated by {M1}, which is also iso- morphic to Cs. For all other positions in the unit cell, the stabilizer group only contains the identity. There- fore these Wyckoff positions have multiplicity two and coordinates (x,−x). Let us now enumerate the elemen- tary band representations [75] for exponentially localized Wannier functions (WFs) sitting at the maximal Wyckoff positions 1a and 1b. They can be induced by considering that in reciprocal space there are two mirror-symmetric momenta in the Brillouin zone (BZ), i.e. Γ = 0 and X = π. Moreover, since the stabilizer group of 1a does not contain any translation, the mirror eigenvalues ±i at Γ and X must be identical. On the contrary, the stabi- lizer group of 1b contains a lattice translation of half a unit cell and therefore the mirror eigenvalues at Γ and X are opposite. The elementary band representations can then be summarized as in Table I. Note that the "com- posite" band representation for two symmetric WFs [57] at the same position with opposite mirror eigenvalues ±i have a representation content in momentum space that is independent on whether they are centered at 1a or 1b. i ⊕ ρ1b−i ↑ G, This yields the equivalence ρ1a i ⊕ ρ1a−i ↑ G (cid:39) ρ1b 3 Wyckoff position Representation Γ X i i −i −i i −i −i i i ↑ G ρ1a ρ1a−i ↑ G i ↑ G ρ1b ρ1b−i ↑ G 1a 1b the one- Table I. Elementary band representation for dimensional space group of a mirror symmetric chain. The first column indicates the maximal Wyckoff positions. The second column the corresponding induced band representa- tion, and the last two columns the mirror eigenvalues at the center and edge of the 1D BZ. which simply states that the corresponding pairs of ex- ponentially localized WFs can be moved anywhere along the line between the 1a and the 1b sites in opposite di- rections. Figure 1. (color online) Evolution of the Wilson loop eigenval- ues for a mirror and time-reversal symmetric Aubry-Andr´e- Harper model [cf. Appendix A and Ref. 76] at NF = 4 by sweeping the dimerization hopping amplitude δt while pre- serving mirror symmetry (a) and changing the phase φV away from the mirror-symmetric point φV = −π/4 (b). The aforementioned composite band representation becomes a physical elementary band representation (PEBR) [61] when time-reversal symmetry Θ is taken into account. This is because Θ requires the complex irreducible one-dimensional representations at Γ and X to double. The corresponding pairs of energy bands, however, do not derive from Wannier states with charge centers at arbitrary positions along the chain. Kramers theorem indeed guarantees that exponentially localized WFs come in Kramers degenerate pairs, in which each pair has the same center. Moreover, while an even num- ber of Wannier Kramers pairs centered at the maximal Wyckoff positions 1a or 1b can be freely moved away without breaking either the mirror or time-reversal sym- metry, with an odd number of Wannier Kramers pairs sit- ting at 1a or 1b the center of at least one pair of Wannier states is unmovable [53]. Put differently, the parity of Wannier Kramers pairs centered at the maximal Wyck- off positions 1a and 1b represent stable topological Z2 in- dices characterizing a one-dimensional time-reversal and mirror-symmetric insulator. More importantly, these sta- ble topological indices cannot be read off from the sym- metry character of the bands since only one PEBR exists. The discrepancy between the existence of real space sta- ble topological indices and the absence of distinct PEBRs can be overcome using the recent finding that Kramers pairs of bands in a mirror symmetric [76], or equivalently C2 twofold rotation symmetric [77], atomic chain possess a Z2 topological index defined in terms of the "partial" polarization introduced by Fu and Kane [78], which is quantized by the presence of these point group symme- tries. In its U (NF ) gauge invariant form it can be written as νM := 1 π π dk TrA(k) + i log Pf [w(π)] Pf [w(0)] 0 mod 2. (1) (cid:20) (cid:21) In the equation above, we have introduced the non- Abelian Berry connection Am,n(k) = (cid:104)um(k) i∂k un(k)(cid:105), and the sewing matrix wm,n(k) = (cid:104)um(−k) Θun(k)(cid:105) that is antisymmetric at the Γ and X points and hence characterized by its Pfaffian Pf(w). The Z2 invariant de- fined above can be related to the charge centers of the Wannier Kramers pairs by introducing the unitary Wil- son loop operator [60, 79] Wk+2π←k = exp k+2π A(k(cid:48))dk(cid:48) , (2) (cid:35) (cid:34) i k (cid:80) j νj mod 1 ≡ 0, and consequently(cid:80) where exp denotes path ordering of the exponential while k is the Wilson loop base point. The eigenvalues of the Wilson loop operator, exp(2πi νj), j labelling the occu- pied bands, are independent of the base point k and uniquely determine the Wannier centers νj. The presence of mirror symmetry translates into a chiral symmetry for the Wilson loop eigenvalues [70], thus implying that the Wannier centers are restricted to the values νj = 0, 1/2 or to "unpinned" pairs (¯ν,−¯ν). Moreover, time-reversal symmetry guarantees that each Wilson loop eigenvalue has to be doubly degenerate. The concomitant presence of mirror and time-reversal symmetry therefore yields j νj mod 2 ≡ νM can only assume the values 0 and 1. Knowing the rela- tion between the Z2 topological invariant and the Wan- nier centers, we can straightforwardly classify the insulat- ing states realized in a one-dimensional mirror-symmetric atomic chain. In fact, with a total number of occupied bands NF = 4n+2, n being integer, an insulating atomic chain for which νM = 0 (νM = 1) will be characterized by the presence of an odd number of Wannier Kramers pairs at 1a (1b). If instead NF = 4n the system can be described in terms of exponentially localized Wannier functions with an even or odd number of Kramers degen- erate pairs centered at 1a and 1b depending on whether νM = 0 or νM = 1, respectively. To verify the relation between the Z2 topological in- variant νM and the Wannier centers distribution, we have computed the Wilson loop spectrum for a time- reversal and mirror symmetric one-dimensional spin- ful Aubry-Andr´e-Harper model [cf. Appendix A and a)b)0.3νν-π2-π40-12012ϕV0-12012δti 1b 1a Wyckoff position Representation Γ X Y M i i −i −i −i −i i −i i −i i −i −i i i −i −i i −i −i i i i −i −i i −i i −i i ↑ G ρ1a ρ1a−i ↑ G i ↑ G ρ1b ρ1b−i ↑ G i ↑ G ρ1c ρ1c−i ↑ G i ↑ G ρ1d ρ1d−i ↑ G 1d 1c i i Table II. Elementary band representation for the p2 wallpa- per group G = (cid:104){Et} ,{C20}(cid:105). The first column indicates the maximal Wyckoff positions; the second column the cor- responding induced band representation, and the last two columns the C2 eigenvalues at the Γ = {0, 0}, X = {π, 0}, Y = {0, π} and M = {π, π} points in the BZ. In time- reversal symmetric systems, the PEBRs obey the equivalence ρ1a ↑ G (cid:39) ρ1b ↑ G (cid:39) ρ1c ↑ G (cid:39) ρ1d ↑ G. Ref. 76], in which the half-filled NF = 4 insulating state undergoes a band gap closing-reopening, accompanied by a change of the Z2 topological invariant, by sweeping the strength of the nearest-neighbor hopping amplitude δt. As explicitly shown in Fig. 1(a), the insulating state can be described in terms of two Wannier Kramers pairs cen- tered at 1a and 1b in the νM = 1 region. On the con- trary, a νM = 0 value of the topological invariant implies the existence of two Wannier pairs centered at two mir- ror related, non-maximal Wyckoff positions in the unit cell. Moreover, by breaking the mirror symmetry of the model [see Fig. 1(b)] the position of the exponentially localized Wannier function can be freely moved at ar- bitrary positions in the unit cell in agreement with the fact that the space group in this case only contains the identity. Finally, we emphasize that the change of the Z2 invariant is associated with a band gap closing-reopening occurring at unpinned points in the BZ [76], which is a restatement of the fact that the topological index charac- terizing a mirror and time-reversal symmetric insulating chain cannot be inferred from the symmetry character of the occupied bands. III. WALLPAPER GROUP p2: INSULATORS WITH TWO OCCUPIED BANDS Having established the Z2 classification of mirror and time-reversal symmetric insulating chains in the absence of symmetry indicators, we next consider the main fo- cus of this work: two-dimensional (2D) crystals possess- ing a C2 twofold rotation symmetry. The smallest two- dimensional wallpaper group containing C2 is p2. It has four maximal Wyckoff positions labelled as 1a = {0, 0}, 1b = {1/2, 0}, 1c = {0, 1/2} and 1d = {1/2, 1/2}. Their stabilizer group is isomorphic to C2, which implies that in systems with time-reversal symmetry the induced band representations have the same symmetry character [cf. 4 Figure 2. (color online) Schematic drawing of a C2 symmetric Brillouin zone spanned by reciprocal lattice vectors g1 and g2 with high-symmetry points Γ, X, Y and M . The contours along which the partial Berry phases γI 2 are calculated are drawn in green, a typical Wilson loop operator contour, discussed in the main text, is drawn in red. 1 and γI Table II]. (cid:110) k2=G2/2 indices are not all k1=G1/2; νM However, the parity of the Wannier Kramers pairs centered at 1a,1b,1c,1d still represent real space sta- ble topological indices that discriminate between non- equivalent atomic insulating states. To classify these different atomic insulators, we first use the fact that in the BZ of a twofold rotation symmetric crystal, 2 H(k)C2 = H(−k) is the C2 symmetry constraint C−1 equivalent to a one-dimensional mirror symmetry con- straint along the time-reversal invariant non-contractible (cid:111) loop lines k1,2 ≡ 0, and k1,2 = G1,2/2. Therefore, we can in principle define a quartet of Z2 invariants k2=0; νM νM k1=0; νM [c.f. Fig. 2]. These topological independent, however, since the differences νM k1,2=0 can be re- lated [80] to the Fu-Kane-Mele (FKM) Z2 topological invariant [6, 78] characterizing a time-reversal invariant 2D topological insulator. This follows from the fact that k1,2=G1,2/2 − νM νM k1,2=0 keeps track of the evolution of the Wannier centers during a time-reversal pumping pro- cess [60]. Therefore, the condition νM k1,2=0 = 1 mod 2 immediately implies a quantum spin Hall (QSH) insulating state. When dealing with insulating crystalline systems without anomalous edge states (trivial FKM in- variant), we are thus left with a Z2×Z2 classification [81], which, as we will show below, is only able to diagnose the atomic insulating states when one Kramers pair of bands is occupied. k1,2=G1,2/2 − νM k1,2=G1,2/2−νM The assertion above can be immediately proved by us- ing the fact that for an atomic insulator with two oc- cupied bands, the exponentially localized Wannier Kra- mars' pair must be centered at one of the maximal Wyck- off positions. Hence, the corresponding center of charge already provides a Z2 × Z2 classification. Furthermore, the center of charge can be straightforwardly connected to the doublet of one-dimensional invariants νM k1,2=0 as follows. Let us consider the Wilson loop operator in the e1 direction W(k1+2π,k2)←(k1,k2) where (k1, k2) is the base k1=0 νM Wyckoff position νM k2=0 0 0 0 1 0 1 1 1 1a 1b 1c 1d Table III. The Z2 × Z2 classification of atomic insulators in the p2 wallpaper group with one occupied Kramers pair, i.e. NF = 2. The first column indicates the maximal Wyckoff position, while the second and third column are the U (2) gauge invariant line invariants. point. Its eigenvalues exp [2πi νj(k2)] (j = 1, 2) depend on the k2 coordinate of the Wilson loop base point and the corresponding phases νj(k2) are the centers of the one-dimensional hybrid Wannier functions [c.f. Fig. 2]. Due to time-reversal symmetry the Wannier bands realize a Kramers related pair [c.f. Appendix B], and therefore can be split into two time-reversed channels s = I, II satisfying νI (k2) ≡ νII (−k2). The additional C2 rotation symmetry mandates the Wilson loop spectrum to be chi- ral symmetric, i.e. νI (k2) ≡ −νII (k2). As a result, the center of charge of the Wannier Kramers pair in the e1 direction is νI (k2)dk2 mod 1 ≡ νI (k2 = 0) mod 1 ≡ νM k2=0 2 . 1 2π Repeating the same argument using the Wilson loop op- erator in the e2 direction, we therefore reach the classi- fication of atomic insulators with one occupied Kramers pair of bands summarized in Table III. Strictly speaking, this classification does not enumer- ate all possible insulating phases with a trivial FKM in- variant. Contrary to 1D systems where all insulating phases can be adiabatically continued to an atomic in- sulating phase [62], in 2D systems there can exist topo- logically non-trivial states that present an obstruction to a representation in terms of symmetric and expo- nentially localized WFs [69]. These topological phases have been dubbed "fragile" topological phases since al- though not admitting a Wannier representation by them- selves, such a representation becomes possible when ad- ditional trivial bands are added to the system. In recent works, the existence and diagnosis of fragile topological phases [70, 74, 82] have been linked to the topological na- ture of disconnected PEBR's [61]. However, the defining characteristic of a fragile topological phase -- the absence of a Wannier gap in the Wilson loop spectrum that con- sequently must display a non-trivial winding -- can exist also in our low-symmetric crystal with a single unsplit- table PEBR. In fact, due to the concomitant presence of the com- muting two-fold rotation symmetry and time-reversal symmetry, a crystal in the p2 space group is also in- variant under the combined antiunitary symmetry op- eration C2Θ with (C2Θ)2 = 1. Assuming a periodic and smooth real gauge can be found [83], this also implies that 5 the Wilson loop operator in the e1,2 direction belongs to the orthogonal group SO(2), with the homotopy group π1 [SO(2)] = Z guaranteeing the existence of an integer winding number invariant [84]. A C2Θ-protected fragile topological phase of this kind has been first discussed in Ref. 85 and dubbed Stiefel-Whitney (SW) insulator since the parity of the winding number corresponds to the second SW class invariant. Note that for a SW insu- lator to exist, the total Berry phases along the k1,2 ≡ 0 lines -- which correspond to the first SW class invari- ant in a smooth and periodic real gauge -- must vanish. This constraint is immediately verified in a C2 crystal with time-reversal symmetry. On the other hand, time- reversal symmetry also guarantees the winding number of the Wilson loop operator to assume 2Z values, which, in the language of Ref. 85 would imply the Z2 second SW class invariant to be trivial. However, in a NF = 2 insulator with time-reversal symmetry a Wilson loop spectrum winding an even num- ber of times cannot be unwinded. Consider the Wil- son loop operator W(k1,k2+2π)←(k1,k2) and assume, for instance, that the line invariant νM k1=0 = 0. The Wil- son loop spectrum has to display two symmetry enforced degeneracies at k1 = 0, π with the corresponding hy- brid Wannier centers at ν = 0. The absence of a Wan- nier gap also implies the existence of two degeneracies at time-reversal related momenta ¯k1,−¯k1 where the hybrid Wannier center ν = 1/2. The C2Θ symmetry mandates that these unpinned degeneracies can be only moved [c.f. Appendix B and Ref. 70] pairwise (as required by time- reversal), and consequently cannot be destroyed. Hence, the winding of the Wilson loop spectrum is robust, which allows for the definition of a fragile topological phase in insulators with one occupied Kramers pair of bands. Fur- thermore, the Wilson loop winding can occur indepen- dent of the Z2 line invariants, thus suggesting that the complete classification in systems with a trivial FKM in- variant is Z2 × Z2 × Z2, where the third Z2 invariant discriminates between gapped and winding Wilson loop spectra. To verify the existence of the fragile topological phase discussed above, we introduce a four-band tight-binding model on a C3 and mirror symmetry broken honeycomb lattice [see Appendix C for the corresponding tight- binding model] with a full spectral gap at half-filling [see Fig. 3(a)]. It can be thought of as being made of two coupled Chern insulators with opposite Chern num- bers C = ±2, thereby respecting time-reversal symmetry. In Fig. 3(b) we show the Wilson loop spectrum along the k1 direction, which displays the non-trivial winding discussed above. We close this section by emphasizing that the existence of the fragile topological phase does not strictly rely on the existence of a single PEBR. In Appendix C, we introduce a C4 symmetric tight-binding model on the square lattice where the NF = 2 atomic insulating states can be generally represented in terms of symmetric WFs centered at the maximal Wyckoff po- sitions 1a = {0, 0} and 1b = {1/2, 1/2}, which possess 6 spectra for two Kramers pairs in Fig. 4, where the red bands are centered around ν = 0 and the green bands around ν = 1/2. The blue bands can be seen as centered around either point [87]. Obviously, the parity of the pairs of Wannier bands belonging to the gapped region centered around ν = 1/2 can be linked to the line invari- ants νM k1,2=0. Considering for instance the spectrum of the Wilson loop W(k1,k2+2π)←(k1,k2) and further splitting the Wannier bands in two time-reversed channels, we imme- diately find that νM k1=0 = 1) if the Wilson loop spectrum region centered at ν = 1/2 is populated by an even [c.f. Figs. 3(b)-(d)] (odd [c.f. Fig. 3(a)]) number of pairs of Wannier bands. Furthermore, we can obtain two distinct Z2 invariants for the two disconnected regions of the k1 dependent Wilson loop spectrum as follows. Let us consider the Wilson loop operator W(k1,k2+2π)←(k1,k2), choosing its base point on the time-reversal and twofold rotation symmetric line k2 = 0 [c.f. Fig. 2]. The corre- , where the subscript e2 sponding eigenstates specifies the k2 direction of the Wilson loop, satisfy k1=0 = 0 (νM (cid:12)(cid:12)(cid:12)νj e2;(k1,0) (cid:69) = e2πiνj (k1)(cid:12)(cid:12)(cid:12)νj (cid:69) (cid:105)n (cid:69)(cid:104) νj e2;(k1,0) (cid:69) , W(k1,2π)←(k1,0) (cid:12)(cid:12)(cid:12)νj (cid:12)(cid:12)(cid:12)un = (cid:80) n (cid:69) (cid:12)(cid:12)(cid:12)wj e2;(k1,0) e2;(k1,0) and allow us to define the Wannier basis [42, 43], (cid:69) (cid:12)(cid:12)(cid:12)un (cid:69) (k1,0) (k1,0) e2;(k1,0) e2;(k1,0) (cid:12)(cid:12)(cid:12)wj , where n = 1, . . . , NF . Since the quantized partial polarization asso- ciated to the Bloch Hamiltonian eigenfunctions is unchanged by a general U (NF ) transformation, it fol- lows that the Z2 invariant νM k2=0 can be equivalently com- puted in the Wannier band eigenbasis . More importantly, working in such a basis allows us to de- compose νM k2=0 into two different Z2 invariants, which we M;1/2 M;0 dub as ν k2=0 , corresponding to the "nested" k2=0 and ν quantized partial polarizations for the two gapped sec- tors of the Wilson loop spectrum (the red and green bands in Fig. 4, respectively). This is because, as men- tioned above, the two gapped regions separately sat- isfy both time-reversal and twofold rotation symmetry, which guarantees that the partial polarization of the corresponding Wannier band eigenstates is quantized. Note that Wannier bands only respect twofold rotation and time-reversal symmetry when the Wilson loop base points lie on a mirror symmetric line. Having obtained three distinct Z2 topological invari- ants, we can now classify the atomic insulating phases enumerated above. Fig. 4(a) schematically shows the k1- dependent Wilson loop spectrum when the two gapped sectors are each populated with one pair of Wannier bands, and thus νM k1=0 = 1. The gapped sector cen- tered around ν = 0 is further characterized by the Z2 M;0 invariant ν k2=0, and its value dictates whether the Wan- nier Kramers pair is centered at the maximal Wyckoff M;0 position 1a (ν k2=0 = 1). The same argument can be applied to the gapped sector centered at ν = 1/2 to set apart Wannier Kramers pairs cen- M;0 k2=0 = 0) or 1b (ν Figure 3. (color online) (a) Band structure of the NF = 2 fragile topological insulator with twofold rotation and time- reversal symmetry. Energies have been measured in units of t. There are no degeneracies other than those required by time- reversal symmetry. (b) The Wilson loop spectrum along the k1 direction for the half-filled insulating state. See Appendix C for more details. distinguishable PEBRs. The symmetry content of the oc- cupied bands of our model is compatible with an atomic insulator with a Wannier Kramers pair centered at 1b. However, inspection of the Wilson loop spectrum firmly establishes it as being a topological insulator of the frag- ile type. IV. Z2 × Z2 × Z2 CLASSIFICATION WITH NF = 4: A NEW FRAGILE TOPOLOGICAL PHASE With the Z2 × Z2 × Z2 classification of NF = 2 insu- lating phases in our hands, we next consider insulators with NF = 4. We will follow the same strategy used in the preceding section, and enumerate and classify all the existing atomic insulating phases. It is easy to see that there exist seven distinct insulating states representable in terms of symmetric WFs. In fact, with two Wannier Kramers pair in the system, their centers will either lie at two C2 related non-maximal Wyckoff positions or at two distinct maximal Wyckoff positions. Therefore, the two Z2 line invariants νM(k1,2 = 0) are insufficient to clas- sify these states. Now we will show, using a procedure similar to the "nested" Wilson loop one of topological multipole insulators [42, 86], that it is possible to obtain an additional Z2 invariant by identifying two sectors in the Wilson loop spectrum, each of which carries its own topological content, i.e. its quantized partial polariza- tion. We recall that the essential characteristic of a generic atomic insulating state is the presence of a Wannier gap in the Wilson loop spectrum. Its chiral symmetry, dic- tated by the C2Θ symmetry, then allows us to distinguish two regions, one symmetrically centered around ν = 0 and one symmetrically centered around ν = 1/2, each possessing both twofold rotation and time-reversal sym- metry, and populated by Kramers related pairs of Wan- nier bands. We have plotted the possible Wilson loop a)b)νEΓXMYΓ-4-20240π2π-12012k17 Next, we consider insulating states where the Wannier bands occupy only one gapped sector of the Wilson loop spectrum, and thus νM k1=0 = 0. Fig. 4(b),(c),(d) show the allowed possibilities for the Wannier bands. They can either realize a connected pair with two protected degen- eracies at time-reversal related momenta (¯k1,−¯k1) or can come in disconnected pairs, in which case the two pairs can be arbitrarily assigned to the ν = 0 or the ν = 1/2 M;0 sector. Let us first inspect the value the invariants ν k2=0 M;1/2 (ν k2=0 ) assume for the connected pair of Wannier bands shown in Fig. 4(b),(c). We can divide the four Wannier bands in two time-reversed channels, that each possess C2Θ symmetry. Then, an essential twofold degeneracy in one channel at ν = 0 (ν = 1/2) implies a π Berry phase [see Appendix B and Ref. 85], and consequently M;1/2 the nested line invariant ν k2=0 ) is enforced to be 1. As a result, the schematic Wannier bands shown in Fig. 4(b),(c) correspond to the atomic insulating phase with Wannier Kramers pairs centered at 1a ⊕ 1b and 1c ⊕ 1d respectively. Using similar arguments, we also find that the disconnected Wannier bands of Fig. 4(d) are characterized by a zero nested partial polarization [see Appendix B]. Therefore, in this atomic insulating state the Wannier Kramers pairs are centered at two non-maximal Wyckoff positions related to each other by the twofold rotation symmetry. All in all, we have thus reached the classification summarized in Table IV of the seven distinct atomic insulating states realizable in the p2 wallpaper group with four occupied bands. M;0 k2=0 (ν M;1/2 k2=0 = 1 with ν When comparing this with the eight allowed config- urations for the three Z2 invariants, one can immedi- ately recognize that an insulating state characterized by the two nested quantization polarization invariants M;0 M;0 ν k1=0 = 0 cannot be rep- k2=0 = ν resented in terms of symmetric exponentially localized Wannier functions. In fact, such a configuration featur- ing essential degeneracies at unpinned momenta k1 both around ν = 0 and ν = 1/2 would necessarily imply the closing of the Wannier gap and hence a non-trivial wind- ing of the Wilson loop. We thus conclude that such an insulator corresponds to a topologically non-trivial phase of the fragile type. Its stability against symmetry- allowed perturbations is rooted in the fact that the pos- sible local annihilation of the degeneracies on the ν = 0 or ν = 1/2 line requires a change of the line invariant νM k2=0 = with a bandgap closing-reopening point. mod 2, which is only possible M;0 k2=0 + ν M;1/2 k2=0 (cid:17) (cid:16) ν Let us now present a model realization of this novel fragile topological insulating phase. The model is built by stacking two quantum spin-Hall insulators on the hon- eycomb lattice -- the so-called Kane-Mele model [5] -- with opposite sign of the spin-dependent next-nearest neighbor hopping t2 parametrizing the spin-orbit cou- pling strength. Inversion symmetry is explicitly broken by considering a chemical potential difference between the two layers while the threefold rotation symmetry breaking due to, e.g., a uniaxial strain [c.f. Fig. 5(a)] is Figure 4. (color online) Schematic drawings of the Wilson loop spectra for the NF = 4 atomic insulating states in the p2 wallapaper group. Panel (a) corresponds to four differ- ent atomic insulating states, where the pair of bands around ν = 0 (ν = 1/2) can have a Wannier center at 1a or 1b (1c or 1d), respectively, which can be determined by calculating their nested partial polarizations. Panel (b) corresponds to an atomic insulator with Wannier Kramers pairs centered at 1a⊕1b, while panel (c) is for 1c⊕1d. In panel (d) the Wannier functions are centered at C2 related generic points in the unit cell. Wyckoff positions νM k1=0 ν 1 1 1 1 0 0 0 1a ⊕ 1c 1a ⊕ 1d 1b ⊕ 1c 1b ⊕ 1d 1a ⊕ 1b 1c ⊕ 1d ν ⊕ −ν M;0 k2=0 ν 0 0 1 1 1 0 0 M;1/2 k2=0 0 1 0 1 0 1 0 Table IV. The classification of atomic insulating states in the p2 wallpaper group when two occupied Kramers pairs of bands are occupied, i.e. NF = 4. The first column indicates the centers of charge of the Wannier Kramers pairs; the second column is the Z2 line invariant of the full Wilson loop spec- trum; the second and third columns are the invariants de- rived from the nested Wilson loops, which obey the sum rule M;0 k2=0. The last row refers to in- ν k2=0 + ν sulators where the Wannier Kramers pairs are centered at C2 related non-maximal Wyckoff positions. mod 2 = νM M;1/2 k2=0 (cid:16) (cid:17) M;1/2 tered at 1c (ν k2=0 = 1). This, in turn, allows us to catalogue four distinct atomic insu- lating states. M;1/2 k2=0 = 0) and 1d (ν a)b)c)d)νν-0π2π12012k1ν-0π2π12012k1ν0π2π-12012k10π2π-12012k18 Wyckoff positions νM k1=0 ν 1 1 0 0 0 0 1 1 1a ⊕ 1b ⊕ 1c 1a ⊕ 1b ⊕ 1d 1a ⊕ 1c ⊕ 1d 1b ⊕ 1c ⊕ 1d 1a ⊕ ν ⊕ −ν 1b ⊕ ν ⊕ −ν 1c ⊕ ν ⊕ −ν 1d ⊕ ν ⊕ −ν M;0 k2=0 ν 1 1 0 1 0 1 0 0 M;1/2 k2=0 0 1 1 1 0 0 0 1 Table V. The Z2 × Z2 × Z2 classification of atomic insulating states in the p2 wallpaper group when three occupied Kramers pairs of bands are occupied, i.e. NF = 6, indicating the re- lation between the Wannier Kramers pairs center of charges and the ("nested") quantized partial polarization topologi- cal invariants. This classification is generically valid for an arbitrary number of occupied bands NF = 4n + 2 with the integer n ≥ 1, which will only include more unpinned pairs of Kramers pairs. generic NF = 6 insulating state, the Z2 × Z2 × Z2 clas- sification is saturated by enumerating the phases with symmetric Wannier function. In fact, with three Wan- nier Kramers pairs in the system, their centers can either lie on three distinct maximal Wyckoff positions, or two Wannier pairs sit at C2-related non-maximal Wyckoff po- sition with a third pair located at one maximal Wyckoff position. Inspecting the possible features of the Wilson loop spectrum and iterating the arguments presented in the former sections we reach the classification summa- rized in Table V. Note that this classification is generally valid for NF = 4n + 2 and n ≥ 1. In fact, by adding two Wannier Kramers pairs to a state with NF = 6, we will end up in one of the NF = 6 configurations [c.f. Table V] with the addition of two Wannier Kramers pair cen- tered at unpinned two-fold rotation symmetric momenta, which do not change the Z2 invariants. Finally, in Table VI we also provide the classification of atomic insulators with four Wannier Kramers, which is also valid for a generic number of occupied bands NF = 4n and n > 1. Note that the distribution of Z2 invariants is strictly equivalent to the case of four occu- pied bands. However, the topological non-trivial fragile phase is substituted by an atomic insulator where the four Wannier Kramers pairs are centered at the four max- imal Wyckoff positions. In this configuration, in fact, the Wilson loop spectrum is the superposition of Fig. 4(b) and Fig. 4(c) which is allowed with a full Wannier gap with a minimum number of eight Wannier bands. VI. CONCLUSIONS In this paper, we presented a classification of gapped insulating phases that cannot be diagnosed using crys- talline symmetry eigenvalues. We have showcased two- dimensional crystals in the wallpaper group p2 where Figure 5. (color online) (a) Top view of the strained honey- comb bilayer realizing the NF = 4 fragile topological phase. The intralayer spin-dependent hopping amplitude t2 has been taken only along the zigzag direction to amplify the threefold rotation symmetry breaking. (b) Bulk bands showing a full spectral gap at half-filling. The parameter set is specified in Appendix D. (c) The corresponding spectrum in a ribbon ge- ometry demonstrate the insulating nature of the edges. (d) Wannier bands along the k2 direction. The Wilson loop in the k1 direction also show a similar winding. incorporated taking direction dependent hopping ampli- tudes t2. We also break the Mz symmetry by introducing a Rashba spin-orbit coupling term. Being composed of two quantum spin Hall insulators, the FKM invariant of the half-filled model is trivial, and with an explicit inter- layer coupling the helical edge states disappear [see Ap- pendix D for the model Hamiltonian and Fig. 5(b) for the ribbon spectrum]. A direct computation of the Wilson loop spectrum [c.f. Fig. 5(d)] shows the non-trivial wind- ing with the line invariants νM k1,2=0 = 0 that present an obstruction to the Wannier representation of this phase. In Appendix D, we also present a spinful model inspired by the px,y orbital model presented in Ref. [82] that also realizes the NF = 4 fragile topological phase discussed above. V. MORE OCCUPIED BANDS Contrary to the NF = 2 topologically non-trivial phase, which is trivialized only when certain Kramers pairs of bands are added , the NF = 4 topological insula- tor discussed above is intensively fragile: it is trivialized by the addition of a generic Kramers pair of bands. This assertion can be immediately proved noticing that for a a)b)EΓXMYΓ-4-20240π2π-4-2024k1Eνc)d)0π2π-12012k2Wyckoff positions 1a ⊕ 1b ⊕ 1c ⊕ 1d ν 1 ⊕ −ν 1 ⊕ ν 2 ⊕ −ν 2 1a ⊕ 1b ⊕ ν ⊕ −ν 1a ⊕ 1c ⊕ ν ⊕ −ν 1a ⊕ 1d ⊕ ν ⊕ −ν 1b ⊕ 1c ⊕ ν ⊕ −ν 1b ⊕ 1d ⊕ ν ⊕ −ν 1c ⊕ 1d ⊕ ν ⊕ −ν M;1/2 k2=0 νM k1=0 ν 0 0 0 1 1 1 1 0 M;0 k2=0 ν 1 0 1 0 0 1 1 0 1 0 0 0 1 0 1 0 Table VI. The Z2 × Z2 × Z2 classification of atomic insulating states in the p2 wallpaper group when four occupied Kramers pairs of bands are occupied, i.e. NF = 8, indicating the re- lation between the Wannier Kramers pairs center of charges and the ("nested") quantized partial polarization topologi- cal invariants. This classification is generically valid for an arbitrary number of occupied bands NF = 4n with the in- teger n > 1, which will only include more unpinned pairs of Kramers pairs. all gapped phases have the same physical elementary band representation, but they can be nevertheless clas- sified with three Z2 topological invariants: the quan- tized nested partial polarizations -- partial Berry phases -- along high-symmetry lines in the two-dimensional Bril- louin zone of the system. Using the ensuing Z2 × Z2 × Z2 classification, we have been able to classify all atomic insulating states and iden- tify non-Wannierazible topological crystalline phases pro- tected by twofold rotation symmetry and time-reversal symmetry. Since the crystal does not possess bound- aries that are left invariant under the protecting twofold rotation symmetry, these topological phases do not dis- play gapless anomalous boundary modes although their bulk is topologically non-trivial. Instead, they represent an example of the recently discovered fragile topology, and thus they can be trivialized with the addition of atomic valence bands. In this respect, we wish to em- phasize that the fragile topological phase realized with two occupied valence bands, which is similar in nature to the fragile phases recently discussed in the literature in other wallpaper groups does not necessarily decay into a Wannierazible atomic insulating state when an addi- tional Kramers related pair of bands are introduced. In fact, such band addition might lead to our novel NF = 4 topological crystalline phase whose Wilson loop winding is strictly protected by the quantization of the nested quantized partial Berry phase in the presence of time- reversal and twofold rotation symmetries. An interesting direction for future research is the ex- tension of the classification presented here to other wall- paper and space groups where the symmetry data of the valence bands could be combined with Berry phase invariants to search for new topological electronic ma- terials. Furthermore, the Berry phase invariants for atomic insulating phases can be also exploited to ob- tain, using the Wannier centers flow of hybrid Wannier functions [53, 57], topological invariants for higher-order topological insulators with helical hinge modes in non- centrosymmetric crystals. 9 ACKNOWLEDGMENTS C.O. acknowledges support from a VIDI grant (Project 680-47-543) financed by the Netherlands Organization for Scientific Research (NWO). This work is part of the re- search programme of the Foundation for Fundamental Research on Matter (FOM), which is part of the Nether- lands Organization for Scientific Research (NWO). S.K. acknowledges support from a NWO-Graduate Program grant. Appendix A: Spin-orbit coupled Aubry-Andr´e-Harper model To analyze 1D atomic chains with time-reversal and mirror symmetry with respect to a mirror point, we con- sider a tight-binding model [76] for spin-1/2 electrons corresponding to a generalized Aubry-Andr´e-Harper model [88 -- 90] (cid:88) j,σ +i (cid:88) (cid:88) j,σ,σ(cid:48) H = [t0 + δt cos (πj + φt)] c † j+1,σcj,σ [λ0 + δλ cos (πj + φλ)] c † j+1,σsy σσ(cid:48)cj,σ(cid:48) + † [V0 + δV cos (jπ/2 + φV )] c j,σcj,σ + H.c., j,σ † where c j,σ is the creation operator for an electron at site j with spin σ (σ =↑,↓), and si are the conventional Pauli matrices. The Hamiltonian contains harmonically modulated nearest-neighbor hopping, spin-orbit coupling and onsite potentials of amplitudes δt, δλ, and δV , and phases φt, φλ and φV . The periodicities of the modu- lated hopping and spin-orbit coupling have been chosen to be of two lattice sites while the periodicity of the on- site potential is four lattice sites. Moreover, t0, λ0 and V0 are the site-independent amplitudes of the hopping, spin-orbit coupling and on-site potential. The model pos- sesses time-reversal symmetry whereas mirror symmetry is preserved only for specific values of the phases φt,λ,V . In Fig. 1 we have chosen the parameter set φt = φλ = π, λ0 = 0.5t0, δλ0 = −0.3t0 and δV = t0. The Wilson loop eigenvalues shown in Fig. 1(a) have been obtain using the mirror-symmetric value φV = −π/4 while sweeping the dimerized hopping amplitude δt. In Fig.1(b), instead, we have fixed δt = −0.25t0 while sweeping φV away from the mirror symmetric point. Appendix B: Nested partial polarization in Wilson loop spectra with C2 and Θ symmetry Here we show that the nested partial polarizations are well-defined quantities in gapped Wilson loop spectra, and that they are quantized in the presence of C2 and Θ symmetry. In addition, we calculate the partial polariza- tions for various Wilson loop spectra. Let us start by examining how the symmetries act on the Wilson loop. Consider the Wilson loop operator W(k1,2π)←(k1,0), C2 and Θ symmetry then require [70, 79] C2W(k1,2π)←(k1,0)C† 2 = W† ΘW(k1,2π)←(k1,0)Θ† = W† (−k1,2π)←(−k1,0), (−k1,2π)←(−k1,0), where the complex conjugate on the right-hand side comes from the fact that both symmetries send k → −k and hence reverse the contour of the Wilson loop opera- tor. Furthermore, C2 relates the eigenvalues of the Wilson loop operator {νi (k1)} = {−νi (−k1)} , and time-reversal relates {νi (k1)} = {νi (−k1)} , where {} denotes the set of eigenvalues. Hence C2Θ en- forces a chiral symmetry in the Wilson loop spectrum. Now let us show, following Ref. [70], that a single cross- ing in the Wilson loop spectrum is locally protected by the combination of C2 and Θ symmetry. Let us work in a basis where C2Θ = K, where K indicates complex con- jugation. The symmetry restriction on the Wilson loop operator is then KW(k1,2π)←(k1,0)K = W(k1,2π)←(k1,0), since C2Θ sends k → k. Since the Wilson loop operator in this basis is an SO (N ) matrix, we can write it as the exponential of an Hermitian matrix HW , which is restricted by C2Θ such that Near a two-band crossing, this restriction means that lo- cally HW (k1) = k1 · σy. A single twofold degeneracy on the ν = 0, 1/2 lines cannot be gapped out without break- ing C2Θ symmetry, but only moved on the line. There- fore, as for a Weyl point, the degeneracy can be only removed by pair annihilation. We now turn to the various possible Wilson loop spec- tra, and compute their partial polarizations. In Fig. 6(a) we have drawn a generic Wilson loop spectrum for one oc- cupied Kramers pair. The corresponding Wannier bands are given by ∗ HW (k1) = −HW (k1) . Now 10 k and ψII where ψI k are the Bloch waves (schematically drawn in Fig. 6(b) along the same contour), and the coefficients are given by the eigenvectors of the Wilson loop matrix [see also Sec. IV]. The Wannier bands in Fig. 6(a) are thus obtained by a unitary transformation on the occupied eigenstates of the Hamiltonian [(]Fig. 6(b)], and will be linear combinations thereof. These Wannier bands satisfy [(]see again Fig. 6(a)], k = eiθ(k)C2ϕI−k, ϕI ϕI k = eiφ(k)ΘϕII−k. The partial polarization is in this case given by the Berry k is C2 symmetric, its Berry phase, phase of ϕI and hence the partial polarization is quantized to 0, π. k. Since ϕI Now consider two occupied Kramers pairs with two crossings at ν = 0 (Fig. 6(c)). The colors indicate the Kramers partners, and the dotted (solid) lines are C2 partners. To find the partial polarization we split the bands into two time-reversal channels. The only pos- sibility that leaves us with periodic subsets of bands is taking the solid blue and dotted red bands together, and the solid red and dotted blue bands together (shown in Fig. 6(c) on the right). Let us denote the solid blue Wannier band by a (k), and define the red dotted band b (k) by b (k) := C2θ (k) . Clearly the bands are not periodic, and we have a (2π) = b (0) , b (2π) = a (0) . We now try to construct a periodic gauge by a basis transformation, under which the partial polarization is invariant. We define a (k) = [a (k) + b (k)] /2, b (k) = [a (k) − b (k)] /2. a (2π) = [a (2π) + b (2π)] /2 = [b (0) + a (0)] /2 = a (0) , hence a (k) is periodic, however b (2π) = [a (2π) − b (2π)] /2 = [b (0) − a (0)] /2 = −b (0) , ϕI k = αψI ϕII k = γψI k + βψII k , k + δψII k , is anti-periodic. Multiplying by a phase and defining b (k) = eik/2b (k) , remedies this situation. Under C2Θ we now have C2Θ a (k) = a (k) , C2Θ b (k) = e−ik/2b (k) = e−ikb (k) . 11 Using this, we can calculate the Berry phase of the two bands separately, γa = = = − = −γa dk a (k) dk a (k) † † † i∂ka (k) (C2θ) † dk a (k) i∂ka (k) (C2θ) i∂ka (k) and γb = = dk dk b (k) † b (k) † b (k) i∂k (C2θ) † b (k) dk ∂kk b (k) † (C2θ) i∂k eiki∂ke−ikb (k) = − dk = −γb − = −γb − 2π, from which we see γa = 0 and γb = π, and thus we find that the partial polarization is π. In particular, this shows that the nested polarization around ν = 0, 1/2 will be π when there are an odd number of crossings in half the Brillouin zone on this line. Let us finally consider two occupied Kramers' pairs with a disconnected Wilson loop spectrum (see Fig. 6(d)- (e)). To calculate the partial polarization of these bands, let us first consider the red Kramers pair in isolation. To calculate the partial polarization we need to again calculate the Berry phase of the red dotted band. This band does not posses C2 symmetry and hence its Berry phase will not be quantized. In order to calculate the partial polarization of the blue bands, we calculate the Berry phase of the blue dotted band. However, since the blue dotted and the red dotted bands are related by C2 symmetry, we find for their Berry phases Red = −γI γI Blue, and hence the partial polarization, which is the sum of the two, is zero. To calculate nested partial polarizations, we need to select symmetric regions centered around ν = 0 and ν = 1/2. Since there are two gaps in the spectrum, we have two choices. We can either include the pair of Kramers pairs, or exclude them from either region. However, we have just seen that the partial polarization of this set of k and ϕII Figure 6. (a) Schematic drawing of a generic Wilson loop spectrum with one occupied Kramers pair. The two Wan- nier bands ϕI k are related to each other by time- reversal symmetry, and are themselves C2 symmetric. (b) Generic band structure corresponding to (a), the Wannier states are obtained by linear combinations of the eigenstates of the Hamiltonian ψI k . (c) Wilson loop spectrum for two occupied Kramers pairs with two crossings at ν = 0. The colors denote the two different Kramers pairs, and the C2 partners have a solid (dotted) line. The two time-reversal channels are depicted on the right, which are by themselves C2Θ symmetric. (d) Disconnected Wilson loop of two occu- pied Kramers pairs, again color denotes Krames pairs, dotted (solid) the C2 partners. (e) Corresponding band structure with two occupied Kramers pairs. k,ψII bands is zero, and hence either choice will yield the same result. Indeed for any NF , the only choice in selecting a subset of bands centered around ν = 0, 1/2, is including or excluding pairs of disconnected bands such as in Fig. 6(e), making the nested partial polarizations well-defined quantities. Finally, gapped Wilson loop spectra for arbitrary NF a)b)ν-0π2π12012k1ΓXΓk1Eνd)e)0π2π-12012k1ΓXΓk1E-0π2π12012k1νc)A topological phase of the fragile type can also be ob- tained in a fourfold rotation symmetric system by consid- ering the following C4 symmetric C = +2 Chern insulator, 12 H = −
1 [cos (kx) + cos (ky)] (τ0 + τz) −
2 [cos (kx) + cos (ky)] (τ0 − τz) − 2t [cos (kx) − cos (ky)] τx − t2 [sin (kx) sin (ky)] τx, where
1,
2, t, t2 are hopping amplitudes. We then again add a time-reversal copy and couple them by Hmix = −λ [sin (kx) τ0sy + sin (ky) τ0sx] . The bulk bands and Wilson loop spectrum are plot- 7. The C4 operator is represented by ted in Fig. C4 = τz ⊗ eisz/4, and time-reversal by Θ = UK with U = I2 ⊗ isy and K is complex conjugation. The symmetry eigenvalues of the occupied bands at Γ are {eiπ/4, e−iπ/4}, and at the M point {−e−iπ/4,−eiπ/4}, which are compatible with a Wannier function centered at the maximal C4 symmetric position 1b = {1/2, 1/2}. Appendix D: Fragile topological insulators with two occupied Kramers pairs of bands To construct a fragile topological insulator with two oc- cupied Kramers pairs, we consider two copies of a quan- tum spin Hall insulator (the Kane-Mele model [5]) with broken C3 symmetry on a honeycomb lattice, cα† i,σcα j,σ Hα KM = −tα (cid:88) − iλα (cid:88) (cid:104)i,j(cid:105),σ − (−1)α itα 2 cα† i,σcα j,σ +
α(cid:88) (cid:88) (cid:104)(cid:104)i,j(cid:105)(cid:105)x,σ i,σ (d × s)σσ(cid:48) cα† ηijcα† z i,σ i,σcα j,σ cα j,σ(cid:48), (cid:104)i,j(cid:105),σ,σ(cid:48) where α = 1, 2 denotes the two copies of the Kane- Mele model, tα denotes the hopping amplitude, (cid:104)i, j(cid:105) the sum over nearest-neighbors, tα 2 the amplitude of intrinsic spin-orbit coupling, (cid:104)(cid:104)i, j(cid:105)(cid:105)x the sum over next-nearest neighbors only in the x-direction, ηij = +1 (−1) for hop- ping in the clockwise (counter-clockwise) direction, λα the Rashba amplitude, d the vector between two sites, s the vector of Pauli matrices and
α an on-site poten- tial. Note that we have only taken intrinsic spin-orbit coupling along one direction, and hence we have broken C3 symmetry. We now couple the two copies with the following term Hmix (t3) = −t3 c2† i,σc1 i,σ + c2† i,σc1 j,σ (cid:88) (cid:88) i,σ c2† i,σc1 j,σ + (cid:104)(cid:104)i,j(cid:105)(cid:105) y,σ (cid:88) (cid:104)i,j(cid:105),σ  , Figure 7. Bulk bands (a) and Wilson loop spectrum (b) for the C4 symmetric fragile topological insulator. Plots are for
1/t = 0.1,
2/t = −0.3, t2/t = 0.8, λ/t = 0.4. will consist of linear combinations of the three cases pre- sented here, and since the partial polarization is an ad- ditive quantity, we know how to calculate it for arbitrary NF . Appendix C: Topological insulator models of the fragile type with NF = 2 To show the existence of fragile topological insulating phases in two-dimensional crystals with two-fold rota- tion and time-reversal symmetries when only one pair of Kramers related pairs are occupied, we start by consid- ering the following two-band model of a Chern insulator with C = +2, H = {−t [1 + cos (ky) + cos (2kx)] − t2 cos (kx)} τx {−t [− sin (ky) − sin (2kx)] − t2 sin (kx)} τy − t3 sin (2kx) τz where t, t2, t3 are hopping amplitudes, and τi are Pauli matrices representing an internal degree of freedom. We now add its time-reversal partner, and couple them with (cid:26)(cid:20) − 1 2 (cid:20) 1 2 HR = −iλ − 1 2 sin (kx) + sin (ky) (cid:21) iτxsx (cid:27) + cos (kx) + cos (ky) iτysx , (cid:21) where λ is a hopping amplitude and si are Pauli ma- trices acting in spin-space. This model consists of two time-reversed copies of Chern insulators with Chern num- bers C = ±2, and C2 symmetry. Taken together, the Kane-Mele invariant is trivial but the Wilson loop spec- tra wind, indicating a fragile topological insulator pro- tected by C2Θ symmetry. The C2 symmetry operator is C2 = iτx ⊗ sz and the time-reversal operator Θ = UK, with U = I2 ⊗ isyand K is complex conjugation. The plots in the main text are for the parameters t/t2 = 0.4 t/t3 = −1.6 λ/t = 0.15. a)b)νE0π2π-12012k1ΓXMΓ-4-202413 and (cid:26) sin [i (k2 − k1)] + sin [ik1] k = − t2 H 1 4 −ρ sin [ik2]} τz ⊗ σy, where tσ and tπ are the hopping amplitudes for the σ and π pairing, t2 is the amplitude of next-nearest-neighbor hopping and σi and τi are Pauli matrices that act in or- bital and sublattice space respectively. α and ρ are two parameters we have introduced to break the C3 symme- try. For α = ρ = 1, the C3, symmetry is preserved and our Hamiltonian is equivalent to the one in Ref. [82]. We now take two copies of two copies of Hpxpy (k), where one copy has spin pointing in the positive x- direction, and the other spin pointing in the negative x-direction. In addition, we shift the momentum along the x-direction of the copies in opposite direction: H = Hpxpy (k − x
)←(cid:105)(cid:104)← + Hpxpy (−k + x
) + Hmix →(cid:105)(cid:104)← , ∗ →(cid:105)(cid:104)→ (D1) where the spins are mixed by Hmix = −iλ sin (kx) τ0 ⊗ σ0. This Hamiltonian has C2 and Θ symmetry, where C2 = −τ0σxeisxπ/2, and Θ = UK with U = −iτ0σ0sy and K complex conjugation, and τi, σi, si Pauli matrices acting in orbital-space, sublattice-space and spin-space respec- tively. Fig. 8 shows the bulk band spectrum and the Wilson loop spectrum. Figure 8. Bulk band spectrum (a) and Wilson loop spectrum (b) of the Hamiltonian Hpxpy for
= 0.5, tπ/tσ = 1.5, t2/tσ = 3.75, λ/tσ = 1.2 2 and t2 where (cid:104)(cid:104)i, j(cid:105)(cid:105)y denotes next-nearest neighbor hopping along the y-direction. When t1 2 have a differ- ent sign, this Hamiltonian can be in a fragile topolog- ical phase. The time-reversal operators is Θ = UK with U = iτ0σ0sy and the twofold rotation operator C2 = iτ0σxsz, where τi, σi and si are Pauli matrices that act in copy-space, sub-lattice space and spin-space respectively. The plots in the main text are made for 2/t1 = −0.9, the parameters t2/t1 = 1.1, t1
1/t1 = −
2/t1 = 0.1, λ1/t1 = λ2/t1 = 0.15, t3/t1 = 0.25. 2/t1 = 1.1, t2 A different way to construct a model exhibiting this fragile topological phase is by considering a model of px,y orbitals on a honeycomb lattice introduced in Ref. [82], (cid:32) (cid:33) Hpxpy (k) = 0 hk † h k 0 + H 1 k , with (cid:0)1 + α e−ik2 + e−ik1(cid:1) (tσ + tπ) (cid:18) (cid:19) (cid:0)−1 + e−ik1(cid:1) (tσ − tπ) σx, + α e−ik2 − 1 2 e−ik1 − 1 2 1 hk = 2 − 1 2 √ + 3 4 (tσ − tπ) σz [1] K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [2] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [3] B. I. Halperin, Phys. Rev. B 25, 2185 (1982). [4] M. Kohmoto, Ann. Phys. (N.Y.) 160, 343 (1985). [5] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). [6] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [7] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). [8] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buh- mann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). [9] H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang, Nat. Phys. 5, 438 (2009). [10] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). [11] C. Brune, C. Liu, E. Novik, E. Hankiewicz, H. Buhmann, Y. Chen, X. Qi, Z. Shen, S. Zhang, and L. Molenkamp, Phys. Rev. Lett. 106, 126803 (2011). [12] B. Rasche, A. Isaeva, M. Ruck, S. Borisenko, V. Zabolot- nyy, B. Buchner, K. Koepernik, C. Ortix, M. Richter, a)b)νEΓXMYΓ-4-20240π2π-12012k114 and J. Van Den Brink, Nat. Mat. 12, 422 (2013). [43] W. A. Benalcazar, B. A. Bernevig, and T. L. Hughes, [13] V. Mourik, K. Zuo, S. M. Frolov, S. Plissard, E. P. and L. P. Kouwenhoven, Science 336, 1003 Bakkers, (2012). [14] R. M. Lutchyn, J. D. Sau, and S. D. Sarma, Phys. Rev. Lett. 105, 077001 (2010). Physical Review B 96, 245115 (2017). [44] F. Schindler, Z. Wang, M. G. Vergniory, A. M. Cook, A. Murani, S. Sengupta, A. Y. Kasumov, R. Deblock, S. Jeon, I. Drozdov, et al., Nat. Phys. 14, 918 (2018). [45] M. Geier, L. Trifunovic, M. Hoskam, and P. W. Brouwer, [15] L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 Phys. Rev. B 97, 205135 (2018). (2008). [16] C. Beenakker, Ann. Rev. Cond. Mat. Phys. 4, 113 (2013). [17] J. Alicea, Rep. Prog. in Phys. 75, 076501 (2012). [18] N. Armitage, E. Mele, and A. Vishwanath, Rev. Mod. Phys. 90, 015001 (2018). [19] S.-M. Huang, S.-Y. Xu, I. Belopolski, C.-C. Lee, G. Chang, B. Wang, N. Alidoust, G. Bian, M. Neupane, C. Zhang, et al., Nat. Comm. 6, 7373 (2015). [20] B. Lv, H. Weng, B. Fu, X. Wang, H. Miao, J. Ma, P. Richard, X. Huang, L. Zhao, G. Chen, et al., Phys. Rev. X 5, 031013 (2015). [21] S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.-C. Lee, et al., Science 349, 613 (2015). [22] E. Haubold, K. Koepernik, D. Efremov, S. Khim, A. Fe- dorov, Y. Kushnirenko, J. van den Brink, S. Wurmehl, B. Buchner, T. Kim, et al., Phys. Rev. B 95, 241108 (2017). [46] C. W. Peterson, W. A. Benalcazar, T. L. Hughes, and G. Bahl, Nature 555, 346 (2018). [47] M. Serra-Garcia, V. Peri, R. Susstrunk, O. R. Bilal, T. Larsen, L. G. Villanueva, and S. D. Huber, Nature 555, 342 (2018). [48] E. Khalaf, Phys. Rev. B 97, 205136 (2018). [49] M. Ezawa, Phys. Rev. Lett. 120, 026801 (2018). [50] M. Ezawa, Phys. Rev. B 97, 155305 (2018). [51] J. Langbehn, Y. Peng, L. Trifunovic, F. von Oppen, and P. W. Brouwer, Phys. Rev. Lett. 119, 246401 (2017). [52] M. Sitte, A. Rosch, E. Altman, and L. Fritz, Phys. Rev. Lett. 108, 126807 (2012). [53] Z. Song, Z. Fang, and C. Fang, Phys. Rev. Lett. 119, 246402 (2017). [54] S. Imhof, C. Berger, F. Bayer, J. Brehm, L. W. Molenkamp, T. Kiessling, F. Schindler, C. H. Lee, M. Greiter, T. Neupert, et al., Nat. Phys. 14, 925 (2018). and S. Wan, arXiv preprint [55] Y. Xu, R. Xue, [23] A. Lau, K. Koepernik, J. van den Brink, and C. Ortix, arXiv:1711.09202 (2017). Phys. Rev. Lett. 119, 076801 (2017). [24] X. Wan, A. M. Turner, A. Vishwanath, Savrasov, Phys. Rev. B 83, 205101 (2011). [56] C.-H. Hsu, P. Stano, J. Klinovaja, and D. Loss, Phys. and S. Y. Rev. Lett. 121, 196801 (2018). [57] G. van Miert and C. Ortix, Phys. Rev. B 98, 081110 [25] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, (2018). 127205 (2011). [58] S. H. Kooi, G. Van Miert, and C. Ortix, Phys. Rev. B [26] A. A. Zyuzin, S. Wu, and A. A. Burkov, Phys. Rev. B 98, 245102 (2018). 85, 165110 (2012). [59] A. A. Soluyanov and D. Vanderbilt, Phys. Rev. B 83, [27] A. Lau and C. Ortix, Phys. Rev. Lett. 122, 186801 035108 (2011). (2019). [28] T. Ojanen, Phys. Rev. B 87, 245112 (2013). [29] A. A. Soluyanov, D. Gresch, Z. Wang, Q. Wu, M. Troyer, X. Dai, and B. A. Bernevig, Nature 527, 495 (2015). [30] A. Altland and M. R. Zirnbauer, Phys. Rev. B 55, 1142 (1997). [60] R. Yu, X. L. Qi, A. Bernevig, Z. Fang, and X. Dai, Phys. Rev. B 84, 075119 (2011). [61] B. Bradlyn, L. Elcoro, J. Cano, M. G. Vergniory, Z. Wang, C. Felser, M. I. Aroyo, and B. A. Bernevig, Nature 547, 298 (2017). [62] H. C. Po, A. Vishwanath, and H. Watanabe, Nature [31] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. Ludwig, Communications 8, 50 (2017). Phys. Rev. B 78, 195125 (2008). [63] J. Kruthoff, J. de Boer, J. van Wezel, C. L. Kane, and [32] A. Kitaev, in AIP Conference Proceedings, Vol. 1134 R.-J. Slager, Phys. Rev. X 7, 041069 (2017). (AIP, 2009) pp. 22 -- 30. [64] Z. Song, T. Zhang, Z. Fang, and C. Fang, Nat. Comm. [33] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 9, 3530 (2018). (2010). [65] T. Zhang, Y. Jiang, Z. Song, H. Huang, Y. He, Z. Fang, [34] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 H. Weng, and C. Fang, Nature 566, 475 (2019). (2011). [35] L. Fu, Phys. Rev. Lett. 106, 106802 (2011). [36] T. H. Hsieh, H. Lin, J. Liu, W. Duan, A. Bansil, and [66] M. Vergniory, L. Elcoro, C. Felser, N. Regnault, B. A. Bernevig, and Z. Wang, Nature 566, 480 (2019). [67] F. Tang, H. C. Po, A. Vishwanath, and X. Wan, Nature L. Fu, Nat. Comm. 3, 982 (2012). 566, 486 (2019). [37] J. Liu, T. H. Hsieh, P. Wei, W. Duan, J. Moodera, and [68] Z. Song, L. Elcoro, N. Regnault, and B. A. Bernevig, L. Fu, Nat. Mat. 13, 178 (2014). arXiv e-prints , arXiv:1905.03262 (2019). [38] T. H. Hsieh, J. Liu, and L. Fu, Phys. Rev. B 90, 081112 [69] H. C. Po, H. Watanabe, and A. Vishwanath, Phys. Rev. (2014). Lett. 121, 126402 (2018). [39] C. Fang and L. Fu, arXiv preprint arXiv:1709.01929 [70] B. Bradlyn, Z. Wang, J. Cano, and B. A. Bernevig, Phys. (2017). [40] E. Khalaf, H. C. Po, A. Vishwanath, and H. Watanabe, Phys. Rev. X 8, 031070 (2018). [41] F. Schindler, A. M. Cook, M. G. Vergniory, Z. Wang, S. S. Parkin, B. A. Bernevig, and T. Neupert, Sci. Adv. 4, eaat0346 (2018). [42] W. A. Benalcazar, B. A. Bernevig, and T. L. Hughes, Science 357, 61 (2017). Rev. B 99, 045140 (2019). [71] Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, and P. Jarillo-Herrero, Nature 556, 43 E. Kaxiras, (2018). [72] Y. Cao, V. Fatemi, A. Demir, S. Fang, S. L. Tomarken, J. Y. Luo, J. D. Sanchez-Yamagishi, K. Watanabe, T. Taniguchi, E. Kaxiras, et al., Nature 556, 80 (2018). [73] J. Ahn, S. Park, and B.-J. Yang, Phys. Rev. X 9, 021013 15 (2019). [74] H. C. Po, L. Zou, T. Senthil, and A. Vishwanath, Phys. Rev. B 99, 195455 (2019). [75] J. Zak, Phys. Rev. B 26, 3010 (1982). [76] A. Lau, J. van den Brink, and C. Ortix, Phys. Rev. B 94, 165164 (2016). Phys. Rev. Lett. 120, 266401 (2018). [83] A real gauge can be formulated as C2Θψ(cid:105) = ψ(cid:105). [84] T. Bzdusek and M. Sigrist, Phys. Rev. B 96, 155105 (2017). [85] J. Ahn, D. Kim, Y. Kim, and B.-J. Yang, Phys. Rev. Lett. 121, 106403 (2018). [77] G. van Miert and C. Ortix, Phys. Rev. B 96, 235130 [86] S. Franca, J. van den Brink, and I. C. Fulga, Phys. Rev. (2017). B 98, 201114 (2018). [78] L. Fu and C. L. Kane, Phys. Rev. B 74, 195312 (2006). [79] A. Alexandradinata, X. Dai, and B. A. Bernevig, Phys. Rev. B 89, 155114 (2014). [87] Since we have to take a region symmetrically centered around ν = 0, 1/2 we have to either include both or nei- ther of the blue bands. [80] J. Kruthoff, J. de Boer, and J. van Wezel, arXiv e-prints [88] P. G. Harper, Proceedings of the Physical Society. Section , arXiv:1711.04769 (2017). A 68, 874 (1955). [81] G. van Miert and C. Ortix, Phys. Rev. B 97, 201111 [89] S. Aubry and G. Andr´e, Ann. Isr. Phys. Soc. 3, 133 (2018). (1980). [82] J. Cano, B. Bradlyn, Z. Wang, L. Elcoro, M. G. Vergniory, C. Felser, M. I. Aroyo, and B. A. Bernevig, [90] S. Ganeshan, K. Sun, and S. Das Sarma, Phys. Rev. Lett. 110, 180403 (2013). 

1109.2941

1

1109

Fragility of multi-junction flux qubits against quasiparticle tunneling

We study decoherence in superconducting qubits due to quasiparticle tunneling which is enhanced by two known deviations from the equilibrium BCS theory. The first process corresponds to tunneling of an already existing quasiparticle across the junction. The quasiparticle density is increased, e.g., because of an effective quasiparticle doping of the system. The second process is quasiparticle tunneling by breaking of a Cooper pair. This can happen at typical energies of superconducting qubits if there is an extended quasiparticle density inside the gap. We calculate the induced energy decay and pure dephasing rates in typical qubit designs. Assuming the lowest reported value of the non-equilibrium quasiparticle density in Aluminum, we find for the persistent-current flux qubit decay times of the order of recent measurements. Using the typical sub-gap density of states in Niobium we also reproduce observed decay times in the corresponding Niobium flux qubits.

cond-mat.mes-hall

cond-mat

Fragility of multi-junction flux qubits against quasiparticle tunneling Institut fur Theoretische Festkorperphysik and DFG-Center for Functional Nanostructures (CFN), Karlsruhe Institute of Technology, D-76128 Karlsruhe, Germany Juha Leppakangas and Michael Marthaler We study decoherence in superconducting qubits due to quasiparticle tunneling which is enhanced by two known deviations from the equilibrium BCS theory. The first process corresponds to tunnel- ing of an already existing quasiparticle across the junction. The quasiparticle density is increased, e.g., because of an effective quasiparticle doping of the system. The second process is quasiparticle tunneling by breaking of a Cooper pair. This can happen at typical energies of superconducting qubits if there is an extended quasiparticle density inside the gap. We calculate the induced energy decay and pure dephasing rates in typical qubit designs. Assuming the lowest reported value of the non-equilibrium quasiparticle density in Aluminum, we find for the persistent-current flux qubit decay times of the order of recent measurements. Using the typical sub-gap density of states in Niobium we also reproduce observed decay times in the corresponding Niobium flux qubits. PACS numbers: 42.55.-f,85.25.Cp,03.65.Yz I. INTRODUCTION A basic building block of a superconducting quan- tum bit is the Josephson junction, which allows coher- ent Cooper-pair tunneling. This Josephson current pro- vides the necessary non-linear element in an electric cir- cuit and enables reduction of the externally controllable quantum dynamics to involve only two eigenstates1. A drawback for superconducting qubits has been that these systems can be very sensitive to various effects in their nearby environments. The influence of two major deco- herence sources, charge fluctuations and two-level fluc- tuators, have been substantially reduced in recent years. Qubits that are less sensitive to changes in background charge have been build2,3, while fluctuators have been re- moved by decreasing the junction size. Recently, trans- mons and persistent-current flux (p-flux) qubits with ex- tremely long decay times have been demonstrated4,5 . It is therefore of great interest to investigate mechanisms of decoherence that were previously unobservable. Temper- ature dependent decay time measurements in phase6 and transmon qubits7 suggest that non-equilibrium quasipar- ticles might have now become the main factor limiting the decay times of superconducting qubits. In this work we analyze qubit decay and dephasing due to quasiparticle tunneling. As the source of tunneling we consider the two following deviations from the equilib- rium BCS state of superconductors. The first one is a non-equilibrium distribution of quasiparticles, which cor- responds to a finite quasiparticle density above the gap, present at all temperatures8 -- 12. Non-equilibrium quasi- particles could originate in a quasiparticle diffusion from higher temperature regions in an experiment, or through stray radiation, and are relatively long lived due to slow quasiparticle recombination rate 13 -- 16. The second pro- cess we consider is quasiparticle tunneling by breaking of a Cooper pair, which can occur at typical energies of su- perconducting qubits if there is an extended quasiparticle density inside the gap. It has long been observed that the number of states inside the gap is vastly larger than the predictions of BCS theory17. This effect is especially pro- nounced in Niobium, a material that is widely used in the production of qubits18. Nonetheless, Aluminum based superconducting devices are observed to posses sub-gap states too10,19, but with considerably smaller density as Niobium. Our results are that the two quasiparticle processes result in a similar type of qubit decay rate. This pro- vides a clear comparison between the magnitudes of de- coherence due to the two sources. For conventional Alu- minum qubits non-equilibrium quasiparticles should have a more significant impact on decoherence, for Niobium qubits the sub-gap density of states. We also find that the sensitivity to quasiparticle tunneling is vastly differ- ent for differing qubit designs. By using the lowest re- ported density for non-equilibrium quasiparticles in Alu- minum10 we obtain p-flux qubit decay times similar to recent experiments5, whereas for other qubit types the results are from two to three orders of magnitude greater than the observed ones. For Niobium p-flux qubits we also reproduce usual experimental decay times by as- suming a Dynes-type of quasiparticle sub-gap states, ob- served in various other experiments20. These central re- Qubit type Transmon21 Phase6 Capacitively-shu. flux22 Persistent-current flux23 Non. QP (Al) Sup-gap DOS (Nb) 1.2 ms 0.6 ms 80 µs 12 µs 20 µs 14 µs 3 µs 0.3 µs TABLE I. Comparison of decay times obtained by assum- ing a non-equilibrium quasiparticle density nqp = 0.033/µm3 (second column, Aluminum) or a Dynes parameter γNb = ΓD/∆ = 10−2 (third column, Niobium). We used qubit pa- rameters specified in each of the citations at each qubit. We also assume ∆Al = ∆Nb/7 = 180 µeV and the validity of the Ambegaokar-Baratoff relation. Typical sub-gap density of states in Aluminum (γAl ∼ 10−4 − 10−7) does not limit the coherence here. sults are summarized in Table I. The paper is organized as follows. In section II we introduce the Hamiltonian and methods describing the qubit-quasiparticle dynamics. The section III is devoted to discussion of decoherence processes and the corre- sponding rate equations. We also discuss decay rates for single-junction qubits obtained by using typical experi- mental parameters of the quasiparticle environments. In section IV we generalize the treatment to multi-junction qubits, where a special concentration is given to flux qubits in their different designs. In section V we sum- marize the results. II. FOUNDATIONS Our starting point is the division of the total Hamilto- nian into three parts, Htotal = Hqubit + Hqp + HT. (1) Here the qubit Hamiltonian Hqubit describes the collec- tive degrees of freedom that constitute our two-level sys- tem. The quasiparticle Hamiltonian Hqp describes the electronic degrees of freedom in the superconductors, treated independently of the collective part. Interaction between these two parts emerges due to single-electron tunneling modeled by the tunneling Hamiltonian HT. It describes, e.g., quasiparticle tunneling across the Joseph- son junction which can cause transitions between the eigenstates of the qubit. In the following we discuss in more detail the central properties of each of the three parts. A. Single-junction qubit Hamiltonian As the general single-junction qubit Hamiltonian we inductive, and use the following sum of capacitive, Josephson coupling energy Hqubit = EC N 2 + EL 2 (ϕ − ϕext)2 − EJ cos ϕ. (2) Here N is the number operator of electron charges on the junction capacitor C. It is a conjugated variable to the phase difference across the junction ϕ, satisfying [ϕ, N/2] = i. This choice of qubit Hamiltonian corre- sponds to the phase and dc-flux qubits, i.e., when the junction is placed in a superconducting loop with induc- tive energy EL and thread by flux ϕext. For the trans- mon one has EL = 0 and periodic commutation relations [N/2, e±iϕ] = ±e±iϕ, related to the 2e-quantization of the charge that tunnels across the Josephson junction. All qubits we consider work in the limit where the Joseph- son coupling energy EJ is larger than the single-electron charging energy EC = e2/2C. Generalization to multi- junction qubits is straightforward, see section IV. 2 B. Quasiparticle degrees of freedom The quasiparticle degrees of freedom in superconduct- ing leads are modeled by the Hamiltonian Hqp. For all further calculations the quasiparticle properties are as- sumed to be independent of the qubit state. We consider here two models as two separated causes of quasiparticle tunneling at energies well below the superconducting gap ∆. We label these as model I and model II. Signatures of the electronic structure in the two models are found from correlation functions of type, G> αα(k, t) =Dckα(t)c† αβ(k, t) =Dc† kα(t)c† kαE , G< −kβE , F < kαckα(t)E , αα(k, t) =Dc† αβ(k, t) = hckαc−kβ(t)i . (3) F > Here c(†) kα is an electron annihilation (creation) operator of the state k with spin α (α 6= β). The Fourier transforms of the correlation functions are related to spectral and distribution functions as24 G≷ F ≷ ↑↓(k, ω) = −F ≷ αα(k, ω) = A(k, ω)f ±(ω), ↓↑(k, ω) = B(k.ω)f ±(ω). (4) Here f − = f and f + = 1 − f . In equilibrium f = feq(ω) = 1/[1 + exp(ω/kBT )]. We define the normalized density of states n and the pair density p as n(ω) = p(ω) = 1 πD Xk πD Xk 1 A(k, ω), B(k, ω). (5) Here D is the density of states nearby the Fermi surface, including spin. These functions, added with the distri- bution function f , have their own characteristic forms in the two models and are all we need to know about the quasiparticle environments in the final forms of the qubit decoherence rates (see section III). 1. Model I: Non-equilibrium quasiparticles In model I we assume that in each lead quasiparticles can be described by the BCS Hamiltonian but exist with a general (non-equilibrium) probability f . In the BCS model the quasiparticle and pair densities have the form n(ω) = p(ω) = Θ(ω2 − ∆2)sgn(ω), Θ(ω2 − ∆2)sgn(ω). (6) ω √ω2 − ∆2 √ω2 − ∆2 ∆ An important quantity is the total density of quasiparti- cles nqp = 2DZ ∞ ∆ n(ω)f (ω)dω. (7) In calculations nqp is assumed to be a given constant. This model has also been investigated in Refs. 11 and 12. III. DECOHERENCE RATES 3 2. Model II: Sub-gap density of states As the other source of quasiparticles we consider the presence of quasiparticle states below the gap. We as- sume that the densities can be expressed in the same form as for the BCS theory, but with the following broadening, ω + iΓD (8) ∆ + i∆2 p(ω + iΓD)2 − (∆ + i∆2)2) sgn(ω), p(ω + iΓD)2 − (∆ + i∆2)2) sgn(ω). n(ω) = Re( p(ω) = Re( Here ΓD ≪ ∆ is the so-called Dynes parameter25 and ∆2 ≪ ∆ is a possible imaginary part of the supercon- ducting gap24. We consider ΓD and ∆2 to be given pa- rameters and independent of energy. The effect of these modifications of the BCS theory are very similar nearby the gap, but differ at low energies ω ≈ 0 (see section III A 2). The broadening through an imaginary part of the superconducting gap26, ∆2 (ΓD = 0), is supported by the microscopic Eliashberg theory. However, a constant imaginary part of the energy, ΓD (∆2 = 0), usually fits better to experiments done with, e.g., Niobium20,25,27. For all further purposes we will only consider finite ΓD and ∆2 = 0. C. Quasiparticle tunneling The interaction between the qubit and the quasiparti- cle environments occurs due to the possibility of quasi- particles to tunnel across the junction(s) and simultane- ously change the charge configuration in the qubit space. This is described by the modified tunneling Hamiltonian HT HT = ¯HT + EJ cos ϕ, where the bare tunneling Hamiltonian has the form ¯HT = tXklα (c† kαclα T + h.c.). (9) (10) Here the states k and l belong to the opposite sides of the junction. We assume a constant tunneling ma- trix element t, related to the normal state tunnel resis- tance as RT = /t2D2πe2. The charge-transfer opera- tor T = PN N + 1ihN = eiϕ/2 accounts for the cor- responding changes in the charge number of the leads. Cooper-pair tunneling, which is a second-order process in quasiparticle tunneling, is already included in the qubit Hamiltonian in an approximative way as the operator −EJ cos ϕ. Therefore we need to subtract it from ¯HT to avoid double counting. The second-order expansion of the qubit's density- matrix equation of motion in the tunneling Hamiltonian leads to contributions describing qubit decoherence and parameter renormalization. Here we discuss in detail the decoherence terms, i.e., terms leading to energy decay and pure dephasing. We then estimate their magnitude for usual single-junction qubits. An exact derivation of the rates, a formulation for the energy-level renormaliza- tion effects, and a discussion for the effect of higher-order terms is given in the Appendix. The system we consider is a two-level system interact- ing with a fermionic bath. The coupling to the bath can be divided into two parts: (i) the part causing transi- tions, ∝ σx, and (ii) the part causing pure dephasing, ∝ σz. The decay rate, Γ1, is a result of coupling to σx. This leads also to the ordinary dephasing rate Γ2 = Γ1/2. Fluctuations through σz lead to the pure dephasing rate Γ2∗ , which can be interpret to be a result of low-frequency fluctuations in the qubit energy splitting due to coupling to the bath1. A. Tunneling processes In leading order we obtain two distinct types of tun- neling processes causing decoherence. The first one is tunneling of an existing quasiparticle. In the case of the BCS state this is described by an operator of the type O1 = ukul T − vkvl T †. (11) This process contributes mainly in model I. The second one is the breaking of a Cooper pair, described by the BCS operator of the type O2 = ukvl T + vkul T †. (12) This process contributes only in model II. Impor- tant here is that the operators Oi are superpositions of two electron-tunneling directions, T and T †, as long as u, v 6= 0. As the qubit states are superpositions of differ- ent charge states, the direction of electron tunneling be- comes indistinguishable and the two tunneling processes interfere. This gives rise to a phase dependence in quasi- particle tunneling, similar to the cos ϕ-term in the classi- cal Josephson effect28. This interference effect has also an important role in decoherence of superconducting qubits, as discussed below. 1. Qubit decay In the discussed low-energy approximation we obtain the following common form for the decay rates, 4 For the qubits considered in this work the qubit decay rate can be written in the form Γ1 = 2ξiM 2 1 RTe2 (1 + ǫi cos ϕ0) , i = I, II , (16) Γ1 = 2 RTe2 (cid:12)(cid:12)(cid:12)h↑ T ↓i(cid:12)(cid:12)(cid:12) 2Z ∞ −∞ dωZ ∞ −∞ × [n(ω)n(ω′) + p(ω)p(ω′) cos ϕ0] δ(ω − ω′ − δE). dω′f −(ω)f +(ω′) 2 . M 2 1 =(cid:12)(cid:12)(cid:12)h↑ T ↓i(cid:12)(cid:12)(cid:12) (13) Important here is that the rates are proportional to ξi, defined for the two cases as Here δE is the energy-level splitting of the qubit and the angle ϕ0 is defined as eiϕ0 = −h↑ T ↓ih↓ T ↑i (cid:12)(cid:12)(cid:12)h↑ T ↓i(cid:12)(cid:12)(cid:12) 2 . (14) The matrix elements depend on the choice of supercon- ducting qubit, but as we will discuss, they are in fact very similar for broad classes of qubits. Expressions such as (13) are often derived in the BCS excitation picture, where integration over only positive (excitation) energies emerges. In this semiconductor- type presentation negative energies appear as well. But the result of the two presentations is the same. The pos- itive integration region of both frequencies ω and ω′ cor- responds to tunneling of an existing quasiparticle to one direction, as the negative integration region to the other region. The contribution with positive ω but negative ω′ describes breaking of a Cooper pair during the single- electron tunneling. It can exist only if there is an ex- tended states below the gap (because δE < 2∆). The contribution with positive ω′ but negative ω contributes only if δE < 0 and would correspond to recombination of two quasiparticles with excitation of the qubit. 2. Qubit decay in the low-energy approximation To simplify equation (13) we assume δE ≪ ∆ and that the width of the quasiparticle distribution (∼ kBT ) is much smaller than the qubit splitting, δE. This means that for model I the distribution f is nonzero above the gap (ω ≥ ∆) practically only in a very narrow region [we have always f (−ω) = 1 − f (ω) = f +(ω)]. On the other hand, for the model II with Dynes broadening (and ∆2 = 0) one can approximate n = p = , ΓD ∆ ω ∆ ΓD ∆ . (15) For a finite imaginary part ∆2 and ΓD = 0 one would have n = (ω/∆)(∆2/∆) and p = (ω/∆)2(∆2/∆). However, in the following we assume a Dynes-type broadening20,25,27 when considering model II. ξI = nqpn(∆ + δE) D , ∆ (cid:19)2 ξII = δE(cid:18) ΓD . (17) The form Γ1 ∝ 1 + ǫi cos ϕ0 for the decay rate (16) is a result of the interference effect in quasiparticle tunnel- ing28 (see the discussion in section III A). Its magnitude and sign are given by ǫI = 1 1 + δE/∆ , ǫII = − 1 ∆ (cid:19)2 6(cid:18) δE . (18) For model I we have ǫI ≈ 1 whereas for model II ǫII ≈ 0. Physical interpretation for this is that the subgap states in model II are close to metallic states and do not show significant pair correlations (p ≈ 0), needed for the effect. 3. Pure dephasing In the leading-order expansion of the qubit's time evo- lution the pure dephasing appears as an extra decay of co- herent oscillations through transition terms proportional to h↑ T (†) ↑i or h↓ T (†) ↓i. The pure dephasing rate is then similar to the rate (13) by setting δE = 0, and changing the matrix element, Γ2∗ = M 2 −∞ −∞ M 2 2 dω′f −(ω)f +(ω′) dωZ ∞ RTe2 Z ∞ × [n(ω)n(ω′) + p(ω)p(ω′) cos ϕ0∗ ] δ(ω − ω′) , 2 =(cid:12)(cid:12)(cid:12)h↑ T ↑i − h↓ T ↓i(cid:12)(cid:12)(cid:12)  eiϕ0∗ = −   = −    2 2 . h↑ T ↑i (cid:12)(cid:12)(cid:12)h↑ T ↑i(cid:12)(cid:12)(cid:12) h↓ T ↓i (cid:12)(cid:12)(cid:12)h↓ T ↓i(cid:12)(cid:12)(cid:12) 2 Here we have assumed that which is the case for most of the qubits considered in this paper. The only exception is the flux qubit away from its symmetry point, discussed more detailed in section IV. The value of the integration in Eq. (19) depends on the distribution f , which depends on the model used. In model II we assume the equilibrium distribution (19) , (20) feq = 1/[1 + exp(ω/kBT )]. Then the pure dephasing corresponds to fluctuations originating from tunneling of thermalized (sub-gap) quasiparticles. For this case the integration can be easily evaluated giving (ΓD/∆)2kBT . This means that for small temperatures kBT ≪ δE the pure dephasing rate is much smaller than the decay rate. In model I the situation is more difficult due to the log- arithmic divergence of the energy integral at the energy gap. Similar singularities appear also in other properties of superconductors29, but stay finite due to finite lifetime effects or gap anisotropy. It is now crucial, which type of broadening is assumed. If one introduces a small imag- inary part ∆2 and accordingly redefines nqp in Eq. (7), then the integral in Eq. (19) remains small. As a result of this the pure dephasing rate stays small compared to the energy decay rate. Other methods for circumventing the divergence can produce larger results for the inte- gral. However, as we will discuss in the next section, pure dephasing is additionally suppressed by small tun- neling matrix elements. B. Discussion for single-junction qubits We will now discuss the qubit part of the decoher- ence rates in the case of single-junction qubits. Specifi- cally we consider the phase qubit, the transmon, the dc- flux qubit, the strongly anharmonic phase qubit30 and the fluxonium. Strictly speaking the fluxonium is not a single junction qubit, but we consider the integrated Josephson junction array simply as an inductance. For a detailed discussion of the Fluxonium including quasi- particle tunneling in the junction array see Ref. 12. The discussion can be structured along two major properties. One is the squared magnitude of the qubit matrix el- ement in Eq. (16) or of the difference between matrix elements in Eq. (19), divided by the tunneling resistance RT. The second is the phase difference ϕ0 in the inter- ference (cos ϕ0) term. When it comes to estimating the matrix elements qubits fall broadly in two classes. One class of qubits have eigenstates located in a single local minimum of the qubit's potential energy, U = EL 2 (ϕ − ϕext)2 − EJ cos ϕ . (21) The lowest eigenstates are then similar to that of a har- monic oscillator and they are symmetric or antisymmet- ric around the local minimum. Here the decay and de- phasing elements are in a good approximation given by M 2 1 = EC δE , M 2 δE(cid:19)2 2 =(cid:18) EC . (22) In this class we have clearly the transmon and the phase qubit. A qubit with a quartic potential as proposed in 5 Ref. 30 (strongly anharmonic phase qubit) has the same matrix element and despite its rather unusual potential, the fluxonium belongs to this group as well for ϕext = 0. The qubit that bucks the trend is the dc-flux qubit, which in fact is very similar to the fluxonium for ϕext = π. With its double well potential it is very different from all the other qubits. We will discuss the form of the matrix ele- ments in detail in the next section, but at the symmetry point they can be shown to have the approximative form M 2 M 2 1 = sin2(cid:16) ϕmin 2 (cid:17) , 2 = e−2ω′ϕ2 min, (23) where ϕmin is the solution to the transcended equation and ELϕmin − EJ sin ϕmin = 0 , ω′ =s EL − EJ cos(ϕmin) 2EC . (24) (25) The decay matrix element is the largest of the discussed qubits. Important is also that this qubit has relatively low tunneling resistance RT. Depending on the interfence effect it can be rather sensitive to quasiparticle tunneling. The qubits have various interference angles, as defined in Eq. (14). For the qubits that are similar to a har- monic oscillator this angle is given by the position of the local minimum. The potentials of the transmon and flux- onium qubits are symmetric around the phase ϕ0 = 0, whereas the potential of the phase qubit is also practi- cally symmetric, for certain ϕ0 6= 0. In the dc-flux qubit (with ϕext = π) the interference angle is given by ϕ0 = π, and the situation is the same for the stronlgy anharmonic phase qubit. This provides protection against qubit de- cay due to non-equilibrium quasiparticles, as then the interference between two electron tunneling directions is destructive. However, this protection is only partial be- cause for typical qubit splittings δE ≈ ∆/10 one has ǫI ≈ 0.9 (< 1). The common decay rate (16) shows that the two quasi- particle sources produce a similar type of decay rate. It is now easy to estimate which source should be domi- nant in typical experimental conditions. A crucial quan- tity is the dimensionless parameter ξi/∆, defined in Eq. (17). The division by the energy gap also accounts for changes in Eq. (16) due to change in the tunneling resistance, if the same qubit (with same EJ) is build from, for example, Niobium instead of Aluminum, as one has RT ∝ ∆. For an Aluminum superconductor the parameter is usually measured to have a value in the range ξI /∆ ∼ 10−8 − 10−5 (non-equilibrium quasi- particles) or ξII /∆ ∼ 10−15 − 10−9 (sub-gap density of states). This means that for Aluminum non-equilibrium quasiparticles have usually an impact that is several or- ders of magnitudes larger than that of sup-gap states. On the other hand, for Niobium using the typically ob- served Dynes parameter20,27 we obtain values ξII /∆ ∼ a) b) c) JS 3 2 y 1 J J 4 J4 x 2a 2b FIG. 1. (a) The original persistent-current flux qubit as pro- posed in Ref. 31. The p-flux qubit has three junctions, two large junctions with Josephson energy EJ and phase differ- ences ϕ1 and ϕ3. The smaller junctions has the Josephson energy EJS and the phase difference ϕ2. Through the loop a flux ϕy is applied. (b) For technical reasons many flux qubits actually have four junctions. The fourth junction has the Josephson energy EJ4 and the phase difference ϕ4. (c) To allow for more tunability often the smaller junction is re- place by a SQUID loop consisting of two junctions with phase differences ϕ2a, and ϕ2b 10−6 − 10−5, corresponding to the upper bound values for non-equilibrium quasiparticles in Aluminum. In table I we compare decay times for various qubits calculated by assuming the lowest reported value for the density of non-equilibrium quasiparticles in Aluminum, given in Ref. 10. For single-junction qubits (transmon and phase) we find decay times of the order of ms, which is two to three orders of magnitude higher than the best experimental observations. For the same qubits but build from Niobium the decay rate, Eq. (16), simplifies to Γ1 ≈ 1 CRT(∆/ΓD)2 , (26) being consistent with the result of the classical limit. In this case we find decay times of the order of 10 µs. The single-junction qubit that is the most sensitive to both quasiparticle sources is the dc-flux qubit, for which we obtain decay times of order 100 µs (Al) and 100 ns (Nb) (not listed in table I). The value for Aluminum includes the protection due to destructive interference which is approximately a factor of 10. All together, in these conditions the Aluminum based single-junction qubits are protected against non-equilibrium quasiparti- cles, most of them due to small EC/EJ-ratio. IV. MULTI-JUNCTION FLUX QUBITS In this section we generalize our analysis to multi- junction qubits. We consider in detail the most common of them, the flux qubit, which comes in many different shapes and forms. The original proposal, Fig. 1(a), is a qubit with three junctions31, but in general it is today used in many variations often with several extra junc- tions. For technical reasons the flux qubits have always been build with a fourth junction, Fig. 1(b), which in most modern flux qubits is of the same size as the two 6 large junctions. In another version of the flux qubit, the small junction has been replaced with a SQUID, meaning by a loop with two junctions, Fig. 1(c). This allows for more tunability. We will start this section with a discussion of the gen- eralization of the decoherence rates to many junctions. Then we will continue with calculating the rates for the p-flux qubit with three junctions and then later discuss the modification by the fourth junction and the tunable third junction. We end the section with the capacitively shunted flux qubit, which by its properties is shown to be similar to an anharmonic oscillator. A. Generalization of the decoherence rates As before we define our total Hamiltonian of the qubit and the quasiparticle environment as H = Hqubit + Hqp + HT, (27) where Hqubit is the Hamiltonian of the flux qubit, cou- pled via the tunneling Hamiltonian HT to a quasipar- ticle environment Hqp. With a total of n junctions we j=1 EJj cos ϕj, where the bare tunnel Hamiltonian is have HT = ¯HT +Pn Xj=1 ¯HT = n (c† jkαcjlα Tj + h.c.). tjXklα (28) jk and c(†) Here c(†) jl are electron annihilation (creation) op- erator of the states k and l on the two sides of the junction j with Josephson coupling EJj , and Tj is the correspond- ing qubit-space charge-transfer operator, given in differ- ent cases below. The quasiparticle environments in each of the leads are assumed to be identical and described by relations (3-8). In the leading order the decay rates can be written into a similar form as for the single-junction case, Eq. (13), and in the low-energy approximation (section III A 2) one gets Γtotal 1 = Γ1j = n Γ1j, Xj=1 2ξiM 2 1j RTje2 (1 + ǫi cos ϕ0j) , (29) 2 . M 2 1j =(cid:12)(cid:12)(cid:12)h↑ Tj ↓i(cid:12)(cid:12)(cid:12) Here RTj and ϕ0j stand for the tunneling resistance and the interference angle of the junction j. Similar general- ization applies also for the pure dephasing rate, Eq. (19). In the following we discuss the form of the matrix el- ements M 2 ij and the phases ϕ0j in the considered flux qubit realizations. B. The persistent-current flux qubit where we defined The first persistent-current flux qubit we consider has three Josephson junctions, see Fig. 1(a). Its Hamiltonian can be presented in the form Hqubit = EC(cid:18) 1 1 + 2αs N 2 + + N 2 −(cid:19) − EJ [2 cos(ϕ+) cos(ϕ−) + α cos(2ϕ+ − ϕext)] . (30) Here EC is the total charging energy of the system, EJ is the Josephson energy of the larger junctions and α = EJS/EJ is a parameter that determines the energy splitting of the qubit. In the following it is assumed that for the corresponding capacitive term it holds αs = α. The ratio of Josephson and charging energy for the p- flux qubit is generally given by EJ /EC ≈ 50. We have three junctions with phase differences ϕ1, ϕ2 and ϕ3, as illustrated in Fig. 1(a). The phases in the Hamiltonian are then defined by ϕ± = 1 2 (ϕ1 ± ϕ3) , ϕext = ϕy . (31) For each dynamical phase we have a conjugate charge variable, [ϕ±, N±/2] = i, and for each junction we have a tunneling operator given by T1 = ei(ϕ++ϕ−)/2 , T2 = ei(ϕext−2ϕ+)/2, T3 = ei(ϕ+−ϕ−)/2 . (32) ω′ =r(4α2 + 1) EJ 2EC . The state in the right well, Ri, has the same form, just centered at ϕ+,min. Under the same approximation we used to derive the potential given by Eq. (34) we can now find the coupling strength t between the two states, which is given by t 2 = hRHqubitLi (38) ≈ [−2 − 3α + 2α2 − 2ϕ2 1 + 2α × exp(−ω′ϕ2 +,min). +,min(4α2 + 1)]EJ Here we assumed ϕext = π and consider the contribution of the order of EJ. There is an additional contribution of the order of √EJEC and EC , but they remain small. Of- ten the coupling strength has been calculated using the WKB approximation. Our calculations underestimates the coupling strength but still gives a good order of mag- nitude approximation. and in the lowest order of δϕext we get We move away from symmetry like ϕext = π − δϕext Hqubit ≈ Hqubit(ϕext = π)+EJ(cid:18)−2αϕ+ + 3 (cid:19) δϕext. (39) If we now compare the energy of the minima with the energy of the minima for δϕext = 0 we find the energy difference 4αϕ3 + 12α√6√2α − 1 (8α − 1)3/2 7 (37) (41) (42) (43) (44) (45) 1. Eigenstates δǫ = EJ δϕext. (40) The potential of the p-flux qubit is given by U (ϕ−, ϕ+)/EJ = −2 cos(ϕ+) cos(ϕ−) − α cos(2ϕ+ − ϕext). (33) In the direction of ϕ− the system behaves similar to a harmonic oscillator that remains in the ground state. Therefore we will focus on the dynamics in the ϕ+ di- rection. Choosing ϕext = π (the symmetry point) and ϕ− = 0 we expand U for small ϕ+, U/EJ ≈ (1 − 2α)ϕ2 + +(cid:20) 2α 3 − 1 12(cid:21) ϕ4 +. (34) We have α & 1/2. This means that the eigenstates of the qubit are formed by symmetric and antisymmetric superpositions of the ground state of two wells. These ground states are centered around the minima ±ϕ+,min, Now we can write the Hamiltonian in the standard way of a two-state approximation for the states Li and Ri, Hqubit ≈ t 1 2 δǫ t −δǫ! . The basic form of the states is therefore given by ↑i = cos(cid:18) θ ↓i = − sin(cid:18) θ 2(cid:19)Li + sin(cid:18) θ 2(cid:19)Li + cos(cid:18) θ 2(cid:19)Ri, 2(cid:19)Ri, with tan θ = t/δǫ. Let us now consider an operator of the form T = eiϕ+/2. We get in a good approximation ϕ+,min = √6√2α − 1 √8α − 1 . (35) hL TLi = e−iϕ+,min/2 , hR TRi = eiϕ+,min/2, and In the left well the eigenstate is of the form hϕ+Li =(cid:18) ω′ π (cid:19)1/4 exp(cid:2)−ω′(ϕ+ + ϕ+,min)2/2(cid:3) , hL TRi = hR TLi = e−ω′ϕ2 +,min. (36) Using these equations we will be able to calculate the effect of quasiparticle tunneling on the flux qubits. 2. Qubit decay 3. Dephasing 8 The dephasing is characterized by the diagonal matrix elements of Tj, which can be shown to have the form To estimate the qubit decay rate we start from the transition element of the operator T = eiϕ+/2. This can be expressed in the form h↓ T ↑i = i sin θ sin(cid:16) ϕ+,min 2 (cid:17) + cos θe−ω′ϕ2 +,min.(46) From this general form we can deduce the results for the tunneling operators T1 and T3, 2 + sin θe−ω′ϕ2 h↑ T1 ↑i = h↑ T3 ↑i = −i cos θ sin(cid:16) ϕ+,min +,min + cos(cid:16) ϕ+,min (cid:17) . h↓ T1 ↓i = h↓ T3 ↓i = +i cos θ sin(cid:16) ϕ+,min +,min + cos(cid:16) ϕ+,min (cid:17) . − sin θe−ω′ϕ2 2 (cid:17) (cid:17) (49) (50) 2 2 The second junction has again an extra phase depen- dence, h↓ T1 ↑i = h↓ T3 ↑i = i sin θ sin(cid:16) ϕ+,min 2 + cos θe−ω′ϕ2 +,min, (cid:17) (47) hl T2 li = eiϕext/2× h±i cos θ sin (ϕ+,min) ± sin θe−ω′ϕ2 +,min + cos (ϕ+,min)i . (51) where we have used the fact that the system stays in the ground state in the direction ϕ−. Important here is that for the symmetry point one has θ = π/2 and the matrix elements (47) are purely imaginary. This is in contrast to the (single-junction) dc-flux qubit where the element is real. It follows that the interference angles in Eq. (29) are ϕ01 = ϕ03 = 0, i.e., one has constructive interference instead of destructive. The junction that is somewhat different is the second junction, for which the matrix element is given by h↓ T2 ↑i =h−i sin θ sin ϕ+,min + cos θe−ω′ϕ2 (48) The element (48) has an additional phase factor, eiϕext/2, and becomes real at the symmetry point ϕext = π. This corresponds to the case of the dc-flux qubit and destruc- tive interference. +,mini eiϕext/2. The interference effect has an important role in de- termining the sensibility to non-equilibrium quasiparticle tunneling. As discussed in section III B, the decay matrix element of this qubit type is the largest, but the single junction dc-flux qubit is protected by destructive interfer- ence. As a result of constructive interference for two junc- tions of the p-flux qubit, this protection is lost. It follows that even with the smallest observed non-equilibrium quasiparticle density10, corresponding to ξI /∆ ∼ 10−8, we obtain a decay time of the order 10 µs. This result coincides with recent experiments5. The presence of a term that does not change sign when changing the state leads effectively to a different angle ϕ0∗ for the two states ↑i and ↓i, and deviations from result (19). However, near the symmetry point such cor- rections stay small and one can neglect this effect. In this region the main contribution to pure dephasing comes from the junction 2. Using Eq. (51) one obtains then for the relevant dephasing matrix element h↑ T2 ↑i − h↓ T2 ↓i2 = e−2ω′ϕ2 +,min. (52) This is usually much smaller than the decay element. Combined with the discussion of section III B, this means that at the symmetry point the dephasing is limited only by the qubit decay. C. The four-junction flux Qubit For technical reasons the p-flux qubit is in fact always build with four junctions instead of the necessary three. The fourth junction has a charging energy EC4 and a Josephson energy EJ4. We can assume EJ4 ≫ EC4. The additional junction adds another dimension to the prob- lem, and we can approximate states along this dimension as states of a harmonic oscillator. Within this approxi- mation the total Hamiltonian can be written as Hqubit = 1 2 δEτz + g (m1τx + m2τz) (a† + a) (53) + ωa†a, Results for Niobium p-flux qubits is similar to dc-flux qubits: using typical subgap densities20 one obtains de- cay times of the order 100 ns, being also similar to recent experiments32. These results are compared with the ones of other qubit types in table I. where τi are the Pauli matrices acting on the eigenstates of the p-flux qubit (42), with the energy splitting δE = √δǫ2 + t2. These states are coupled with the coupling strength g = EJ α(2EC4/EJ4)1/4 to the fourth junction, modeled as a harmonic oscillator with frequency ω = √8EC4EJ4. The coupling is determined by the following matrix elements, m1 = h↑ sin (2ϕ+ − ϕext) ↓i, 2m2 = h↑ sin (2ϕ+ − ϕext) ↑i − h↓ sin (2ϕ+ − ϕext) ↓i . (54) If the fourth junction is large, EJ4 → ∞, the coupling goes to zero. In this case the fourth junction is decoupled and always stays in the ground state. However, in many flux qubits the fourth junction has the Josephson energy EJ4 ≈ EJ. In this case the coupling is strong, g ≈ ω. The tunneling operator for the fourth junction is sim- ply given by T4 = eiϕ4/2. Close to the symmetry point we have m2 ≈ 0 and additionally we assume ω ≫ δE. In this case we can approximate the matrix element in the lowest order of (2EC4/EJ4)1/4 by gm1 pω2 + 4g2m2 1 , (55) hgT4ei = −i(cid:18) 2EC4 EJ4 (cid:19)1/4 where we have labeled the two lowest eigenstates of the four-junction flux qubit as gi and ei. In the limit of weak coupling EJ4 ≫ EJ, we find ω ≫ gm1. Then the matrix element simplifies to hgT4ei = −iαEJm1/8EJ4. In general we have EC4/EJ4 ≪ 1 and therefore transi- tions due to tunneling across the fourth junction are well suppressed compared to tunneling across the other junc- tions. Similarly we can calculate the tunneling through the third junction with the tunneling operator T3 = ei(φ+−φ−)/2 and, using the same approximation as for Eq. (55), we get hgT3ei ≈ h↓ T3 ↑i . (56) In conclusion we can say that the fourth junction should have little effect on the overall decoherence rate of the p-flux qubit. 9 it is straightforward to calculate the tunneling matrix elements. The decay rates for quasiparticle tunneling through the large junctions are the same as for the stan- dard persistent-current flux qubit and we therefore get the similar overall rates. The only difference is that the tunneling across the small junctions, that form the SQUID, have now different type of interference effect if ϕx 6= 0. E. Capacitively shunted Flux qubit Another version of the same qubit, the capacitively- shunted flux qubit22 has the same Hamiltonian as the persistent-current flux qubit (30). The major difference is that it is operated in the regime α ≤ 1/2 and the charging energy along the dimension of the double well potential is significantly reduced by introducing a shunt capacitance, αs ≫ α. This qubit has excellent coherence times and still preserves relatively large anharmonicity. We will discuss the system at the symmetry point ϕext = π and for the case where it deviates most strongly from a harmonic oscillator, α = 1/2. After this we are able to generalize this result to other parameter regimes. Under the conditions we specified above we can write the Hamiltonian as Hqubit(ϕext = π, α = 1/2) ≈ E′ C N 2 + + 1 3 EJϕ4 +, (59) with E′ C = EC /(1 + 2α + Cs/CJ ). We note that the Hamiltonian is similar to the one of the strongly anhar- monic phase qubit30. We find the eigenstates to be hϕ+ ↓i =(cid:18) ω′ hϕ+ ↑i =(cid:18) ω′ e−ω′ϕ2 + , π (cid:19)1/4 π (cid:19)1/4 √2ω′ ϕ+ e−ω′ϕ2 +/2, (60) (61) D. Tunable gap flux qubit with ω′ = (3EJ/16E′ the matrix elements for the tunneling operators (32), C)1/4. It is now simple to calculate In the tunable gap flux qubit the small junction is re- placed by a SQUID, see Fig. 1(c). In this case the pa- rameter α can be tuned by changing the field ϕx. In Hamiltonian (30) this changes α = α0 cos(cid:16) ϕx 2 (cid:17) , ϕext = ϕy + ϕx 2 . (57) The four tunneling operators are given by T1 = ei(ϕ++ϕ−)/2 , T2b = ei(ϕext−ϕx/2−2ϕ+)/2 , T3 = ei(ϕ+−ϕ−)/2, T2a = ei(ϕext+ϕx/2−2ϕ+)/2, (58) where Ti corresponds to the junction with phase ϕi. Since the eigenstates are the same as for the p-flux qubit, h↓ T1 ↑i = h↓ T3 ↑i ≈ i C 2(cid:18) 2E′ 3EJ(cid:19)1/6 , (62) matrix element as h↓ T1 ↑i = pE′ and for T2 one obtains again the additional phase fac- tor eiϕext/2 = i. The energy splitting between the eigenstates is given in a good approximation by δE = 2EJ )1/3. Using this result we can rewrite the 2(12E′ C C /δE. This is a well known result for a harmonic oscillator and is therefore also valid for the complete relevant parameter regime of the capacitively shunted flux qubit. The interference an- gles ϕ0j of the three junctions are the same as for the p-flux qubit at the symmetry point: In the case of non- equilibrium quasiparticles and qubit decay the junctions 1 and 3 have constructive (ϕ01 = ϕ03 = 0) and the junc- tion 2 has destructive interference (ϕ02 = π). Vice versa for the dephasing. As mentioned, the c-flux qubit is very similar to the strongly anharmonic phase qubit, with the difference that the latter qubit has only a single junction, which corre- sponds to the second junction of the c-flux qubit. As this junction shows destructive interference against qubit de- cay due to non-equilibrium quasiparticles, the strongly anharmonic phase qubit is better protected against this decoherence source. However, the matrix elements of the qubits are generally similar to that of harmonic oscilla- tors, and this means that the c-flux qubit is relatively well protected against quasiparticle tunneling, too. In table I we compare the qubit decay times between the capacitively shunted flux and the other discussed qubits. V. CONCLUSION In this work we analyzed decoherence in supercon- ducting qubits due to quasiparticle tunneling. We con- sidered two types of sources of quasiparticle tunneling relevant for low temperatures and low qubit energies: non-equilibrium quasiparticles and sub-gap density of states. Using typically observed values for their densi- ties we estimated the resulting qubit decay times and compared them with the experimentally measured ones. In the best case scenario, i.e., with the lowest reported quasiparticle densities, we showed that only the decay times of persistent-current flux qubits were similar to re- cent experiments. The multi-junction flux qubits have achieved recently excellent coherence times and have the remarkable feature of extremely large anharmonicity which makes it the best approximation to a real two level system of all superconducting qubits. However, the pre- sented analysis shows that such qubits are also very frag- ile against quasiparticle tunneling induced decoherence. To protect the qubits against creation of non-equilibrium quasiparticles, for example, a careful isolation from the nearby environments should be realized10,13. ACKNOWLEDGMENTS We thank A. Heimes, W. D. Oliver, P. Kotetes, G. Schon, J. Liesenfeld and G. Johansson for useful dis- cussions. This work was supported by the CFN of DFG and the U.S. ARO under Contract No. W911NF-09-1- 0336. 10 where the superconductor quasiparticle density of states at a given energy cannot be mapped back into certain quasiparticle energy in the normal state, for example, due to energy-level broadening effects. We consider a single Josephson junction qubit and write the time evolution of the reduced density matrix, when interacting with quasiparticle environment, gener- ally as ρmn(t) = i(En − Em) ρmn  +Xab Z t t0 dt′σa→m b→n (t − t′)ρab(t′). (63) Here the (generalized transition rate) tensor σa→m b→n in- cludes the effect of qubit-quasiparticle interaction. Its calculation is similar to the case of metallic reservoirs33 with a difference that when tracing out the environment also two annihilation (creation) operators c(†) can also k contract to pairs. Their contribution leads to the pair density p, whereas diagonal contributions, such as c† kck, lead to the density of states n. In the leading order we obtain for the transition from σi→f the state ii to fi s→0Z t i→f = 2Re(cid:26)lim ¯σ(t − t′) =(cid:10)hi ¯HT(t)fi(cid:10) f ¯HT(t′)iii −∞ dt′e−(t−t′)s ¯σ(t − t′)(cid:27) , The trace over the initial distribution of quasiparticles leads to four nonvanishing contributions (64) ¯σ(t) = t2 2 eiδωt(a + b + c + d), (65) where δω = δE/ and G> αα(k, t − t′)G< G< αα(k, t′ − t)G> αα(l, t′ − t), αα(l, t − t′), a = hf Tii2Xklα b = hf T †ii2Xklα c = −hi Tfihf TiiXklαβ d = −hi T †fihf T †iiXklαβ F > αβ(k, t − t′)F < αβ(l, t′ − t), F < αβ (k, t′ − t)F > αβ (l, t − t′). (66) APPENDIX The correlation functions G and F are defined in Eq. (3). We define their Fourier transforms as A. Derivation of the decoherence rates G≶(k, t) = 1 2π Z ∞ −∞ dωe−iωtG≶(k, ω). (67) Here we formulate the qubit decay rate as a function of electron correlation (Green's) functions of the leads. Our aim is to consider the response of the superconductor as a sum over all energy states, rather than consider the response of single states. By this we can model situations The Fourier tranforms are related to the spectral densi- ties as in Eq. (4). In our analysis we have omitted the overall phase appearing usually in F -functions, as it is treated in the qubit part. In compared to Refs. 24 and 34 our definition of F > corresponds to F > (and F < to F <). Provided by the assumption of constant tunneling am- plitude t we can now write the contributions to the tran- sition rate coming from the different terms as 11 D2t2 2 hf Tii2 lim s→0 a → Re(cid:26)eiδωt′Z t −∞ dt′e−(t−t′)sZ ∞ −∞ dωZ ∞ −∞ dω′n(ω)n(ω′)f +(ω)f −(ω′)ei(ω′−ω)t′(cid:27) . (68) iP (1/(ω)), one obtains We have defined the normalized density of states n as in Eq. (5). Using the relation lims→0R 0 dω′f +(ω)f −(ω′)n(ω)n(ω′)δ(ω − ω′ − δω). a → D2πt2hf Tii2Z ∞ −∞ dωZ ∞ −∞ −∞ eiωt+st = πδ(ω) − (69) Note the symmetry of the equation around ω = 0: contri- bution of negative energies (ω, ω′ < 0) gives similar con- tribution as positive energies (ω, ω′ > 0). As all processes described by a correspond to given electron tunneling di- rection, negative energies correspond to opposite tunnel- ing direction of the quasiparticle. The contribution from b follows the same calculation and is similar to a, with a difference that the qubit matrix element is changed to hf T †ii2. It corresponds to tunneling of an electron to the opposite direction. The contribution from c and d exists only if a factor hi Tfihf Tii = hi T †fi(hf T †ii)∗ is finite. It means that it exists if two electron tunneling directions can lead to the same final state, in a single quasiparticle tunneling process. Then the processes interfere. The contribution from c is similar as before, but replaces n functions by p functions, c → − D2πt2 2 hf Tii2Re(cid:26)eiϕ0Z ∞ −∞ dωZ ∞ −∞ dω′f +(ω)f −(ω′)p(ω)p(ω′)(cid:20)δ(ω − ω′ − ωf i) − i P ω − ω′ − ωf i(cid:21)(cid:27) , (70) where we have used the fact that for the considered qubits hf Tii = hf T †ii and then defined hi Tfihf Tii = −eiϕ0hf Tii2. Noticing, that the contribution from d is equal to contribution from c, except with an opposite phase factor e−iϕ0, one obtains that the principal value part gives no contribution. Therefore one gets c + d → 2D2πt2 2 hf Tii2 cos ϕ0Z ∞ −∞ dωZ ∞ −∞ dω′f +(ω)f −(ω′)p(ω)p(ω′)δ(ω − ω′ − ωf i). (71) We can write now a + b + c + d → 2πt2 2 hf Tii2Z ∞ −∞ dωZ ∞ −∞ dω′f +(ω)f −(ω′) [n(ω)n(ω) + cos ϕ0p(ω)p(ω′)] δ(ω − ω′ − ωf i), (72) leading to equation (13). This expression is similar to equation (9) in Ref. 34, giving also insight to the inter- pretation of the different terms. The main term that is missing, when compared to Ref. 34, is the coherent Josephson term (sin ϕ-term), as it does not contribute in the process considered here. (This term is treated exactly in the qubit Hamiltonian). Such coherent terms would contribute through the principal value integration. The dissipative terms are similar. Especially, the famous cos ϕ-term corresponds to interference between electron and hole-like tunneling in our calculation, described by terms c and d. A new type of measurement of this term with a superconducting charge qubit has been considered in Ref. 28. B. Coherent terms: Parameter renormalization Taking into account the tunneling Hamiltonian HT by perturbation theory includes not only incoherent pro- cesses but also coherent terms leading to renormaliza- tion of the qubit parameters. Their contribution is usu- ally small but can depend, for example, on the non- equilibrium quasiparticle density. This has been studied in detail in Ref. 12. The Josephson coupling EJ in Eq. (2) is calculated as the expectation value over the quasiparticle distribution of the operator EJ 2 − =(cid:28)hN + 2 ¯HT 1 Hqp ¯HTNi(cid:29) . (73) Here the quasiparticle (BCS) Hamiltonian has no depen- dence on the charge number N . Using fermi distribu- tions for the excitation occupation probabilities one ob- tains the famous Ambegaokar-Baratoff relation for EJ. This corresponds to the usual choice of the Josephson coupling. In addition, there exists a similar second-order contribution describing reactive behaviour of quasiparti- cles (usually referred to as the capacitance renomaliza- tion) HC′ =XN NihN ¯HT 1 Hqp ¯HTNihN. (74) To calculate this one can introduce a cut-off function D(E) = 1/[1 + (E/Eco)2], where Eco is larger than any of the relevant energy scales in the system introduced to avoid divergence of the energy corrections. At this point this term produces a constant energy shift for all N and therefore does not bring anything new on the system. Insted of including these terms into the time- dependent problem, it is easier to estimate their effect by considering energy-level changes using time-independent perturbation theory. Using the second-order theory again, but now for the modified tunneling Hamiltonian HT, one obtains a correction to the energy of the state i δH J ii = EJhi cos ϕii +(cid:28)hi ¯HT(cid:20) 1 Ei − Hqp − Hqubit(cid:21) ¯HTii(cid:29) , (75) 12 where Ei is the energy of the qubit states ii. At this point we see that, for example, the Ambegaokar-Baratoff result is a good approximation as long as the approxima- tive Josephson term EJ cos ϕ and the corresponding term coming from the second term on the right-hand side of Eq. (75) almost cancel each other. It occurs if the tran- sition probabilities to virtual states vi (v 6= i) are small, or if the energies of the virtual states stay small Ev ≪ ∆. In the opposite case extra anharmonicity (non-constant energy-level spacing) could emerge due to this correction, as the effective tunneling coupling can become different for different states. C. Decoherence due to higher-order effects: Andreev tunneling The decoherence processes I and II correspond to leading order (incoherent) tunneling effects in the system. The most important process in higher orders is usually the Andreev tunneling. This involves, for example, tun- neling of two electrons from one side, with leaving two excitation behind, to form a Cooper pair on the other side. Such processes dominate the sub-gap conductance of normal metal-insulator-superconductor junctions for ideal BCS state on the superconductor side. However, for the case of perfect SIS junctions no such process exist until the energy 2∆ is somehow provided for creation of two excitations. The process appears as a step-like be- haviour in the current-voltage characteristics of single JJs nearby eV = ∆. In experiments the height of the step is usually considerably higher than obtained by theory. The reasons for this are still unclear35. For the case of superconducting qubits no energy is available to create two excitations, unless sub-gap den- sity exists at one side of the junction. Through creation of excitations into subgap region with creating a Cooper- pair on the other side of the junction one obtains a pos- sible contribution to the decoherence. However, a simple analysis indicates that the rates for such processes are proportional to (Γ/∆)2/N , where N ≫ 1 is the effective number of parallel tunneling channels. It results that contribution from this process should be much smaller than of the process II considered in this paper. We con- clude that the contribution from higher-order tunneling effects to decoherence in superconducting qubits should stay small. 1 Y. Makhlin, G. Schon and A. Shnirman, Rev.Mod. Phys. 73, 357 (2001) 2 J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schus- ter, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin and R. J. Schoelkopf, Phys. Rev. A 76, 042319 (2007). 3 V. E. Manucharyan, J. Koch, L. I. Glazman and M. H. Devoret, Science 326, 113 (2009). 4 H. Paik, D. I. Schuster, L. S. Bishop, G. Kirchmair, G. Catelani, A. P. Sears, B. R. Johnson, M. J. Reagor, L. Frunzio, L. Glazman and R. J. Schoelkopf, arxiv/1105.4652 5 J. Bylander, S. Gustavsson, F. Yan, F. Yoshihara, K. Harrabi, G. Fitch, D. G. Cory, Y. Nakamura, J.-S. Tsai and W. D. Oliver Nat. Phys. 7, 565 (2011). 6 J. M. Martinis, M. Ansmann and J. Aumentado, Phys. Rev. Lett. 103, 097002 (2009). 7 A. Palacios-Laloy, F. Mallet, F. Nguyen, F. Ong, P. Bertet, D. Vion and D. Esteve, Phys. Scr. T137, 014015 (2009). 8 B. S. Palmer, C. A. Sanchez, A. Naik, M. A. Manheimer, J. F. Schneiderman, P. M. Echternach, and F. C. Well- stood, Phys. Rev. B 76, 054501 (2007). 13 9 M. D. Shaw, R. M. Lutchyn, P. Delsing, and P. M. Echter- nach, Phys. Rev. B 78, 024503 (2008). 10 O.-P. Saira, A. Kemppinen, V. F. Maisi, and J. P. Pekola, arXiv:1106.1326. 11 G. Catelani, J. Koch, L. Frunzio, R. J. Schoelkopf, M. H. Devoret and L. I. Glazman, Phys. Rev. Lett. 106, 077002 (2011). 12 G. Catelani, R. J. Schoelkopf, M. H. Devoret and L. I. Glazman, arXiv:1106.0829. 13 R. Barends, J. Wenner, M. Lenander, Y. Chen, R. C. Bial- czak, J. Kelly, E. Lucero, P. O'Malley, M. Mariantoni, D. Sank, H. Wang, T. C. White, Y. Yin, J. Zhao, A. N. Cleland, J. M. Martinis and J. J. A. Baselmans, arXiv:1105.4642 14 G. Catelani, L. I. Glazman and K. E. Nagaev, Phys. Rev. B 82, 134502 (2010) 15 M. Lenander, H. Wang, Radoslaw C. Bialczak, Erik Lucero, Matteo Mariantoni, M. Neeley, A. D. O'Connell, D. Sank, M. Weides, J. Wenner, T. Yamamoto, Y. Yin, J. Zhao, A. N. Cleland and J. M. Martinis, Phys. Rev. B 84, 024501 (2011). 16 C. S. Owen and D. J. Scalapino, Phys. Rev. Lett. 28, 1559 (1972). skinen, and G. S. Paraoanu, Phys. Rev. B76, 172505 (2007). 21 J. Majer, J. M. Chow, J. M. Gambetta, J. Koch, B. R. Johnson, J. A. Schreier, L. Frunzio, D. I. Schus- ter, A. A. Houck, A. Wallraff, A. Blais, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Nature 449, 443 (2007). 22 M. Steffen, S. Kumar, D. P. DiVincenzo, J. R. Rozen, G. A. Keefe, M. B. Rothwell and M. B. Ketchen, Phys. Rev. Lett. 105, 100502 (2010). 23 I. Chiorescu, Y. Nakamura, C. J. P. M. Harmans, and J. E. Mooij, Science 299, 1869 (2003). 24 A. Barone and G. Paterna, Physics and Applications of the Josephson Effect (Wiley, New York, 1982). 25 R. C. Dynes, J. P. Garno, G. B. Hertel, and T. P. Orlando, Phys. Rev. Lett. 53, 2437 (1984). 26 B. Mitrovic and L. A. Rozema, J. Phys.:Cond. Matt. 20, 015215 (2008). 27 Yu. A. Pashkin, H. Im, J. Leppakangas, T. F. Li, O. Astafiev, A. A. Abdumalikov Jr., E. Thuneberg, and J. S. Tsai, Phys. Rev. B83, 020502(R) (2011). 28 J. Leppakangas, M. Marthaler and G. Schon, Phys. Rev. B 84, 060505(R) (2011). 17 K. K. Likharev, Dynamics of Josephson Junctions and Cir- 29 M. Tinkham, Introduction to Superconductivity, 2nd cuits (OPA, Amsterdam, 1986). 18 M. W. Johnson, M. H. S. Amin, S. Gildert, T. Lanting, F. Hamze, N. Dickson, R. Harris, A. J. Berkley, J. Johansson, P. Bunyk, E. M. Chapple, C. Enderud, J. P. Hilton, K. Karimi, E. Ladizinsky, N. Ladizinsky, T. Oh, I. Perminov, C. Rich, M. C. Thom, E. Tolkacheva, C. J. S. Truncik, S. Uchaikin, J. Wang, B. Wilson and G. Rose, Nature 473, 194 (2011). 19 J. P. Pekola, V. F. Maisi, S. Kafanov, N. Chekurov, A. Kemppinen, Yu. A. Pashkin, O.-P. Saira, M. Motto- nen, and J. S. Tsai, Phys. Rev. Lett. 105, 026803 (2010). 20 J. J. Toppari, T. Kuhn, A. P. Halvari, J. Kinnunen, M. Le- ed. (McGraw-Hill, New York, 1996). 30 A. B. Zorin and F. Chiarello, Phys. Rev. B 80, 214535 (2009). 31 T. P. Orlando, J. E. Mooij, L. Tian, C. H. van der Wal, L. S. Levitov, S. Lloyd and J. J. Mazo, Phys. Rev. B 60, 15398 (1999). 32 private communication, M. D. Oliver. 33 H. Schoeller and G. Schon, Phys. Rev. B 50, 18436 (1994). 34 V. Ambegaokar and A. Baratoff, Phys. Rev. Lett. 10, 486 (1963). 35 T. Greibe, M. P. V. Stenberg, C. M. Wilson, T. Bauch, V. S. Shumeiko, and P. Delsing, Phys. Rev. Lett. 106, 097001 (2011). 

1704.08877

1

1704

Negative Poisson's Ratio in Phagraphene

We present the results of numerical simulation of elastic properties of phagraphene, a recently predicted but not synthesized yet quasi-two-dimensional allotrope of graphene. We show that the Poisson'ss ratio is positive for the planar configuration of phagraphene and negative for the nonplanar one. Both the Poisson's ratio and the Young's modulus are isotropic for the planar phagraphene and strongly anisotropic for the nonplanar phagraphene.

cond-mat.mes-hall

cond-mat

Negative Poisson's Ratio in Phagraphene L.A. Openov and A.I. Podlivaev National Research Nuclear University "MEPhI", Moscow 115409, Russian Federation E-mail address: [email protected] ABSTRACT We present the results of numerical simulation of elastic properties of phagraphene, a recently predicted but not synthesized yet quasi-two-dimensional allotrope of graphene. We show that the Poisson's ratio is positive for the planar configuration of phagraphene and negative for the nonplanar one. Both the Poisson's ratio and the Young's modulus are isotropic for the planar phagraphene and strongly anisotropic for the nonplanar phagraphene. 1. INTRODUCTION The response of isotropic materials to mechanical stress is completely described by two scalar quantities, the Young's modulus Y, which characterizes the ability of a material to withstand longitudinal tension or compression, and the Poisson's ratio ν, which is equal to the ratio of the transverse compressive strain to the longitudinal tensile strain [1]. The elastic properties of anisotropic solids are described using a more sophisticated mathematical apparatus: In general case, the equations of the linear theory of elasticity contain the tensor of elastic moduli. Nevertheless, often it is possible to restrict ourselves to just two parameters, Y and ν. In particular, for a quasi-two-dimensional graphene monolayer [2], the following values of the Young's modulus and the Poisson's ratio have 1 been reported in the literature: Y ~ 1.0 TPa [3, 4] and ν = 0.15-0.45 (see [5] and references therein). For the majority of materials, the Poisson's ratio is positive. However, there are some exceptions, including fractal structures and other heterogeneous media (see [6] and references therein). Recently, it has been shown that for sufficiently narrow weakly deformed graphene nanoribbons the Poisson's ratio ν can be negative and reach the value of ~‒1.5 [7]. The reason for this effect is that compressive edge stresses lead to the formation of transverse displacement waves localized near the boundaries of the sample [8]. When the sample is stretched in the longitudinal direction, the wavy portions are straightened, and the sample is expanded (rather than compressed) in the transverse direction, which leads to a negative value of ν. As shown in Ref. [9], phagraphene – a recently predicted planar allotrope of graphene [10] – is unstable with respect to transverse atomic displacements, which generates the transverse displacement waves with an amplitude of ~1 Å and results in the formation of a nonplanar configuration of phagraphene [9, 11]. By analogy with graphene nanoribbons [7, 8], it can be expected that the Poisson's ratio for a nonplanar phagraphene will be negative. It should be noted, however, that there is a difference between graphene ribbons and phagraphene. While the negative Poisson's ratio for graphene ribbons is determined by boundary effects, in phagraphene it is caused by bulk effects (transverse displacement waves propagate throughout the entire sample). The purpose of our work is to perform numerical calculations of the Poisson's ratio for planar and nonplanar configurations of phagraphene, as well as to calculate the Young's modulus for these configurations. 2 2. COMPUTATIONAL DETAILS The planar phagraphene was simulated by a rectangular supercell consisting of 4 x 4 = 16 primitive 20-atom cells (Figs. 1a, 1b). In this supercell, the formation of transverse displacement waves led to a decrease in the energy of the supercell. The minimum energy was observed in the supercell with two waves (Figs. 2a, 2b), which we used to simulate the nonplanar configuration of phagraphene. We chose the periodic boundary conditions in both directions within the (XY) plane and the free boundary conditions in the transverse direction Z. Figure 1. A 320-atom supercell for the planar configuration of phagraphene. Top view (a) and side view (b). 3 The energy of the supercell with specified atomic coordinates was calculated in the framework of the nonorthogonal tight-binding model [12], which takes into account all four valence orbitals of each carbon atom and has been shown to work well for the simulation of graphene and other carbon structures (see [9, 11, 13] and references therein). (a) Figure 2. The same as in Figure 1, for the nonplanar configuration of phagraphene. The energy of this supercell is by 3.1 eV (~0.01 eV/atom) lower than that of the planar supercell. 4 3. RESULTS AND DISCUSSION We start with formulas for the calculation of the Poisson's ratio ν and the Young's modulus Y for quasi-two-dimensional materials which take into account the possible anisotropy of these quantities in the (XY) plane [14]. Let us assume that the sample has a rectangular shape with initial length L0 (along the X axis) and width W0 (along the Y axis). If, after stretching the sample along the X axis, its length and width become equal to L and W, respectively, the Poisson's ratio in the Y direction can be calculated according to the formula , (1) where X = (L-L0)/L0 and Y = (W-W0)/W0 are the relative strains of the sample in the X and Y directions, respectively. Similarly, the Poisson's ratio in the X direction after stretching the sample along the Y axis is determined as . (2) In our case, the length and the width of the sample are equal to the periods of the supercell a and b, respectively. It should be noted that, when calculating the Poisson's ratio νYX, the strain εX of the sample is specified as an input parameter, while the strain εY is found by minimization of the supercell energy, whereas in the calculation of νXY, on the contrary, the strain εY is fixed, while the strain εX is an output parameter and should be determined. Generally, the Poisson's ratio ν of the sample depends on the strain [7]. However, we restrict ourselves to the case of small strains and, for certainty, take the input parameters εX and εY to be equal to 0.001 in all calculations. 5 /YXYX/XYXYUnder weak deformations, the Young's moduli in the X and Y directions are calculated, respectively, according to the following formulas (similar to the case of nanotubes [15]): , (3) where S = ab is the area of the supercell, ΔE is the increment of the energy of the supercell under the corresponding deformation, and d is the thickness of the monolayer. Here it should be kept in mind that, the concept of the thickness of a sample does not have a clear meaning for quasi-two-dimensional systems, Nevertheless, this quantity has often been used to express the Young's modulus in conventional units (Pa). In particular, for graphene and carbon nanotubes, the sample thickness is usually taken to be 3.35 Å [3, 15], i.e., the distance between the adjacent graphene layers in graphite. We also set d = 3.35 Å in expressions (3). This makes it possible to compare the mechanical stiffnesses of phagraphene and graphene. For the verification of our computational algorithms and the subsequent comparison with the available data on graphene, in the first stage we calculated the Poisson's ratio ν and the Young's modulus Y in single-layer graphene making use of both rectangular and rhombic supercells. We found νXY = νYX = 0.35 and YX = YY = 0.97 TPa, which agree well with the values ν = 0.34 [7] and Y = 1.0 TPa [3] present in the literature. Despite the apparent anisotropy of the planar configuration of phagraphene (Fig. 1a), we obtained for it (as for graphene) equal (within the limits of computational error) values of the Poisson's ratios νXY = νYX = 0.38 and very close values of the Young's moduli YX = 0.84 TPa and YY = 0.86 TPa. Thus, elastic characteristics of the planar phagraphene 6 2222,XYXYEEYYSdSddiffer little from graphene. The slightly less (by ~15%) value of the Young's modulus for the planar phagraphene, as compared to graphene, is apparently due to the fact that the planar phagraphene contains pentagons and heptagons formed by the C-C bonds, which are absent in graphene and which make the two-dimensional crystal lattice more "soft." For practical applications, this is not very important. Indeed, of primary interest are not elastic but electronic characteristics of planar phagraphene, which are determined by the presence of the so-called Dirac cones in its band structure [10]. The anisotropy of the atomic structure of the nonplanar phagraphene in the (XY) plane is associated with the presence of transverse displacement waves in it, the minima and maxima of which alternate along the X axis (Fig. 2b). This gives rise to a strong anisotropy of the Poisson's ratio ν and the Young's modulus Y after stretching the sample along the X and Y axes. Thus, we obtained the Poisson's ratios νYX = -0.04 and νXY = -0.50 and the Young's moduli YX =0.06 TPa and YY = 0.75 TPa. Note that, despite the almost tenfold difference in the absolute values of νXY and νYX, both these quantities are negative, whereas they are positive in the planar phagraphene. The reason for the abnormally low Young's modulus of the nonplanar phagraphene along the X axis lies in the corresponding orientation of the transverse displacement waves softening the sample in this direction (Fig. 2b). The Young's modulus of the nonplanar phagraphene along the Y axis is an order of magnitude larger, though slightly smaller, than that in graphene and planar phagraphene. 7 4. CONCLUSIONS The calculations of the Poisson's ratio and the Young's modulus of phagraphene show that its elastic characteristics are isotropic in the planar configuration and strongly (tenfold) anisotropic in the nonplanar one. The main result is that the Poisson's ratios in the planar and nonplanar phagraphenes have different signs, regardless of the direction of the deformation. This can be used in practice to determine the type of atomic configuration of phagraphene samples. Acknowledgments. This work was supported by the Russian Foundation for Basic Research (project 15-02-02764). REFERENCES 1. L.D. Landau and E.M. Lifshitz, Course of Theoretical Physics, Vol. 7: Theory of Elasticity (Nauka, Moscow, 1987). 2. K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y. Zhang, S.V. Dubonos, I.V. Grigorieva, and A.A. Firsov, Science 306, 666 (2004). 3. C. Lee, X. Wei, J.W. Kysar, and J. Hone, Science 321, 385 (2008). 4. A.E. Galashev and O.R. Rakhmanova, Phys.-Usp. 57, 970 (2014). 5. C.D. Reddy, S. Rajendran, and K.M. Liew, Nanotechnology 17, 864 (2006). 6. V.V. Novikov and K.W. Wojciechowski, Phys. Solid State 41, 1970 (1999). 7. J.-W. Jiang and H.S. Park, Nano Lett. 16, 2657 (2016). 8. V.B. Shenoy, C.D. Reddy, A. Ramasubramaniam, and Y.W. Zhang, Phys. Rev. Lett. 101, 245501 (2008). 8 9. A.I. Podlivaev and L.A. Openov, JETP Lett. 103, 185 (2016). 10. Z. Wang, X.-F. Zhou, X. Zhang, Q. Zhu, H. Dong, M. Zhao, and A.R. Oganov, Nano Lett. 15, 6182 (2015). 11. L.A. Openov and A.I. Podlivaev, Phys. Solid State 58, 1705 (2016). 12. M.M. Maslov, A.I. Podlivaev, and K.P. Katin, Mol. Simul. 42, 305 (2016). 13. L.A. Openov and A.I. Podlivaev, Phys. Solid State 58, 847 (2016). 14. B.D. Annin and N.I. Ostrosablin, J. Appl. Mech. Tech. Phys. 49, 998 (2008). 15. O.E. Glukhova and O.A. Terent'ev, Phys. Solid State 48, 1411 (2006). 7 9 This is a preprint of the Work accepted for publication in Physics of the Solid State. © 2016 A.F.Ioffe Physics Technical Institute, Russian Academy of Sience. Publisher http://pleiades.online/. 10 

1612.06666

1

1612

Lifetime enhancement for multi-photon absorption in intermediate band solar cells

A semiconductor structure consisting of two coupled quantum wells embedded into the intrinsic region of a {\it p-i-n} junction is proposed to be implemented as an intermediate band solar cell with ratchet state. The localized conduction subband of the right-hand side quantum well is thought as the intermediated band, while the excited conduction subband of the right-hand side quantum well, coupled to right-rand side one, is thought to acts as the ratchet state. The photo-excited electron in the intermediate band can tunnel out the thin barrier separating the wells and accumulate into ratchet subband. This might raise the electron probability of being hit by a second photon and exiting out to the continuum, increasing solar cell current. Is presented a temporal rate model for describing the charge transport properties of the cell. Calculations are carried out by solving the time-dependent Schr\"odinger equation applying the time evolution operator within a pertinent choice of the non-commuting kinetic and potential operators. The efficiency in the generation of current is analyzed directly by studying the occupation of the subbands wells in the p-i-n junction, taking into account the injection and draining dynamic provided by the electrical contacts connected to the cell. As a result, the efficiency in the generation of current was found to be directly correlated to the relationship between optical generation and recombination rates regarding to the scattering to the ratchet state rate. This suggests that a good coupling between the intermediate band and the additional band is a key point to be analyzed when developing an efficient solar cell.

cond-mat.mes-hall

cond-mat

Lifetime enhancement for multi-photon absorption in intermediate band solar cells 1 Departamento de F´ısica, Instituto de Ciencias Exatas, UNIFAL, 37130-000, Alfenas, MG, Brazil A T Bezerra1,3, N Studart2,3 2 Centro de Ciencias Naturais e Humanas, UFABC, 09210-580, Santo Andr´e, SP, Brazil and 3 DISSE, Instituto Nacional de Ciencia e Tecnologia de Nanodispositivos Semicondutores, Brazil A semiconductor structure consisting of two coupled quantum wells embedded into the intrinsic region of a p-i-n junction is proposed to be implemented as an intermediate band solar cell with ratchet state. The localized conduction subband of the right-hand side quantum well is thought as the intermediated band, while the excited conduction subband of the right-hand side quantum well, coupled to right-rand side one, is thought to acts as the ratchet state. The photo-excited electron in the intermediate band can tunnel out the thin barrier separating the wells and accumulate into ratchet subband. This might raise the electron probability of being hit by a second photon and exiting out to the continuum, increasing solar cell current. Is presented a temporal rate model for describing the charge transport properties of the cell. Calculations are carried out by solving the time-dependent Schrodinger equation applying the time evolution operator within a pertinent choice of the non-commuting kinetic and potential operators. The efficiency in the generation of current is analyzed directly by studying the occupation of the subbands wells in the p-i-n junction, taking into account the injection and draining dynamic provided by the electrical contacts connected to the cell. As a result, the efficiency in the generation of current was found to be directly correlated to the relationship between optical generation and recombination rates regarding to the scattering to the ratchet state rate. This suggests that a good coupling between the intermediate band and the additional band is a key point to be analyzed when developing an efficient solar cell. I. INTRODUCTION process [3]. The demand for renewable energy sources has been promoting the research on semiconductor solar cells [1- 12]. Basically, a solar cell is formed by a p-i-n junc- tion shed with light in order to excite charge carriers be- tween valence and conduction bands. The carriers are then drained by the contacts by a drift electric field, giving rise to a net electric current [15]. However, to a single band gap cell, the detailed balance determines the current generation to be fundamentally limited by the absorption of photons with energies greater than the cell's band gap [1]. To overcome such limit, intermediate bands solar cells has been proposed [2]. A set of electronic states, called intermediate band, is introduced within the semiconductor band gap. This affords new pathways to the carriers [2 -- 4], allowing additional carrier generation due to the extension of the spectral response in the en- ergy range bellow the host-semiconductor band gap [7, 8]. Consequently, the addition of the intermediate band in- creases the light-to-current conversion efficiency of the solar cell. An additional condition to ensure such an effi- ciency enhancement, is that intermediate band needs to be radiatively connected to the valence band but electri- cally isolated from the other bands [9]. The use of the localized subbands of quantum wells, embedded into the intrinsic region of p-i-n solar cells, are promising candidates to be used as the intermediate bands [3]. They has been suggested to effectively enhance solar cell efficiencies [7, 10]. However, extracting carriers from localized states often requires inelastic processes, as multiphoton excitation [11, 12]. That becomes a new problem due to the low carrier lifetime into the interme- diate subbands regarding to the interband recombination Hence, structures with ratchet states has been pro- posed [3 -- 5]. The ratchet are states optically- and electrically-coupled to conduction band and decoupled from the valence band, acting as scattering channels for electrons excited into the intermediate band. The cou- pling between intermediate band and ratchet states, al- lows for excited electrons to increase their lifetimes within the intermediate band. This might increases the multi- photon transition probability and, consequently, the solar cell efficiency [13]. The use of quantum cascade-like scattering to work as the ratchet transport dynamics has been theoretically [5] and experimentally [6] proposed. It showed to be feasi- ble and promising for increasing the cell's efficiency. In such approach, the energy difference between the inter- mediate and hatchet states are set to be in resonance with the host-semiconductor LO-phonon energy, allowing for a phonon-assisted scattering. This spatially shift the opti- cally generated electron-hole pair, decreasing the recom- bination probability, and also increases the electron's life- time within the intermediate band. However there still has a lack of information about the transport dynamics within the excitation/scattering/excitation process. Obviously, studies of this kind of structures must rely on their quantum properties, once we are dealing with quantum devices. Therefore the analyses of such a sys- tems under the quantum mechanics perspective is funda- mental, once we consider the quantum well subbands as the basis of the cell's operation [8, 14, 15]. In what follows, we propose an p-i-n layered structure with two coupled quantum wells embedded into the in- trinsic semiconductor region, to be implemented as an intermediate band solar cell with ratchet state. The 2 while the right-hand side one (rQW) is thinner having only one. The intention was to use the rQW subband as the intermediate band and the lQW's excited subband as the ratchet state. Being the latter an excited state, the selection rules for interband transition uncouple it for the valence band ground state [16], as required [3]. This also enables new pathways to the intersubband absorp- tion process hopefully increasing even more the current generation. In details, the resulting intermediate band solar cell under analysis, was based on a layered structure within material and parameters found in the literature [4]. The 1 µm wider intrinsic layer and the doped contacts has been chosen to be GaxAl1−xAs, with x = 0.3. The left- and the right-hand side quantum wells are formed by GaAs layers with widths 110 A and 40 A, respectively, coupled each other by a 80 A GaxAl1−xAs barrier, as shown in the 1. The energy shift between the rQW ground state and the lQW excited state was set to be around 32 meV, the GaAs LO-phonon energy [18]. It has been chosen the drain and injection contacts, n and p layers, respectively, to be doped with concentra- tions of N = 1017 cm−3. Without any external electric potential applied to the device, at thermal equilibrium, the Fermi level is the same throughout the structure (see the bold black-dashed line in the 1(a)). This gives rise to a built-in potential, Vbi = 4.23 eV, by the depletion of the potential profile at intrinsic region. Then a drift field is established, responsible for extracting photogener- ated carriers towards the drain (n) contact [15]. In order to restore the thermal equilibrium, the same amount of drained charge is re-injected by the injection (p) contact into the cell, having as a net effect the generation of a short circuit current on the external circuit. Under external bias, the potential profile changes due to the equilibrium within the contacts' electrochemical potentials, as shown in the 1(b) at backward bias con- dition, and in the 1(c) at the forward bias condition. It was considered for the levels occupation the equilibrium with the quasi-Fermi levels established by the contacts. At forward bias condition, we do not we expect optical generation of current, since the conduction band states, at thermal equilibrium, are kept fulled. Under such a bias condition, current is expected only in the presence of inelastic scattering processes (common to a p − i − n diode) [15], not considered in the present model. The system was modeled using uncoupled parabolic valence and conduction bands within the effective mass approximation [16]. The eigenfunctions were numeri- cally obtained by solving the time-dependent Schrodinger equation through the Split Operator method, within the imaginary time evolution [17]. In the method, an initially guessed wave function is evolved in time by successive ap- plications of the time-evolution operator, exp(−iHdτ /¯h), choosing dτ = −idt. However, the system's Hamiltonian, H = K + U , formed by the non-commuting kinetic (K) and potential operators (U ), imposes an intrinsic error when applied to the time-evolution operator. The error FIG. 1. (color online) Schematics of the potential profile of the intermediate band solar cell with ratchet states. (a) Structure without aplying external potential, the equilibrium between doped contacts gives rise to a built-in potential eVbi. The arrows represent the optical transitions responsible to gener- ation of photocurrent. (b) Structure under backward external bias eVf . µ(cid:63) n(p) represent the quasi-fermi levels for conduction (valence) band. (c) Structure under forward external bias eVb. transport dynamics is based on the quantum cascade-like scattering at the intermediate band. We analyze the cur- rent response of the cell by means of a rates model using the recombination and generation rates, directly obtained through the system eigenfunctions. We find a direct cor- relation between the scattering rate to the ratchet state and the recombination rates, rising the solar cell cur- rent. However, we observe conditions that even having the scattering to the ratchet states, the resulting current could result in a decrease of the net current. II. METHODS A. Structure As discussed earlier, the major problem of using quan- tum well subbands as intermediate bands is the lowest excitation time for the electron at the intermediate band regarding to its recombination (back to valence band) time. Allowing such an electron to relax into a ratchet state might increase its lifetime and, consequently, the two-photon process efficiency [5, 13]. In order to enhance the generation of current, the cho- sen potential profile was thought to have two quantum wells, and the work is based on dealing with their ab- sorption and emission dynamics. As we can observe in the 1, the left-hand side quantum well (lQW) is wider, having two confined subbands in the conduction band, can be easily handled by properly splitting the exponen- tial argument. Choosing 3 e−i Hdt ¯h = e−i U dt 2¯h e−i Kdt ¯h e−i U dt 2¯h + O(dt3), (1) the error in time evolution can be arbitrarily controlled by appropriate choice of the time increment dt, once it is proportional to the third order of dt. The ex- cited subbands are obtained carrying out the time evolu- tion together with the Gram-Schmidt orthonomalization scheme. Within the eigenfunctions, the subbands' ener- gies are obtained by the mean of the system's Hamilto- nian. B. Rates Model In order to model the electronic excitation and recom- bination dynamics, it was developed a semiclassical rates model taking into account the electronic subbands as sim- ple levels, as shown in the 2. The optical processes lead to changes in the carrier con- centration of the levels within time. The spaced-hatch orange boxes in the 2, represent the valence levels at the lQW and rQW, within carrier concentrations of Nvl and Nvr, respectively. The closed-hatch green boxes repre- sent the intermediate band levels, within carriers con- centrations of Nir, N gnd (considering the left well having a ground and an excited level). Nc is the car- rier concentration of the conduction band level. G(R)s ij is the generation (recombination) rate for the absorption between bands i to j at the s-hand side quantum well. Ti(v) is the transition rate between lQW and rQW at in- termediate (valence) band. This rate is related to the phonon-assisted scattering process between the interme- diate band and the ratchet state. It was set as a control parameter for the simulation. , and N ex il il The changes of the carriers concentrations within time are viNvr + Rr Nir = Gr N ex il = TiNir + Rl il = Gl Nc = Gr N gnd ciNc − (Gr ciNc + Gl viNvl + Rl icNir + Gl ic + Rr il − (Gl iN gnd il − (Gl iN ex il − (Rl icN ex iv)Nir − TiNir (2) il (3) (4) i)N ex iv)N gnd ic + Rl i + Rl ci + Rr ci)Nc, (5) il where Nband is a short representation to dNband/dt. The change with time in the electrons' concentration within a specific band. Let's begin with the 2. At steady state condition, Nir = 0, so FIG. 2. (color online) Levels schematics showing the possi- ble transitions and their respective rates. The spaced-hatch orange boxes represent the valence levels at the lQW and rQW, within carrier concentrations of Nvl and Nvr, respec- tively. The closed-hatch green boxes represents the interme- diate levels, within carriers concentrations of Nir, N gnd , and N ex il , considering the left well having a ground and an ex- cited level. Nc is the conduction level carrier concentration. G(R)s ij is the generation (recombination) rate for the absorp- tion between bands i to j at the s-hand side quantum well. Tv(i) is the transition rate between lQW and rQW at valence (intermediate) band. il 3, also at steady state condition ( N ex il = 0), yields N ex il = Gl iN gnd il + Rl ciNc + TiNir Gl ic + Rl i . (7) We can use 4 to determine the relationship between excited subband in the lQW, considering once more the steady state condition, il = 0. This gives N gnd N gnd il = Gl viNvl + Rl i + Rl iv Gl iN ex il . (8) Backing to the 7, and after some simple algebra N ex il = Gl iGl viNvl + (Gl iv)(Rl ciNc + TiNir) i + Rl γ , (9) where γ = (Gl i + Rl iv)(Gl ic + Rl i) + Gl iRl i. The current extracted from well's region was consid- ered to be proportional to the change of the carriers con- Nc. Substituting 6 and 9 centration at conduction band, into 5, yields Nc (cid:104) Nir = Gr viNvr + Rr Gr ic + Rr ciNc iv + Ti . (6) where = κNc + Gl i+Rl iv)Gl ic + (Gl Gr γ iGl icGl γ vi icTi (cid:105) (10) Nvl + Gr vi ic+Rr iv+Ti Gr Nvr, (cid:104) (Gl i+Rl iv)Gl γ ic (cid:105) − 1 + i+Rl ic+Rr icTi iv)Gl iv+Ti)γ + (Gr 1 ic+Rr iv+Ti Gr = Rl ci (cid:104) (Gl κ Rr ci (cid:105) − 1 , (11) Gl(r) vi = ω0 nphEp ¯hω (cid:88) m,n vc f (Em)[1 − f (En)], Dqw (15) 4 depends on the recharge of the wells via conduction band states, with rates Rl(r) . ci where Ep = 23 eV is the Kane energy for GaAs [18]. The sum is done over the valence (m) and conduction band states (n), weighted by their occupations given by the Fermi distributions f(cid:0)Em(n) (cid:1). The quasi-Fermi level C. Transition rates were determined by the contacts electrochemical poten- tials [15]. The photo-generation process is, mainly, a result of two interband transitions, one at the bulk region and other at the quantum wells region. However, in the latter one, to extract carriers from the confined subbands a second transition needs to take place [3, 4]. Both the tunnel- ing and the intersubband transition to continuum states, extended throughout the contacts region, can be used as such a secondary process [5, 6]. Therefore, the knowledge of the transitions rates interplay is fundamental. We have determined the rates by means of the Fermi's golden Rule [16, 18], using the eigenstates resulting from the Split-Operator method. The generation rate for elec- trons excited from valence band to conduction band at bulk region, by photons with energy ¯hω, is given by [18] (cid:18) 2p2 vc m0 (cid:19) 2 3 Gbulk vc (¯hω) = ω0 nph ¯hω Dvc(¯hω), (12) where ω0 = πe2¯h/m0ε, e and m0 are the electron charge and mass, respectively, ε is the GaAs dielectric constant, nph is the photons' density set to unity, and pvc is the mo- mentum matrix element, which can determined using the Kane approximation [16]. Dvc is the three-dimensional electronic density of states given by √ 2 Dvc(¯hω) = (m∗)3/2(cid:112)¯hω − Eg π2¯h3 being m∗ the electron's effective mass, and Eg the semi- conductor band gap. At the quantum wells region, the generation rates, Gl(r) , are obtained by changing the three-dimensional vi density of states to a two-dimensional one, taking into account the overlap of the envelope functions. So (cid:88) m∗ π¯h2Lqw m,n Dqw vc (¯hω) = (cid:104)φm v φn c (cid:105)2Θ(Emn − ¯hω), (14) where the sum is done over the valence φm v and con- duction band φn c eigenstates, separated in energy by Emn = Eg − Ec n − Ev m. Lqw is the well's width, and Θ is the Heaviside step function, related to the two- dimensional density of states. The interband generation rates, Gl(r) vi are obtained by The interband recombination rates (Rbulk iv ) are similar to the interband generation ones. However, we might to change the density of photons to nph + 1, and to replace the electronic density of states by the photons density of states ρ(¯hω) = (¯hω)2/π2(¯hv)3, where v is the light velocity within the semiconductor [18]. vc , Rl(r) The intersubband generation rates at the quantum wells region, Gl(r) ic , are given by Gl(r) ic (¯hω) = ω0 nph ¯hωLqw pif2f (Ei)[1 − f (Ef )], (16) (cid:88) i,f i f zψl(r) (cid:105)/Lqw, where ψl(r) where the sum is done over the quantum well local- ized subbands (i,j), again taking into account their occupations by using the Fermi distributions f (E). The intersubband momentum matrix element is pif = −i¯h(cid:104)ψl(r) is the left (right) quantum well i-th subband eigenfunction, with energy Ei. The intersubband recombination rate at the lQW, Rl i was considered to be constant, within a value of 10 GHz. With the evaluated generation and recombination rates we are able, to feed the rates model and determine the level dynamics. The results are presented in the next section. i In the solar cell, the electrons excited both from va- lence and intermediate band, towards conduction band, should be extracted from quantum wells regions by the drain contact, restoring the thermal equilibrium. Hence, in the model, we considered the infinity mobility regime, allowing effectively collection of charge by the drain con- tact [2]. The regime is achieved preventing electrons to relax back from continuum to the wells' subbands, choos- ing Rl ci = 0 in the 11, so κ = 0. ci = Rr To analyze the contribution of adding the left-hand side quantum well to the structure, the quantum wells were first assumed to be uncoupled from each other, mak- ing Ti = 0 in the 10, yielding Nc(Ti=0) = Gl icGl iGl vi γ Nvl + Gr icGr vi ic + Rr iv Gr Nvr. (17) , (13) III. RESULTS 5 pronounced peak close to the built-in potential value, for potentials between 3 to 4 eV. Peak intensity increases with increasing wells' coupling rate until saturates for Ti greater than 500 GHz. Such a behavior was understood as follows. As given by 15, the interband generation rate (Gr vi), and the interband recombination rate (Rr iv) are deter- mined by a conjunction of three factors - the Θ step function, the Fermi distributions, and the overlap be- tween valence and intermediate subbands eigenfunctions [18]. Therefore, due to the step function, low energetic photons (regarding to the bands separation) are unable to excite electrons from valence to intermediate subband. The excitation of electrons by higher energetic photons is conditioned by the external bias applied to the cell (which controls the quasi-Fermi levels), and by the over- lap between the valence and intermediate band eigen- states (which decreases with increasing the energy sepa- ration from each other). For backward bias condition, encompassing the built- in potential condition, the hole subbands are considered to be kept filled through the equilibrium within the in- jection contact, while the electron subbands are consid- ered to be kept empty by the drain (see the quasi-Fermi levels in the 1(b)). Instead, for the forward bias condi- tion, the electron subbands are kept filled while the hole subbands are kept empty (see the quasi-Fermi levels in the 1(c)). In the latter condition, the current through the cell is non-null only in the presence of inelastic scat- tering processes [15]. Therefore, the quasi-Fermi levels, determined by the potential applied to the cell, is a fun- damental factor for determining both the generation and the recombination rates. That is directly reflected in the behavior of the cell's current, as discussed next. As noted, the liquid current is non-null only for back- ward bias, as expected by the levels occupations. Analyz- ing 19 in details, the first term inside the square-brackets, related to the transitions into the lQW, presents a step- like shape being close to unit for potentials lesser than ∼ 4 eV, and null elsewhere. The second term, related to the transitions into the rQW, presents the same step- like fashion whose intensity is close to unit for potentials lesser than ∼ 3 eV, and null elsewhere. Therefore, the subtraction between such a terms determines the peak width. The peak edges are given by the intensities rela- tionship between the generation and recombination pro- cesses at the lQW and rQW regions. The step-like profile, discussed above, is mainly due to the step-like shape of the recombination rates Rl iv and Rr iv. Even though the intensities of the recombination rates are three to four orders of magnitude lower then the generation ones, the last are peaked functions with max- ima for specific biases, null elsewhere. For such a biases, both the first and the second terms inside the square- brackets of the 19 are close to the unit, since the domi- nant terms are the generations rates. However, for biases greater than 3 eV (4 eV), Rr iv) is the dominant rate, becoming the square-brackets term null. Consequently, iv (Rl FIG. 3. (color online) Liquid current as a function of the potential applied to the cell and the transition rate Ti between the lQW e rQW. The inset shows the profile of liquid current for Ti = 132, 5 GHz. Clearly, when uncoupled, both the lQW and the rQW might independently contribute for changing the concen- tration of carriers in the conduction band. The contri- bution is proportionally to their valence band concentra- tions, Nvl and Nvr weighted by their specific generation Gl(r) iv . The greater the recombination rate regarding to the gener- ation one, the lesser effective is the current generation process. ic , regarding to the recombination rates Rl(r) We now may think in terms of a liquid current, Il, proportional to the change in the concentration of con- duction carriers (times the electronic charge), when com- paring the system within coupled (Ti (cid:54)= 0) and uncoupled (Ti = 0) quantum wells, (cid:104) NcTi(cid:54)=0 − NcTi=0 (cid:105) Il = e . (18) Subtracting 17 from 10, Il reads Il (cid:104) (Gl = eGr ic+Rr Gr i+Rl iv)Gl γ ic viTi iv+Ti × − Gr ic Gr ic+Rr iv (cid:105) . (19) Such a liquid current was interpreted as the contribu- tion (to the current) of adding the additional scattering channel, charging the ratchet state. It would enhances the carriers lifetime at the intermediate band, increasing the cell's efficiency. In the 3, we presented the liquid current, given by the 19, as a function of the transition rate Ti and the cell's potential. The potential was varied from backward to forward biases relative to the built-in potential, uni- formly spread through the cell's intrinsic region (see 1 (b) and (c)). For Ti greater than 1 GHz, we observed an effective enhancement of the liquid current, with a the liquid current is non-null only in that specific poten- tial range. Then the inclusion of the lQW and its relationship with the rQW determines the presence of an enhancement in the cell's current. The peak width in the liquid current is, therefore, controlled by subbands occupation of both wells, which determines the recombination rates cutoffs with the potential. So, the liquid current peak could in principle be enlarged by increasing the energy sepa- ration between the rQW state (intermediate band) and the ground state of the lQW, responsible to uncouple the ratchet state to the valence band. The saturation, which determines the liquid current peak intensity, is mainly drive by the first term at the right-hand side of 19. It takes into account the relation- ship between both the rQW recombination (Rr iv) and generation (Gr ic) rates, regarding to the transition rate between the wells (Ti) itself. Once more the recombina- tion rate Rr iv, with the same step-like shape and max- imum intensity of around 100 GHz, is responsible for controls the liquid current intensity. Increasing Rr iv de- creases the electrons concentration at the intermediate band and, consequently, the probability of such electrons to be scattered to the lQW, within rate Ti. Therefore, for Ti lesser than such a recombination rate, we expected a lay back in the cell's current as observed. Our results has been shown a clear enhancement in the generation of current by the intermediate band so- lar cell, within the excited state of the lQW acting as a quantum ratchet state. This is in consonance with Noda and coworkers [4], who developed a thermodynam- ical model and concluded the use of intermediate band solar cells, adding the ratchet state is expected to al- ways enhance solar cell's efficiency beyond the Shockley- Quiesser limit [1]. They also showed that the inclusion of the intermediate band by itself is not a determining condition for increasing efficiency. The work provided conditions where the intermediate band solar cells could be less efficient than singled junction ones. According to them, an intermediate band solar cell will always be more efficient then a singled junction one, and overcome the Shockley-Queisser limit, only when adding ratchet states. However, we observed that even introducing the quan- tum ratchet state, such a behavior can be changed. As discussed earlier, the liquid current behavior is a conse- quence of the relationship of several generation and tran- sition processes. 4, presents the liquid current behavior, for Ti = 10 GHz and a backward potential of 3.8 eV, as a function of the intersubband recombination rate Rl i. We can observe the change in the liquid current signal, which becomes negative for higher Rl i, indicating that increas- ing such a rate can be harmful to the liquid current for specific conditions. Independently of how effective is the coupling between the intermediate band and the ratchet state, regarding of Ti rate, our model shows there are conditions in which the introduction of the ratchet states is not a guarantee 6 FIG. 4. Liquid current as a function of the transition rate between the ground and excited subband at the rQW, RL i , for Ti = 10 GHz and 3.8 eV. We can observe a current's signal change around Rl i = 70 GHz. of increasing cell's current. Allowing electrons to be scat- tered into the ratchet state should increase their lifetime into the intermediate band, but the relationship between the recombination and the scattering rates can both in- crease or decrease the solar cell current, depending on the other rates involved in the cell's dynamics. IV. CONCLUSION In summary, we have demonstrated the possibility of using the subbands of quantum wells as the interme- diate band and ratchet state of an intermediate band solar cell with ratchet state, using a quantum cascade like approach. The cell was considered to be a double quantum well structure within GaAlAs and GaAs layers sandwiched between heavily doped GaAs contacts. The system's eigenfunctions were numerically simulated by solving the time-dependent Schrodinger equation. The recombination and generation rates, used in a semiclas- sical rates model, were obtained by means of the Fermi golden rule, within the simulated eigenstates. With al- lowing electron to scattering inside the ratchet state, as a LO-phonon assisted scattering, we have shown an effec- tive increase in the cell's current for specific values of elec- trical potential, close to the built-in potential of the cell. Agreeing with other works, the inclusion of the ratchet state increases the solar cell current, by increasing the electron's lifetime into the intermediate band. However, the addition of the ratchet state is not the unique condi- tion determining the cell's efficiency. The recombination and generation dynamics plays a fundamental role, and we can achieve situations in which the solar cell current is decreased even with the inclusion of the ratchet state. 7 ACKNOWLEDGMENTS We thank CNPq and CAPES for research fellowships and for direct and indirect research funding. N.S. thanks CAPES for a Visiting Senior Professorship at UFABC. The authors thank CNPq for providing funding for this research through INCT-DISSE consortium. [1] Shockley W, and Queisser H J 1961 Journ. Appl. Phys. Phys. Rev. Lett. 97 247701 32 510 [2] Luque A, and Mart´ı A 1997 Phys. Rev. Lett. 78 5014 [3] Yoshida M, Ekins-Daukes N J, Farrell D J, and Phillips C C 2012 Appl. Phys. Lett. 100 263902 [4] Noda T, Elborg M, Mano T, Kawazu T, Han L and Sakaki H 2016 Journ. Appl. Phys 119 085105 [5] Curtin O J, Yoshida M, Pusch A, Hylton N P, Ekins- Daukes N J, Phillips C C, and Hess O 2016 IEEE J. Photovoltaics 6 673 [6] Sugiyama M S, Wang Y, Watanabe K, Morioka T, Okada Y, and Nakano Y 2012 IEEE J. Photovoltaics 2 298 [7] Barnham K W J, and Duggan G 1990 Journ. Appl. Phys 67 3490 [8] Cabrera C I, Rimada J C, Courel M, Hernandez L, Con- nolly J P, Enciso A, and Contreras-Solorio D A 2013 Natu. Resour. 4 235 [12] Elborg M, Noda T, Mano T, Jo M, Sakuma Y, Sakoda K, and Han L 2015 Sol. Ener. Mat. and Sol. Cell. 134 108 [13] Pusch A, Yoshida M, Hylton N P, Mellor A, Phillips C C, Hess O, and Ekins-Daukes N J 2016 Prog. Photovolt: Res. Appl. 24 656 [14] Datas A, L´opez E, Ramiro I, Antol´ın A, Mart´ı A, Luque A, Tamaki R, Shoji Y, Sogabe T, and Okada Y 2015 Phys. Rev. Lett. 114 157701 [15] Datta S 2012 Lessons for Nanoelectronics: A new per- spective on transport (Singapure: World Scientific Pub- lishing Company) [16] Bastard G 1992 Wave Mechanics Applied to Semiconduc- tor Heteroestructures (Paris: Les Editions de Physique, JOUVE) p 247 [17] Degani M H and Maialle M Z 1010 J. Comput. Theor. [9] Luque A, Mart´ı A, and Stanley C 2012 Nat. Photon. 6 Nanosci. 7 454 146 [10] Anderson N G 1995 Journ. Appl. Phys 78 1853 [11] Mart´ı A, Antol´ın E, Stanley C R, Farmer C D, L´opez N, D´ıaz P, C´anovas E, Linares P G, and Luque A 2006 [18] Singh J 2003 Electronic and Optoeletronic Properties of Semiconductors Structures (New York, NY: Cambridge University Press) p 222, 360, 362 

1212.1355

1

1212

Adaptive tuning of Majorana fermions in a quantum dot chain

We suggest a way to overcome the obstacles that disorder and high density of states pose to the creation of unpaired Majorana fermions in one-dimensional systems. This is achieved by splitting the system into a chain of quantum dots, which are then tuned to the conditions under which the chain can be viewed as an effective Kitaev model, so that it is in a robust topological phase with well-localized Majorana states in the outermost dots. The tuning algorithm that we develop involves controlling the gate voltages and the superconducting phases. Resonant Andreev spectroscopy allows us to make the tuning adaptive, so that each pair of dots may be tuned independently of the other. The calculated quantized zero bias conductance serves then as a natural proof of the topological nature of the tuned phase.

cond-mat.mes-hall

cond-mat

Adaptive tuning of Ma jorana fermions in a quantum dot chain. Ion C. Fulga,1 Arbel Haim,2 Anton R. Akhmerov,1, 3 and Yuval Oreg2 1 Instituut-Lorentz, Universiteit Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands 2Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot, 76100, Israel 3Department of Physics, Harvard University, Cambridge, MA 02138 (Dated: February 5, 2014) We suggest a way to overcome the obstacles that disorder and high density of states pose to the creation of unpaired Ma jorana fermions in one-dimensional systems. This is achieved by splitting the system into a chain of quantum dots, which are then tuned to the conditions under which the chain can be viewed as an effective Kitaev model, so that it is in a robust topological phase with well-localized Ma jorana states in the outermost dots. The tuning algorithm that we develop involves controlling the gate voltages and the superconducting phases. Resonant Andreev spectroscopy allows us to make the tuning adaptive, so that each pair of dots may be tuned independently of the other. The calculated quantized zero bias conductance serves then as a natural proof of the topological nature of the tuned phase. PACS numbers: 74.45.+c, 74.78.Na, 73.63.Kv, 03.65.Vf I. INTRODUCTION Ma jorana fermions are the simplest quasiparticles pre- dicted to have non-Abelian statistics.1,2 These topo- logically protected states can be realized in condensed matter systems, by making use of a combination of strong spin-orbit coupling, superconductivity, and bro- ken time-reversal symmetry.3–5 Recently, a series of ex- periments have reported the possible observation of Ma- jorana fermions in semiconducting nanowires,6–9 attract- ing much attention in the condensed matter community. Associating the observed experimental signatures ex- clusively with these non-Abelian quasiparticles, however, is not trivial. The most straightforward signature, the zero bias peak in Andreev conductance10,11 is not unique to Ma jorana fermions, but can appear as a result of var- ious physical mechanisms,12–18 such as the Kondo effect or weak anti-localization. It has also been pointed out that disorder has a detrimental effect on the robustness of the topological phase, since in the absence of time- reversal symmetry it may close the induced supercon- ducting gap.19 This requires experiments performed with very clean systems. Additionally, the presence of multi- ple transmitting modes reduces the amount of control one has over such systems,20–23 and the contribution of extra modes to conductance hinders the observation of Ma jo- rana fermions.24 Thus, nanowire experiments need setups in which only few modes contribute to conductance. In this work we approach the problem of realizing systems in a non-trivial topological phase from a dif- ferent angle. We wish to emulate the Kitaev chain model25 which is the simplest model exhibiting unpaired Ma jorana bound states. The proposed system consists of a chain of quantum dots (QDs) defined in a two- dimensional electron gas (2DEG) with spin orbit cou- pling, in proximity to superconductors and sub jected to an external magnetic field. Our geometry enables us to control the parameters of the system to a great extent by varying gate potentials and superconducting phases. FIG. 1: Examples of systems allowing implementation of a Kitaev chain. Panel (a): a chain of quantum dots in a 2DEG. The QDs are connected to each other, and to superconductors (labeled SC), by means of quantum point contacts. The first and the last dots are also coupled to external leads. The normal state conductance of quantum point contacts (QPCs) between adjacent dots or between the end dots and the leads is G(cid:107) , and of the QPCs linking a dot to a superconductor is G⊥ . The confinement energy inside each QD can be controlled by varying the potential Vgate . Panel (b): Realization of the same setup using a nanowire, with the difference that each dot is coupled to two superconductors in order to control the strength of the superconducting proximity effect without the use of QPCs. We will show how to fine tune the system to the so- called “sweet spot” in parameter space, where the Ma- jorana fermions are well-localized at the ends of the sys- tem, making the topological phase maximally robust. A sketch of our proposed setup is presented in Fig. 1a). The setup we propose and the tuning algorithm is not restricted solely to systems created in a two-dimensional electron gas. The essential components are the ability to form a chain of quantum dots and tune each dot sep- arately. In semiconducting nanowires the dots can be formed from wire segments separated by gate-controlled tunnel barriers, and all the tuning can be done by gates, except for the coupling to a superconductor. This cou- pling, in turn, can be controlled by coupling two super- conductors to each dot and applying a phase difference to these superconductors. The layout of a nanowire im- plementation of our proposal is shown in Fig. 1b). This geometry has the advantage of eliminating many of the problems mentioned above. By using single level quantum dots, and also quantum point contacts (QPC) in the tunneling regime, we solve issues related to mul- tiple transmitting modes. Additional problems, such as accidental closings of the induced superconducting gap due to disorder, are solved because our setup allows us to tune the system to a point where the topological phase is most robust, as we will show. We present a step-by-step tuning procedure which fol- lows the behavior of the system in parallel to that ex- pected for the Kitaev chain. As feedback required to control every step we use the resonant Andreev conduc- tance, which allows to track the evolution of the system’s energy levels. We expect that the step-by-step structure of the tuning algorithm should eliminate the large num- ber of non-Ma jorana explanations of the zero bias peaks. A related layout together with the idea of simulating a Kitaev chain was proposed recently by J. D. Sau and S. Das Sarma.26 Although similar in nature, the geom- etry which we consider has several advantages. First of all, coupling the superconductors to the quantum dots in parallel, allows us to not rely on crossed Andreev reflec- tion. More importantly, being able to control inter-dot coupling separately from all the other properties allows to address each dot or each segment of the chain elec- trically. This can be achieved by opening all the QPCs except for the ones that contact the desired dots. This setup can also be extended to more complicated geometries which include T-junctions of such chains. Benefiting from the high tunability of the system and the localization of the Ma jorana fermions, it might then be possible to implement braiding27,28 and demonstrate their non-Abelian nature. The rest of this work is organized as follows. In sec- tion II we briefly review a generalized model of Kitaev chain, and identify the ”sweet spot” in parameter space in which the Ma jorana fermions are the most localized. The system of coupled quantum dots is described in sec- tion III. For the purpose of making apparent the resem- blance of the system to the Kitaev chain, we present a simple model which treats each dot as having a single spinful level. We then come up with a detailed tuning procedure describing how one can control the parameters of the simple model, in order to bring it to the desired 2 point in parameter space. In section IV our tuning pre- scription is applied to the suggested system of a chain of QDs defined in a 2DEG, and it is shown using numeri- cal simulations that at the end of the process the system is indeed in a robust topological phase. We conclude in section V. II. GENERALIZED KITAEV CHAIN In order to realize unpaired Ma jorana bound states, we start from the Kitaev chain25 generalized to the case where the on-site energies as well as the hopping terms are not uniform and can vary from site to site. The gen- (cid:104)(cid:16) L−1(cid:88) eralized Kitaev chain Hamiltonian is defined as † † n+1a† (cid:17) (cid:105) n+1an + ∆n eiφn a tn eiθn a n n=1 + h.c. + εna† nan HK = , (2.1) where an are fermion annihilation operators, εn are the on-site energies of these fermions, tn exp(iθn ) are the hopping terms, and ∆n exp(iφn ) are the p-wave pairing terms. The chain supports two Ma jorana bound states local- ized entirely on the first and the last sites, when (i): εn = 0, (ii): ∆n = tn , and (iii) φn+1 − φn − θn+1 − θn = 0. The larger values of tn lead to a larger excitation gap. The condition (iii) is equivalent, up to a gauge transfor- mation, to the case where the hopping terms are all real, and the phases of the p-wave terms are uniform. The en- ergy gap separating the Ma jorana modes from the first excited state then equals Egap = 2 min {tn}n . (2.2) The above conditions (i)–(iii), constitute the “sweet spot” in parameter space to which we would like to tune our system. Since all of these conditions are local and only involve one or two sites, our tuning procedure in- cludes isolating different parts of the system and mon- itoring their energy levels. For that future purpose we will use the expression for excitation energies of a chain of only two sites with ε1 = ε2 = 0: E12 = ±(t1 ± ∆1 ). (2.3) III. SYSTEM DESCRIPTION AND THE TUNING ALGORITHM The most straightforward way to emulate the Kitaev chain is to create an array of spinful quantum dots, and apply a sufficiently strong Zeeman field such that only one spin state stays close to the Fermi level. Then the operators of these spin states span the basis of the Hilbert space of the Kitaev chain. If we require normal hopping (cid:17) (3.1) between the dots and do not utilize crossed Andreev re- flection, then in order to have both tn and ∆n nonzero we need to break the particle number conservation and spin conservation. The former is achieved by coupling each dot to a superconductor, the latter can be achieved by spatially varying Zeeman coupling,29,30 or more con- ventionally by using a material with a sufficiently strong spin-orbit coupling. Examples of implementation of such a chain of quantum dots in a two dimensional electron gas and in semiconducting nanowires are shown in Fig. 1. We neglect all the levels in the dots except for the one closest to the Fermi level, which is justified if the level spacing in the dot is larger than all the other Hamilto- nian terms. We neglect the Coulomb blockade, since we assume that the dot is strongly coupled to the supercon- ducting lead. The general form of the BdG Hamiltonian describing such a chain of spinful single-level dots is then (cid:88) given by: (µnσ0 + Vz σz ) c† (cid:16) n,s cn,s(cid:48) HS = n,s,s(cid:48) + (cid:0)wn eiλnσ c† n,s cn+1,s(cid:48) + h.c.(cid:1) , † ∆ind,n eiΦn iσy c† 1 + n,s c n,s(cid:48) + h.c. 2 where c† n,s and cn,s are creation and annihilation oper- ators of a fermion with spin s in the n-th dot, and σi are Pauli matrices in spin space. The physical quanti- ties entering this Hamiltonian are the chemical poten- tial µn , the Zeeman energy Vz , the proximity-induced pairing ∆ind,n exp(iΦn ), and the inter-dot hopping wn . The vector λn characterizes the amount of spin rotation happening during a hopping between the two neighbor- ing dots (the spin rotates by a 2λ angle). This term may be generated either by a spin-orbit coupling, or by a position-dependent spin rotation, required to make the Zeeman field point in the local z -direction.29–31 The in- duced pairing in each dot ∆ind,n exp(iΦn ) is not to be confused with the p-wave pairing term ∆n exp(iφn ) ap- pearing in the Kitaev chain Hamiltonian (2.1). In order for the dot chain to mimic the behavior of the Kitaev chain in the sweet spot, each dot should have a single fermion level with zero energy, so that εn = 0. Di- agonalizing a single dot Hamiltonian yields the condition (cid:113) for this to happen: µn = When this condition is fulfilled, each dot has two (cid:17) (cid:16)(cid:112)Vz − µn c −iΦn (cid:112)Vz + µn cn↓ fermionic excitations n↑ − e † (cid:16)(cid:112)Vz − µn c (cid:17) −iΦn (cid:112)Vz + µn cn↑ † n↓ + e The energy of an is zero, the energy of bn is 2Vz . If the hopping is much smaller than the energy of the excited ei Φn 2√ 2Vz ei Φn 2√ 2Vz z − ∆2 V 2 ind,n . . (3.4) an = bn = (3.2) (3.3) 3 state, wn (cid:28) Vz , we may pro ject the Hamiltonian (3.1) onto the Hilbert space spanned by an . The resulting pro jected Hamiltonian is identical to the Kitaev chain Hamiltonian of Eq. (2.1), with the following effective pa- rameters: sin (αn+1 + αn ) cos(δΦn/2) εn = 0, (cid:20) tn eiθn = wn (cos λn + i sin λn cos ρn ) × (cid:21) , +i cos (αn+1 − αn ) sin(δΦn/2) (cid:20) ∆n eiφn = iwn sin λn sin ρn eiξn × cos (αn+1 + αn ) cos (δΦn/2) +i sin (αn+1 − αn ) sin (δΦn/2) where (cid:21) , (3.5a) (3.5b) (3.5c) (3.6) (3.7) µn = Vz sin(2αn ), ∆ind,n = Vz cos(2αn ), λn = λn (sin ρn cos ξn , sin ρn sin ξn , cos ρn )T , and δΦn = Φn − Φn+1 . It is possible to extract most of the parameters of the dot Hamiltonian from level spectroscopy, and then tune the effective Kitaev chain Hamiltonian to the sweet spot. The tuning, however, becomes much simpler if two out of three of the dot linear dimensions are much smaller than the spin-orbit coupling length. Then the direction of spin-orbit coupling does not depend on the dot number, and as long as the magnetic field is perpendicular to the spin-orbit field, the phase of the prefactors in Eqs. (3.5) becomes position-independent. Additionally, if the dot size is not significantly larger than the spin-orbit length, the signs of these prefactors are constant. This ensures that if δΦn = 0, the phase matching condition of the Kitaev chain is fulfilled. Since δΦn = 0 leads to both tn and ∆n having a minimum or maximum as a function of δΦn , this point is straightforward to find. The only remaining condition, tn = ∆n at δΦ = 0, requires that αn + αn+1 = λn . The above calculation leads to the following tuning algorithm: 1. Open all the QPCs, except for two contacting a single dot. By measuring conductance while tuning the gate voltage of a nearby gate, ensure that there is a resonant level at zero bias. After repeating for each dot the condition εn = 0 is fulfilled. 2. Open all the QPCs except the ones near a pair of neighboring dots. Keeping the gate voltages tuned such that εn = 0, vary the phase difference between the neighboring superconductors until the lowest resonant level is at its minimum as a function of phase difference, and the next excited level at a maximum. This ensures that the phase tuning con- dition φn+1 − φn − θn+1 − θn = 0 is fulfilled. Repeat for every pair of neighboring dots. 3. Start from one end of the chain, and isolate pairs of dots like in the previous step. In the pair of n-th and n + 1-st dots tune simultaneously the coupling of the n + 1-st dot to the superconductor and the chemical potential in this dot, such that εn+1 stays equal to 0. Find the values of these parameters such that a level at zero appears in two dots when they are coupled. After that proceed to the following pair. Having performed the above procedures, the coupling between all of the dots in the chain is resumed, at which point we expect the system to be in a robust topological phase, with two Ma jorana fermions located on the first and last dots. In practice one can also resume the cou- pling gradually by, for instance, isolating triplets of adja- cent dots, making sure they contain a zero-energy state, and making fine-tuning corrections if necessary, and so on. IV. TESTING THE TUNING PROCEDURE BY NUMERICAL SIMULATIONS We now test the tuning procedure by applying it to a numerical simulation of a chain of three QDs in a 2DEG. The two-dimensional BdG Hamiltonian describing the (cid:18) p2 (cid:19) entire system of the QD chain reads:  (σxpy − τz σy px ) α 2m + ∆ind (cos(Φ)τy + sin(Φ)τx ) σy + Vz τz σz . (4.1) HQDC = + V (x, y) τz + Here, σi and τi are Pauli matrices acting on the spin and particle-hole degrees of freedom respectively. The term V (x, y) describes both potential fluctuations due to dis- order, and the confinement potential introduced by the gates. The second term represents Rashba spin-orbit cou- pling, ∆ind (x, y) · exp (Φ(x, y)) is the s-wave superconduc- tivity induced by the coupled superconductors, and Vz is the Zeeman splitting due to the magnetic field. Full de- scription of the tight-binding equations used in the sim- ulation is presented in Appendix A. The chemical potential of the dot levels µn is tuned by changing the potential V (x, y). For simplicity we used a constant potential Vn added to the disorder potential, such that V (x, y) = Vn + V0 (x, y) in each dot. Varying the magnitude of ∆ind,n is done by changing conductance G⊥ of the quantum point contacts, which control the coupling between the dots and the superconductors. Fi- nally, varying the superconducting phase Φ(x, y) directly controls the parameter Φn of the dot to which the su- perconductor is coupled, although they need not be the same. 4 FIG. 2: Andreev conductance measured from the left lead as a function of bias voltage and QD potential (measured relative to quarter filling) for the second dot. Changing the chemical potential allows to tune quasi-bound states to zero energy (white circle). (4.2) The tuning algorithm required monitoring the energy levels of different parts of the system. This can be achieved by measuring the resonant Andreev conduc- tance from one of the leads. The Andreev conductance is given by33,34 † G/G0 = N − Tr(ree r† ee ) + Tr(rhe r he ), where G0 = e2/h, N is the number of modes in a given lead, and ree and rhe are normal and Andreev reflection matrices. Accessing parts of the chain (such as a single dot or a pair of dots) can be done by opening all inter- dot QPCs, and closing all the ones between dots and superconductors, except for part of the system that is of interest. We begin by finding such widths of QPCs that G(cid:107) ≈ 0.02 and G⊥ ≈ 4G0 . This ensures that conductance be- tween adjacent dots, is in tunneling regime and that the dots are strongly coupled to the superconductors such that the effect of Coulomb blockade is reduced.35 The detailed properties of QPCs are descrbed in App. A and their conductance is shown in Fig. 8. First step: tuning chemical potential. We sequentially isolate each dot, and change the dot potential Vn . The Andreev conductance as a function of Vn and bias voltage for the second dot is shown in Fig. 2. We tune Vn to the value where a conductance resonance exists at zero bias. This is repeated for each of the dots and ensures that µn = 0. Second step: tuning the superconducting phases. We now set the phases of the induced pairing potentials Φn to constant. As explained in the previous section, this oc- curs when ∆n and tn experience their maximal and min- imal values. According to Eq. (2.3) this happens when the separation between the energy levels of the pair of dots subsection is maximal. Fig. 3 shows the evolution of these levels as a function of the phase difference be- tween the two superconductors. The condition δΦ1 = 0 is then satisfied at the point where their separation is 5 FIG. 3: Conductance as a function of bias voltage and super- conducting phase difference for a two-dot system. The two lowest energy levels are given by Eq. (2.3) of a two site Ki- taev chain, as indicated. At the point where their separation is maximal (SC phase difference 0 in the plot), the phase dif- ference δΦn of the induced superconducting gaps vanishes. maximal. Third step: tuning the couplings. Finally we tune tn = ∆n . This is achieved by varying G⊥ , while tracking the Andreev conductance peak corresponding to the tn − ∆n eigenvalue of the Kitaev chain we are emulating. After every change of G⊥ we readjust Vn in order to make sure that the condition εn = 0 (or equivalently V 2 z = µ2 n +∆2 n ) is maintained. This is necessary because not just ∆n , but also µn depend on G⊥ . Therefore, successive changes of G⊥ and Vn are performed until the smallest bias peak is located at zero bias. The tuning steps of the first two dots are shown in Fig. 4. We repeat steps 2 and 3 for each pair of dots in the system. Finally, having full all three conditions required for a robust topologically non-trivial phase, we probe the presence of localized Ma jorana bound state in the full three-dot system by measuring Andreev conductance (see Fig. 5). In this specific case, the height of the zero bias peak is approximately 1.85 G0 , signaling that the end states are well but not completely decoupled. Increasing the transparency of the QPC connecting the first dot to the lead brings this value to G = 1.98 G0 . V. CONCLUSION In conclusion, we have demonstrated how to tune a linear array of quantum dots coupled to superconductors in presence of Zeeman field and spin-orbit coupling to resemble the Kitaev chain that hosts Ma jorana bound states at its ends. Furthermore, we have presented a de- tailed procedure by which the system is brought to the so-called “sweet spot” in parameter space, where the Ma- jorana bound states are the most localized. This proce- dure involves varying the gates potentials and supercon- ducting phases, as well as monitoring of the excitation FIG. 4: Conductance as a function of bias voltage during simultaneous tuning of G⊥ and Vn for the first pair of dots. The three different plots represent the situation before (dotted line), at an intermediate stage (dashed line), and after (solid line) the tuning. The arrow indicates the evolution of the first peak upon tuning, and the number of successive changes of G⊥ and Vn are shown for each curve. By bringing the first peak to zero, the third tuning step is achieved. FIG. 5: Conductance as a function of bias voltage for a sys- tem composed of three tuned quantum dots (dashed line). The zero bias peak signals the presence of Ma jorana bound states at the ends of the chain. The first and second excited states are consistent with those expected for a three-site Ki- taev chain, namely E1 = 2t1 and E2 = 2t2 (vertical dashed lines), given the measured values of t1 = ∆1 and t2 = ∆2 , obtained after finalizing the two dot tuning process. As de- scribed in the main text, after increasing the transparency of the lead QPC leads to a zero bias peak having a height G = 1.98G0 (solid line). spectrum of the system by means of resonant Andreev conductance. We have tested our procedure using numerical simula- tions of a system of three QDs, defined in a 2DEG, and found that it works in systems with experimentally reach- able parameters. It can be also applied to systems where 6 Quantum point contacts have a longitudinal dimension of LQPC = 42 nm, which is the same as the Fermi wave- length at quarter filling. The value of the hopping integral becomes t = 2 /(2ma2 ) = 55.8 meV, with a = 7 nm. Disorder is introduced in the form of random uncorrelated on- site potential fluctuations, leading to a mean free path lmfp = 218.8 nm. The system is placed in a perpendic- ular magnetic field characterized by a Zeeman splitting Vz = 336 µeV, which, given a g -factor of 35K/T , cor- responds to a magnetic field Bz = 111 mT. Each dot is additionally connected to a superconductor characterized by a pairing potential ∆SC = 0.86 meV. The potential profile across a quantum point contact (cid:18) (cid:19)(cid:19) (cid:18) (cid:101)s(cid:101)L (cid:18) (cid:101)h is given by x + (cid:101)w (cid:18) (cid:19)(cid:19)(cid:19) (cid:18) (cid:101)s(cid:101)L 2 − tanh x − (cid:101)w 2 2 , (A1) + tanh 2 where x ∈ [− (cid:101)L/2, (cid:101)L/2] is the transverse coordinate across the quantum point contact, (cid:101)h is the maximum height of VQPC , (cid:101)s fixes the slope at which the potential changes, and (cid:101)w is used to tune the QPC transparency. Two ex- amples of potential profiles are shown in Fig. 7. VQPC (x) = FIG. 7: Potential profile VQPC (x) across the transverse direc- tion of a quantum point contact. For the maximum value of this potential, no states are available for quasiparticles in the QPC transparencies, corresponding to (cid:101)s = 17 and (cid:101)w = 87.4, 2DEG. The two curves show potential profiles for two different 39.5 nm for the solid and dashed curves respectively. FIG. 6: Geometry of the quantum dot chain. The quantum dots have a width WDOT and length equal to LDOT . Quantum point contacts have a longitudinal size LQPC and a transverse dimension equal to either LDOT or WDOT . Leads are semi- infinite in the x direction, and superconductors are modeled as semi-infinite systems in the y direction. quantum dots are defined by other means, for example formed in a one-dimensional InAs or InSb wire. Acknowledgments The numerical calculations were performed using the kwant package developed by A. R. Akhmerov, C. W. Groth, X. Waintal, and M. Wimmer. We would like to acknowledge discussions with J. Alicea, L. P. Kouwen- hoven, C. M. Marcus, F. von Oppen, and J. D. Sau. We are grateful for partial support by SPP 1285 of the Deutsche Forschungsgmeinschaft (YO), for grants of ISF and TAMU (YO), to the Dutch Science Foun- dation NWO/FOM and an ERC Advanced Investigator Grant (ICF and AA). AA was partially supported by a Lawrence Golub Fellowship. Appendix A: System parameters in numerical simulations In this section, we describe the parameters used throughout the numerical simulations. The quantum dots and quantum point contacts are modeled using a tight-binding model defined on a square lattice, with leads and superconductors taken as semi-infinite. The characteristic length and energy scales of this sys- tem are the spin-orbit length lSO = 2/mα, and the spin- orbit energy ESO = mα2/2 . We simulate an InAs sys- tem in which the effective electron mass is m = 0.015me , where me is the bare electron mass, taking values of ESO = 1 K = 86 µeV and lSO = 250 nm. We consider a setup composed of three quantum dots, like the one shown in Fig. 6. Each of the three dots has a length of LDOT = 208 nm and a width WDOT = 104 nm. 1 J. Alicea, Rep. Prog. Phys. 75, 076501 (2012). 2 C. W. J. Beenakker, arXiv:1112.1950 (2011). 3 J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, Phys. Rev. Lett. 104, 040502 (2010). 4 R. M. Lutchyn, J. D. Sau and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010). 5 Y. Oreg, G. Refael and F. Von Oppen, Phys. Rev. Lett. 105, 177002 (2010) 7 14 K. Flensberg, Phys. Rev. B 82, 180516(R) (2010). 15 G. Kells, D. Meidan, and P. W. Brouwer, Phys. Rev. B 85, 060507(R) (2012). 16 S. Tewari, T. D. Stanescu, J. D. Sau, and S. Das Sarma, Phys. Rev. B 86, 024504 (2012). 17 F. Pientka, G. Kells, A. Romito, P. W. Brouwer, and F. von Oppen, arXiv:1206.0723. 18 J. Liu, A. C. Potter, K. T. Law, and P. A. Lee, arXiv:1206.1276. 19 P. W. Anderson, J. Phys. Chem. Solids 11, 26 (1959). 20 A. C. Potter and P. A. Lee, Phys. Rev. Lett. 105, 227003 (2010). 21 T. D. Stanescu, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. B 84, 144522 (2011). 22 P. W. Brouwer, M. Duckheim, A. Romito and F. von Op- pen, Phys. Rev. Lett. 107, 196804 (2011). 23 M.-T. Rieder, G. Kells, M. Duckheim, D. Meidan, and P. W. Brouwer, Phys. Rev. B 86, 125423 (2012). 24 M. Wimmer, A.R. Akhmerov, J.P. Dahlhaus, C.W.J. Beenakker, New J. Phys. 13, 053016 (2011). 25 A. Yu. Kitaev, Phys.-Usp. 44, 131 (2001). 26 J. D. Sau, S. Das Sarma, Nat. Comm. 3, 964 (2012). 27 J. Alicea, Y. Oreg, G. Refael, F. Von Oppen, M. P. A. Fisher, Nat. Phys. 7, 412 (2011). 28 J. D. Sau, D. J. Clarke, S. Tewari, Phys. Rev. B 84, 094505 (2011). 29 T.-P. Choy, J. M. Edge, A. R. Akhmerov, C. W. J. Beenakker, Phys. Rev. B 84, 195442 (2011). 30 M. Kjaergaard, K. Wolms, K. Flensberg, Phys. Rev. B 85, 020503(R) (2012). 31 B. Braunecker, G. I. Japaridze, J. Klinova ja, D. Loss, Phys. Rev. B 82, 045127 (2010). 32 Q.-F. Sun, J. Wang, H. Guo, Phys. Rev. B 71, 165310 (2005). 33 A. L. Shelankov, Zh. Eksp. Teor. Fiz., Pisma. 32, 2, 122- 125 (1980). 34 G. E. Blonder, M. Tinkham, and T. M. Klapwijk, Phys. Rev. B 25, 4515 (1982). 35 H. Grabert and M. H. Devoret, Single Charge Tunnel- ing: Coulomb Blockade Phenomena in Nanostructures (Plenum, New York, 1992) of (cid:101)w of Eq. (A1), for a single QPC. The vertical lines indicate FIG. 8: Conductance of a quantum point contact as a function the values at which QPCs are set after tuning. The inter- dot QPCs are all set to the tunneling regime while the ones connecting the dots to the superconductors are set to higher transparencies. 6 V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science 336, 1003 (2012). 7 M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff, and H. Q. Xu, arXiv:1204.4130. 8 A. Das, Y. Ronen, Y. Most, Y. Oreg, M. Heiblum, and H. Shtrikman, arXiv:1205.7073. 9 L. P. Rokhinson, X. Liu, arXiv:1204.4212 (2012). 10 C. J. Bolech and E. Demler, Phys. Rev. Lett. 98, 237002 (2007). 11 K. T. Law, P. A. Lee, and T. K. Ng, Phys. Rev. Lett. 103, 237001 (2009). 12 S. Sasaki, S. De Franceschi, J. M. Elzerman, W. G. van der Wiel, M. Eto, S. Tarucha, and L. P. Kouwenhoven, Nature 405, 764 (2000). 13 D. I. Pikulin, J. P. Dahlhaus, M. Wimmer, H. Schomerus, and C. W. J. Beenakker, arXiv:1206.6687. and J. K. Furdyna, 

1810.00426

1

1810

Non-Born effects in scattering of electrons in a clean conducting tube

Quasi-one-dimensional systems demonstrate Van Hove singularities in the density of states $\nu_F$ and the resistivity $\rho$, occurring when the Fermi level $E$ crosses a bottom $E_N$ of some subband of transverse quantization. We demonstrate that the character of smearing of the singularities crucially depends on the concentration of impurities. There is a crossover concentration $n_c\propto |\lambda|$, $\lambda\ll 1$ being the dimensionless amplitude of scattering. For $n\gg n_c$ the singularities are simply rounded at $\varepsilon\equiv E-E_N\sim \tau^{-1}$ -- the Born scattering rate. For $n\ll n_c$ the non-Born effects in scattering become essential despite $\lambda\ll 1$. The peak of the resistivity is asymmetrically split in a Fano-resonance manner (however with a more complex structure). Namely, for $\varepsilon>0$ there is a broad maximum at $\varepsilon\propto \lambda^2$ while for $\varepsilon<0$ there is a deep minimum at $|\varepsilon|\propto n^2\ll \lambda^2$. The behaviour of $\rho$ below the minimum depends on the sign of $\lambda$. In case of repulsion $\rho$ monotonically grows with $|\varepsilon|$ and saturates for $|\varepsilon|\gg \lambda^2$. In case of attraction $\rho$ has sharp maximum at $|\varepsilon|\propto \lambda^2$. The latter feature is due to resonant scattering by quasistationary bound states that inevitably arise just below the bottom of each subband for any attracting impurity.

cond-mat.mes-hall

cond-mat

Non-Born effects in scattering of electrons in a clean conducting tube Condensed-matter physics laboratory, National Research University Higher School of Economics, Moscow 101000, Russia, and L. D. Landau Institute for Theoretical Physics, Moscow 119334, Russia A.S. Ioselevich∗ N.S.Pescherenko† Moscow Institute of Physics and Technology, Moscow 141700, Russia and Skolkovo Institute of Science and Technology, Moscow 121205, Russia (Dated: October 2, 2018) Quasi-one-dimensional systems demonstrate Van Hove singularities in the density of states νF and the resistivity ρ, occurring when the Fermi level E crosses a bottom EN of some subband of transverse quantization. We demonstrate that the character of smearing of the singularities crucially depends on the concentration of impurities. There is a crossover concentration nc ∝ λ, λ (cid:28) 1 being the dimensionless amplitude of scattering. For n (cid:29) nc the singularities are simply rounded at ε ≡ E − EN ∼ τ−1 -- the Born scattering rate. For n (cid:28) nc the non-Born effects in scattering become essential despite λ (cid:28) 1. The peak of the resistivity is asymmetrically split in a Fano-resonance manner (however with a more complex structure). Namely, for ε > 0 there is a broad maximum at ε ∝ λ2 while for ε < 0 there is a deep minimum at ε ∝ n2 (cid:28) λ2. The behaviour of ρ below the minimum depends on the sign of λ. In case of repulsion ρ monotonically grows with ε and saturates for ε (cid:29) λ2. In case of attraction ρ has sharp maximum at ε ∝ λ2. The latter feature is due to resonant scattering by quasistationary bound states that inevitably arise just below the bottom of each subband for any attracting impurity. I. INTRODUCTION In this study we consider clean multichannel quasi-one- dimensional metallic systems: wires, tubes, strips etc. We revisit a seemingly well-understood problem of semi- classical (i.e. without localization effects) resistivity for such systems in the presence of weak short-range impu- rities at low concentration. It is well known, that this resistivity (as well as the density of states at the Fermi level) has a square-root Van Hove singularities as a func- tion of the Fermi level position E, occurring when E crosses a bottom EN of certain subband of transverse quantization1. These singularities are expected to be smeared due to scattering of electrons by impurities and (at least in the Born approximation) the width of the peak ΓB ∼ τ−1 min can be estimated as an electronic scat- tering rate at maximum of resistivity. This smearing was theoretically studied within the self-consistent Born ap- proximation by different groups of authors2 -- 5,7,8. We demonstrate that the above picture is valid only if the concentration of impurities is relatively high while for low concentration due to specifics of the quasi-one- dimensional systems the non-Born effects become essen- tial despite the nominal weakness of scattering. These effects lead to dramatic restructuring of the Van Hove singularities. Complex asymmetric features were experimentally ob- served in many quasi-one-dimensional systems, such as nanotubes (both single-wall9,10 and multi-wall11,12 ones). ∗e-mail: [email protected] †e-mail: [email protected] These features were attributed to Fano resonance13, aris- ing due to interference of the scattering at some narrow resonant state with the scattering at background con- tinuum. The E-dependence of resistivity ρ at the Fano resonance is usually described by the formula ρ(E) ∝ (E − EN + qΓ/2)2 (E − EN )2 + (Γ/2)2 (1) with phenomenological parameters q and Γ (see, e.g.,14). There were attempts9 -- 11 to fit the experimental data on the Van Hove singularities in nanotubes with the for- mula (1) with an appropriate choice of Γ and q. We will show, however, that this phenomenological expression is not sufficient to describe the entire zoo of possible ρ(E) shapes. In this paper we will give a microscopic deriva- tion of the actual ρ(E) dependence. The main ingredient of our theory is the non-Born effects in scattering. A. Statement of the problem In this paper we restrict our consideration to only one simple realization of quasi-one-dimensional system: a single-wall metallic tube in a strong longitudinal mag- netic field. The zero (or weak) field case is more difficult for theoretical study because of the chiral degeneracy of the electronic states that is only lifted due to an inter- action with an impurity. Other quasi-one-dimensional variants such as a conducting strip involve some addi- tional complications due to nonequivalence of positions of different impurities with respect to the nodes of the wave function. All these effects are quite interesting and they will be discussed elsewhere. 8 1 0 2 p e S 0 3 ] l l a h - s e m . t a m - d n o c [ 1 v 6 2 4 0 0 . 0 1 8 1 : v i X r a Besides the simplicity of the theoretical interpretation the case of strong magnetic field is convenient practically since the changing of magnetic field is an effective in- strument for tuning the distance E − EN , so it is easy to sweep the Van Hove singularity in a controllable way. 2 Oscillations of the longitudinal resistivity with the magnetic flux Φ threading the tube is a well known ef- fect that was experimentally observed in various tubes and wires (especially semimetallic)15 -- 17. These oscilla- tions are the direct manifestation of the Aharonov-Bohm effect18 -- the interference of electronic waves with oppo- site chiralities. From the semiclassical point of view it is instructive to write the resistivity ρ in the form of Fourier series: (cid:32) ∞(cid:88) (cid:33) ρ = ρ0 1 + An cos(πnΦ/Φ0) (2) n=1 where Φ0 = πc/e = ch/2e is the flux quantum. The oscillations can be observed in both dirty and clean sys- tems. As it was shown in a seminal paper by Altshuler, Aronov and Spivak19 in dirty (diffusive) systems the odd- n harmonics of the Aharonov-Bohm oscillations (2) are washed out due to strong variations in the length of dif- ferent diffusive trajectories that lead to randomisation of the non-magnetic part of the phases of electronic wave- functions. The even-n harmonics -- the oscillations asso- ciated with a special sort of trajectories (the ones contain- ing closed topologically non-trivial loops on the cylinder) survive the randomisation. This effect was observed in experiments (see20,21). The odd-n harmonics are in gen- eral very fragile: they may be suppressed also in nomi- nally clean systems17 due to the fluctuations of the tube's parameters: e.g., the radius22. The even-n harmonics are less fragile but still, in the presence of any kind of disor- der the amplitudes An rapidly decrease with n so that the oscillations in the imperfect systems usually look roughly harmonic. It is not the case for the geometrically perfect clean systems where An decrease with n only as n−1/2 so that the series diverges at Φ → 2M Φ0 with integer M . This divergency is nothing else but the Van Hove singularity (see, e.g.,22). So, the shape of oscillations is very different for perfect and for imperfect systems (see Fig.1). It this work we concentrate on geometrically perfect tubes with low concentration of weak short-range impu- rities, where one can expect strongly unharmonic oscilla- tions dominated by the Van Hove singularities as in the upper panel of Fig.1. Thus, we consider a single-wall tube of radius R threaded by magnetic flux Φ. The tube is supposed to be cut from a sheet of two-dimensional metal with simple quadratic spectrum23 of electrons E = 2k2/2m∗. Im- purities are embedded in this sheet with two-dimensional concentration n2. They are supposed to be short-range and weakly scattering ones. FIG. 1: ρ(Φ) dependence for clean and dirty cases. Top: clean case, ρ(Φ) is periodic with a period 2Φ0 and Van Hove square root singularities are present for Φ = 2nΦ0. Bottom: dirty case, ρ(Φ) is periodic with a period Φ0 - odd harmonics are suppressed. B. Principal approximations It is convenient to measure all the energies in the units of E0 = 2/2m∗R2 and we will assume the semiclassical condition throughout this paper: ε0 (cid:29) 1, ε0 ≡ E/E0, N ∼ ε1/2 0 (cid:29) 1. (3) where N is the label of a subband whose bottom is the closest to the Fermi level and has the meaning of the number of open channels in the system. The magnetic field is assumed to be strong enough so √ that the splitting between EN and E−N is larger than the ε0 width of the peaks Γ. Besides that, the parameter 2 √ should not be close to any integer K to avoid resonance between the subbands with m = N and m(cid:48) = N ± 2 ε0. All the interesting effects associated with the Van Hove singularities occur in the range where ε (cid:28) 1, ε ≡ (E − EN )/E0. (4) There are two important dimensionless small parame- ters in this problem: (i) The dimensionless concentration of impurities n ≡ n2(2πR)2, n (cid:28) 1. (5) It is assumed to be small which in particular means that the average distance between impurities is larger than the transverse size of the system. (ii) Dimensionless scattering amplitude Λ2d = λ − iλ2 (6) of the background two-dimensional problem (λ > 0 cor- responds to repulsion, λ < 0 -- to attraction). It is also supposed to be small: λ (cid:28) 1. (7) The imaginary part of complex Λ2d in (6) is necessary to fulfil the unitarity requirement24 (the optical theorem): Im Λ2d = −Λ2d2. (8) Actually we will need this imaginary part only for proper treatment of quasistationary states arising in the case of attracting λ < 0. In all other cases we can simply put Λ2d → λ. (iii) The third condition is imposed on the length L of the tube: it should satisfy the inequality l(ε) (cid:28) L (cid:28) Lloc(ε), (9) where l(ε) is the mean free path and Lloc ∼ N l(ε) is the localization length. The large parameter N (cid:29) 1 assures at least the possibility for inequality (9) to be fulfilled. Indeed, it is very well known that weak localization effects in quasi-one-dimensional systems lead (in the ab- sence of inelastic processes) to an ultimate localization on all electronic states. However, for the tubes with lengths L in the range (9) the localization corrections are still small so that the results obtained throughout this paper are well justified and should give a valid expressions for the resistivity ρ(ε). Moreover, these results provide a possibility to estimate the dependence of the localization length on the parameter ε: Lloc(ε) =(cid:2)e2ρ(ε)(cid:3)−1 3 approximation. In particular we demonstrate that in this approximation the resistivity vanishes exactly at the Van Hove singularity. In Section VII we estimate the effects of interference between scattering events at different im- purities, resolve the zero-resistivity paradox of the pre- ceeding section and estimate the minimal resistivity. In Section VIII we discuss the inhomogeneous broadening of the peaks in the resistivity that arise due to resonant scattering at quasistationqry states. In Section IX we explore the effects that should arise if impurity with dif- ferent effective scattering amplitudes are present in the system. In Section X we summarize the results and out- line the directions of future research. In the Appendices A and B we evaluate the behaviour of the system in the immediate vicinity of the Van Hove singularity (where the single-impurity approximation breaks down) using the self-consistent approximation and explain the effect of a "catastrophic drop" (almost a jump!) of resistivity just below the smeared singularity. II. THE PRINCIPAL RESULTS The number of physical scenarios and distinct ranges of parameters considered in this paper is large. Therefore we find it reasonable to start with the list of different regimes and principal results. A. The Born approximation Away from the Van Hove singularities (at ε (cid:29) 1) the applicability of the Born approximation requires only the condition (7). Here the system behaves simply as a classic piece of the background two-dimensional material. The density of states, the resistivity and the scattering rate (the latter is being measured in units of E0) are (cid:18) λ (cid:19)2 . (10) ν0 = m∗R, ρ0 = 1 e2ε0 1 τ0 , 1 τ0 = 2n π . (11) C. The structure of the paper The structure of the paper is as follows: In Section II we bring together all the principal results of the paper. In Section III we briefly remind the well-known facts about quantum mechanics of an electron on a tube threaded by magnetic field. In Section IV we discuss the scattering of electrons near the Van Hove singularity within the Born approximation. In Section V we discuss the non-Born effects for scattering of electrons in the cylinder geome- try and derive the corresponding renormalization of the scattering amplitude. In the case of attracting potential we find the poles in the scattering matrix that are related to quasistationary states under the bottom of each sub- band. In Section VI we consider the manifestations of the non-Born effects in resistivity in the single-impurity In all cases the main contribution to the current comes from the one-dimensional subbands with labels m that are not very close to N because for m ≈ N the longitu- dinal velocity of electrons with energy E tends to zero. However, the role of the N -band becomes very important near the singularity when E → EN . Indeed, when the total density of states (cid:18) (cid:19) √ θ(ε) ε π ν(ε) = ν0 1 + , (12) is dominated by the second term (the contribution of the resonant N -band), the electrons from the current- carrying bands (those with labels m ∼ N/2) are scat- tered predominantly to the resonant one (with m = N ). Near the singularity the properties of electrons from the resonant band differ from the properties of all others. For a general quasi-one-dimensional system there are in prin- ciple two distinct scattering amplitudes and correspond- ing rates: λ1 and τ−1 1 (ε) describe the scattering from the current-carrying bands to the resonant one while λ2 and τ−1 2 (ε) correspond to scattering within the resonant band. The rate τ−1 1 (ε) directly determines the mean free path and the resistivity ρ(ε) = 1 1 e2ε0 τ1(ε) = 1 e2N l(ε) , l(ε) = N τ1(ε), (13) where we have used the obvious relations l ∼ vF τ and vF ∼ E1/2 ∼ N . The rate τ−1 2 (ε) is responsible for smearing of the singularity in the density of states and is relevant only in the immediate vicinity of the singularity. However, we show in Section IV that for the case of a tube τ2(ε) = τ1(ε) ≡ τ. (14) We will show, that close to the singularity the Born ap- proximation remains valid only if the dimensionless con- centration n of impurities is relatively high. Let us first assume that this condition is fulfilled and estimate the width of the smeared Van Hove singularity. The scattering rate is proportional to the density of final states so that From (19) it is clear that the energy scale εnB = (λ/π)2 (cid:28) 1, 4 (20) measures the range near the singularity where the non- Born effects are considerable. In particular, we see that if, due to low concentration of impurities, the Born scat- tering rate is low enough: ΓB < εnB, (21) then the non-Born effcts have chance to come into play in the range Γ < ε < εnB. Substituting the explicit formulas (17) and (20) to the condition (21), we arrive at the criterion n < nc, nc = λ/π. (22) of the breakdown of the Born approximation in the vicin- ity of the singularity. Under the opposite condition the Born approximation is sufficient for all ε. It is convenient to rewrite (19) in the form λ → Λ(
) ≈ λ 1 − [sign(λ) − iλ](−
)−1/2 ,
≡ ε εnB . (23) 1 τ1(ε) = 1 τ0 ν(ε) ν0 . C. The non-Born effects in resistivity: repulsing (15) impurities The width ΓB of the peak in the density of states (and in the resistivity at the same time) may be estimated from the condition τ−1 1 (ε ∼ ΓB) ∼ ΓB. (16) As a result, the Van Hove singularity is smeared on the scale ε ∼ ΓB (cid:17)2/3(cid:18) λ ΓB ∼(cid:16) n (cid:17)2/3(cid:18) λ (cid:16) n π π (cid:19)4/3 (cid:29) 1 (cid:19)4/3 (cid:29) ρ0. τ0 , π π B ∼ 1 ρmax e2ε0 At low concentration of impurities n (cid:28) nc the shape of the ρ(
) dependence in the vicinity of Van Hove singulari- ties is strongly modified by non-Born effects in scattering. A narrow peak at
= 0 is replaced by a broad one slightly above the bottom -- with the maximum at
∼ 1 nB ∼ εnB, independent of the concen- and the width Γ(+) tration n. The shape of this broad peak can be found (cid:16) n explicitly: (cid:17)(cid:18) λ (cid:19) 1 F (
),
≡ ε/εnB, (24) = 2 π π τ (ε) (17) (18) F (
) = (
1/2 +
−1/2)−1 The maximal (in the range ε > 0) resistivity (cid:16) n (cid:17)(cid:18) λ (cid:19) π π (25) (26) B. The origin of non-Born effects ρmax(+) nB ∼ 2 e2ε0 Fmax (cid:28) ρmax B , The origin of the special importance of non-Born ef- fects in quasi-one-dimensional systems is renormalization of the scattering matrix that is dramatically enhanced near a Van Hove singularity. In the case of tube this ma- trix can be effectively reduced to a single complex con- stant Λ(ε) that can be found from the Dyson equation. As a result (cid:26) (cid:27)−1 λ → Λ(ε) ≈ λ √ 1 − Λ2d ε π (19) is reached at
=
max, where
max = 1 Fmax = 1/2, (27) The function F (
) is shown in Fig. 2. At
(cid:29) 1 it has asymptotics F (
) ≈
−1/2 that corresponds to the stan- dard Van Hove singularity. The height of the broad peak is much less than it would be within the Born approx- imation but still is much higher than the background resistivity ρ0. 5 It can be shown that, besides the true bound state with the energy below the bottom of the lowest subband of the electronic spectrum of the cylinder, a weakly attracting short-range impurity produces an infinite series of qua- sistationary states: one such state below the bottom of each band. In this paper we concentrate on the quasista- tionary states associated with the quasiclassic subbands (those, with large N (cid:29) 1). In particular we show that for λ < 0 the scattering amplitude (23) has a pole at
= −1 + 2iλ (or at ε = (−1 + 2iλ)εnB in other nota- tion). This pole corresponds to a quasistationary state with a relatively small decay rate. In the case of cylin- der these poles are identical for all impurities and, since electrons can be scattered by these resonances, the latter lead to formation of sharp maxima in resistivity for ε < 0 and λ < 0: F (
) ≈ 1 (1 −
−1/2)2 for 1 −
(cid:29) λ, , 4 (1 −
)2 + 4λ2 , for 1 −
(cid:46) λ, (31)  This result is illustrated by Fig. 3: FIG. 3: The same as in Fig. 2 but for attracting impurities. The sharp maximum at
< 0 arises due to resonant scattering at quasistationary states. The maximal (in the range
< 0) resistivity is reached at
= −1, (32) ρmax(−) ∼ 1 e2ε0 the width of this maximum is Γ(−) nB 2n π2 , nB = 4λεnB. The physical origin of the quasistationary states that exist slightly below each of the subbands is as follows. Semiclassical trajectories of electrons with energies near the bottom of subband are almost closed; if an electron with such energy has passed near certain impurity once then it will do so again, and many times. Therefore the attraction to impurity is strongly enhanced and the bound state is formed. An alternative way of thinking is just to neglect in the leading approximation all the tran- sitions from the resonant band to all others. The arising strictly one-dimensional problem grants a bound state FIG. 2: Dependence of the resistivity on the position of the Fermi level for repulsing impurities in the case of strong non- Born regime low concentration of impurities n (cid:28) nc. Note that ρ(
) vanishes as
→ 0: it is an artefact of the single- impurity approximation that is not applicable at
(cid:46)
min (cid:28) 1. The horizontal asymptote (dashed line) corresponds to ρ = ρ0. The behaviour of the resistivity above the Van Hove singularity (for ε > 0), described by (24), does not de- pend on the sign of λ, it is the same for attracting and repulsing impurities. It is not the case for the range ε < 0 below the singularity. For repulsing impurities we obtain (cid:16) n (cid:17)(cid:18) λ (cid:19)2 π π 1 τ (ε) = 2π F (
), (28) F (
) = [1 + (−
)−1/2]−2, (29) so that ρ(ε) monotonically increases with ε and satu- rates at ρ = ρ0. It is easy to see that, as it formally follows from (29), the resistivity ρ(ε) should vanish for ε → 0 from either side. Indeed, for
(cid:28) 1 F (
) ≈
1/2 F (
) ≈ −
. (30) Of course we immediately suspect that in reality the decrease of resistivity will be ultimately stopped by some additional effect (and this is indeed so, see below). But anyway, a dramatic suppression of resistivity in the nar- row vicinity of the Van Hove point is an important phe- nomenon. Physically it is a result of destructive inter- ference of partial electronic waves with different winding numbers. D. Attracting impurities, quasistationary states and resonant scattering As we have already mentioned in previous section, the behaviour of resistivity above the singularity is identical for repulsing and attracting impurities. However, below the singularity the attracting impurities introduce some nice additional physics. for arbitrary weak attraction. Taking the transitions to nonresonant bands into account perturbatively leads to the finite decay rate of the state. E. The minimum of resistivity 6 All the effects described above are the single impurity ones. Their origin is the coherent multiple scattering by the same impurity which fact is manifested in the linear dependence of resistivity on the concentration n. To reveal the mechanism that limits the suppression of resistivity at
→ 0 and to estimate the resistivity at its minimum one has to find the scattering rate τ−1(ε) from (28) in the range ε (cid:28) εnB: (cid:18) (cid:19) = ε τ0 τ (ε) 1 + λ√ θ(ε) ε FIG. 4: Thin conducting tube, threaded by magnetic field H. Impurities (shown as stars) are embedded in the tube. Electrons live on the surface of the cylinder. (33) Em = E0(m + Φ/2Φ0)2, E0 = 2 2m∗R2 (40) where m ∈ Z is the azymuthal quantum number, k is the momentum along the cylinders axis and Φ0 = πc/e = ch/2e is the flux quantum. Em has the meaning of po- sition of the bottom of m-th one-dimensional subband. Actually we have introduced the magnetic field as a tool of easy shifting of the Fermi level in the system but all the physics described below is present already in the case H = 0. The density of states in each subband (cid:19) = (cid:18) δ (cid:90) dk (cid:115) 2π 2 2π E − Em − k2 2m∗ m∗ 2(E − Em) θ(E − Em), (41) νm(E) = = The factor 2 arises because the equation E−Em− k2 0 has two roots k = ±(cid:112)2m∗(E − Em). 2m∗ = The width ΓnB of the peak in the density of states can be estimated from the condition ΓnB ∼ τ−1 (ε ∼ +ΓnB) , and we get ΓnB ∼ εmin ≡ (n/π)2 (cid:28) εnB. (34) (35) Note that this width does not depend on λ. At ε < 0 the resonant contribution to the density of states rapidly drops on the same energy scale so that the factor ν(ε) becomes of order of ν0 already at ε ∼ −εmin. As a result, the resistivity has a minimum at ε = εdip, where εdip < 0, εdip ∼ εmin. (36) The scattering rate and the resistivity at minimum are ∼ n3, 1 τdip ρdip ∼ n3 e2ε0 , (37) and do not depend on the scattering amplitude λ. Thus, there is a deep and narrow universal minimum of resis- tivity slightly below the bare Van Hove singularity, the resistivity in the minimum depends on n superlinearly. III. AN IDEAL TUBE We consider a tube of radius R threaded by a magnetic flux Φ = πR2H (the magnetic field H is oriented along the axis of a cylinder z). Electrons in the tube have the following spectrum and wave functions: ψmk(φ, z) = (2π)−1/2 exp{ikz + imφ}, Emk = 2k2 2m∗ + Em, (38) (39) FIG. 5: Spectrum of an electron on a surface of an ideal cylinder. Subbands of the transverse quantization are shown. The Fermi level E crosses all the subbands with m ≤ N . The total density of states ν(E) = Im g(E), (42) νm(E) = − 1 π g(E) ≡ G(0) E (0, 0) = (cid:88) (cid:90) dk (cid:88) (cid:88) 2π m m = m 1 E − Ekm + i0 = (cid:115) m∗ 2(Em − E) , (43) G(0) E (0, 0) being the one-point retarded Green function of an ideal wire. Strictly speaking, the real part of g diverges. The recipe how to deal with this divergency will be discussed somewhat later. Now we just mention that the divergent part is energy-independent and therefore can be removed by a constant shift of the energy. In the main part of this paper we will measure all en- ergies in the units of E0 and all distances in the units of 2πR: E ≡ E0ε0, E − Em ≡ E0εm, νm(ε) = 1√ εm θ(εm), g(ε) = (cid:88) m νm(E) ≡ νm(ε) 2πRE0 , (44) π√−εm . (45) We are interested in semiclassical case when E0 (cid:28) E or ε (cid:29) 1. Then, in the leading semiclassical approximation ∞(cid:88) m=−∞ ν(ε) = νm(ε) ≈ ν0 = (cid:90) ε0 0 (cid:112)εm(ε0 − εm) dεm = π. (46) This result is valid for all ε except narrow intervals in the vicinity of points where εm = 0 for some m. The condition of strong magnetic field reads εN − ε−N ∼ √ ε0Φ/Φ0 (cid:29) Γ, where Γ is the broadening of peaks and N = the closest to Fermi level E subband. √ (47) ε0 denotes In the entire range of variation of
one can write (cid:18) (cid:19) ν(0)(ε) ≈ ν0 1 + √ θ(ε) π ε where we have introduced ε ≡ εN for brevity. Under the semiclassical condition N (cid:29) 1 the result (46) is not valid in the vicinity of the Van Hove singularity for ε (cid:46) 1 where the second -- resonant -- term in (48) is anomalously large. We see that for ε (cid:28) 1 (49) the inequality νN (ε) (cid:29) ν0 holds: the density of states is indeed dominated by the second term in (48) -- the contribution of the N -subband. Note that in the semi- classical limit N (cid:29) 1 the different peaks in the function ν(ε) are strictly identical. ε > 0, 7 IV. BORN SCATTERING BY SHORT-RANGE IMPURITIES Our first step is finding the longitudinal resistivity of the tube using the Drude and Born approximations. We consider weak short range impurities with the hamilto- nian H = H0 + V δ(r − r0), H0 = −∇2/2m∗ (50) where δ(r) ≡ 1 R δ(z − z0)δ(φ − φ0) is a two-dimensional delta-function and r0 denotes the position of the impurity on the wall of the tube. Let us find the self-energy for an electron (cid:90) dk(cid:48) (cid:88) 2πRn(2) imp Σkm = E0 m(cid:48) 2π Vkk(cid:48)mm(cid:48)2G(0) k(cid:48)m(cid:48) (51) Since for the short range potential (50) V 2πR exp{i(m − m(cid:48))φ0 + i(k − k(cid:48))z0}, Vkk(cid:48)mm(cid:48) = Vkk(cid:48)mm(cid:48)2 ≡ V 2 depends neither on km, nor on k(cid:48)m(cid:48), we conclude that Σkm = Σ(Ekm) is a function of only the total energy E. In our dimensionless variables we get: (52) Σ(ε) = g(ε) 2πν0τ0 , (53) where τ0 is the dimensionless scattering time for an elec- tron away from the resonance (i.e., for ε (cid:29) 1): τ−1 0 = m∗V 2n2 E0 = 2n(λ/π)2, (54) the dimensionless Born scattering amplitude λ = m∗V /2, λ (cid:28) 1, (55) is assumed to be small and may have either signs (the positive sign corresponds to repulsion, the negative -- to attraction). The dimensionless concentration n is also assumed small. 1 τk,m = 1 τ (Ekm) = −2Im Σ = 1 τ0 ν(ε) ν0 . (56) For point impurities the scattering is isotropic and there- fore the transport time coincides with the simple decay time. Thus, if (49) is fulfilled, the particle is scattered pre- dominantly (though not completely) to the upper band. In particular, if the particle was already in the upper sub- band then the scattering event most probably will not remove it from there. It means that in the zero approx- imation the upper subband is almost decoupled from all others. , (48) The Born decay rate Under the condition (49) the electrons in the N - subband states have low longitudinal velocity and there- fore do not contribute much to the current. The latter is dominated by the states in all other bands. However, the singularity in the N -band is manifested also in the resis- tivity ρ through the scattering rate that is proportional to the density of the final states on the Fermi surface: ρ ρ0 = ν(ε) ν0 , ρ0 = 1 e2ε0τ0 . (57) These final states predominantly belong to the N -band. A. Smearing of the Van Hove singularities within the Born approximation It is clear that the scattering should somehow smear the singularities both in the density of states and in the resistivity. In this section we will discuss the mechanism of this smearing within the Born approximation. The condition of its applicability will be discussed in the fol- lowing section. Within the Born approximation one can write (see, e.g.,6,8) (cid:104)ν(ε)(cid:105) = ν(0)[ε − Σ(ε)] (58) where Σ(ε) is given by (53). We are interested in the behavior of (cid:104)ν(ε)(cid:105) in the vicinity of the Van Hove singu- larity where ε (cid:28) 1 and the scattering is dominated by the resonant band. It is convenient to discuss this prob- lem separately for the cases ε > 0 (above the singularity) and ε < 0 (below the singularity). 1. Above the Van Hove singularity Here the self-energy is almost purely imaginary: the scattering is more important than the energy shift. The scattering obviously leads to decay of the plane waves and the decrement of this decay is just τ−1 given by the formula (56). For τ−1 (cid:28) ε the average density of states is almost insensitive to the scattering. In the narrow vicinity of the singularity, for τ−1 ∼ ε, the scattering becomes effectively strong, the density of states is strongly changing on the scale of the width of the relevant states. Thus, at τ−1 ∼ ε the density of states saturates and we conclude that the corresponding peaks are smoothed at ε ∼ εmin ∼ τ−1. However, τ−1 itself depends on ε and so we arrive at the self-consistency condition: 1 τ (ε) = 1 τ0 ν(ε) ν0 √ ∼ 1 πτ0 ε ≈ ε (59) from where we can easily get ε ∼ εmin = (2πτ0)−2/3 = [(λ/π)2(n/π)]2/3, (60) τ0 τmin ∼ νmax ν0 ∼ ρmax ρ0 ∼ (cid:34)(cid:18) λ (cid:19)2(cid:16) n (cid:17)(cid:35)−1/3 π π 8 . (61) 2. Below the Van Hove singularity Now we have to find the density of states for ε < 0. It seems clear that for ε (cid:46)
min the value of ν(
) can not change considerably so that one can expect ν(ε) ∼ νmax, τ (ε) ∼ τmin, for ε (cid:46) εmin (62) On the other hand, in the range ε (cid:29) εmin the correction to the density of states can be found with the help of perturbation theory which gives ν(ε) − ν0 ∼ ν0 (−ε)−3/2 τ0 ∼ ν0 (cid:18) εminε (cid:19)3/2 (cid:28) ν0. (63) It is important to note that the correction (63) is rela- tively small already for ε (cid:46) εmin. It means that ν(−εmin) ∼ ν0 (cid:28) ν(εmin) (64) and direct smooth matching of (63) and (62) is impossi- ble! To resolve this paradox one should in principle go be- yond the estimates made above, and accurately solve the problem in the range ε (cid:46) εmin. However, for a qual- itative understanding it is enough to note that there is practically only one scenario for such a giant drop in the density of states: a "quasifold" -- an inflection point with almost vertical slope, see Fig. 6. FIG. 6: The "quasifold": in the vicinity of the bifurcation point ε = εbi the slope of the curve ρ(ε) is anomalously steep. In the dependence ν(ε) at some point εbi there should be very large positive first derivative ν(cid:48)(εbi) (cid:29) ν(εmin)/εmin zero second derivative and rather small third derivative. An example of such a behaviour is pro- vided by the results of the self-consistent Born approxi- mation, given in Appendix A. Although these results can not be taken too seriously (since the self-consistent Born approximation is not rigorous), the main message seems  to be reliable: the entire domain ε ∼ εmin is split into two basic subdomains: ε < εbi where ν ∼ ν0 and ε > εbi where ν ∼ ν(+εmin) (cid:29) ν0. Between these two subdo- mains there is a narrow intermediate layer around εbi in which ν(ε) undergoes a dramatic change. Then the results can be roughly summarized as follows: ρ(ε) ρ0 = π 1 + √ 1 , ε ∼ λ−2/3n−1/3, ∼ 1, for εmin (cid:28) ε, ε > 0, for εbi < ε (cid:46) εmin, for ε < εbi, (65) with certain εbi < 0, εbi ∼ εmin. A schematic plot of (65) is shown in FIG.7. FIG. 7: The shape of a smeared Van Hove singularity within the Born approximation. V. BEYOND THE BORN APPROXIMATION The above considerations seem plausible and straight- forward. However, the analysis below shows that they are only applicable if the concentration of impurities is high enough, i.e. for n (cid:29) nc ∼ λ. For n (cid:28) nc the scattering that determines the form of smeared Van Hove singulari- ties is strongly modified by non-Born effects that dramat- ically grow upon approaching the singularity. We start our discussion from the properties of an exact amplitude of scattering by a single short-range impurity, placed on the wall of the tube. A. A single impurity problem in two dimensions: non-Born effects Properties of short-range impurities or defects in two- dimensional systems are well studied. In this subsection we briefly remind the main facts. 9 In particular, it is known that a weakly-attracting short-range impurity always forms a bound state24. Writ- ing the hamiltonian of the system in the form (50) with λ < 0 one finds that there is a single bound state with small binding energy (cid:18) (cid:19) bound ≈ − 2 E(2d) m∗a2 0 exp − π λ , (66) where a0 is the ultraviolet cutoff ("radius of the delta- function") The wave-function of the ground state ψ0(r) ∼ exp(−r/a(2d)), a(2d) = (2m∗E(2d) bound)−1/2, (67) being the radius of the ground state wave function. A scattering of a particle with positive energy E (cid:28) is isotropic. For r (cid:29) a0 one can write the "scatter- 2 0 m∗a2 ing wave-function" in the form24 ψp(r) = exp{i(p · r)} − iλH (1) 0 (pr), E = p2/2m∗, (68) where H (1) pr (cid:29) 1 one can write 0 (x) is the Hankel function. Moreover, for ψp(r) ≈ exp{i(p · r)} − λ exp(ipr), (69) (cid:114) 2 −iπpr The above results should be modified if one wants to go beyond the Born approximation. If the condition E (cid:28) U0 (or ka0 (cid:28) 1) is fulfilled then the scattering remains isotropic even beyond the Born approximation; it means that the scattering amplitude is still charac- terised by a single dimensionless constant: small real λ in the result (69) should be replaced by not necessar- ily small complex Λ -- the nonperturbative dimensionless scattering amplitude. The latter should obey the optical theorem: ImΛ = −Λ2, (70) hence the scattering amplitude can be parametrised by a single real constant λ (cid:16)(cid:112) (cid:17) Λ = λe−i arcsin λ ≡ λ 1 − λ2 − iλ . (71) In particular, for weak interaction (λ (cid:28) 1) Λ ≈ λ − iλ2. (72) Note that parameter λ in (71) is related to the poten- tial amplitude V by formula (55) only for λ (cid:28) 1. In general case it is not true and λ is just a convenient pa- rameter for expressing the phenomenological scattering amplitude. B. A single impurity problem on a cylinder: semiclassical treatment of non Born effects Let us place a single weakly attracting impurity on the surface of the cylinder. Clearly, there are two distinct cases with respect to the bound state of an electron: (i) Wide cylinder or strong scattering: R (cid:29) a(2d). In this case the bound state will not differ much from the purely two-dimensional case and the formula (66) applies. (ii) Narrow cylinder or weak scattering: R (cid:28) a(2d). This is an effectively one-dimensional case, the bound state can also be studied easily. In this paper we will be interested, however, not in the ground state but in the scattering matrix for an electron with energy E > 0 in the range E0, Ebound (cid:28) E (cid:28) 2 m∗a2 0 . (73) Under this condition the scattering process can be con- veniently described in semiclassical terms. To find the scattering amplitude beyond the Born approximation one has to solve the Dyson equation G(r1, r2) = G0(r1, r2) + G0(r1, r0)V G(r0, r2) for the retarded Green function defined as G = E − H + i0 G0 = E − H0 + i0 (cid:111)−1 (cid:110) (cid:110) (cid:111)−1 (74) (75) where r0 is the position of the impurity. In particular, putting r1 = r2 = r0 we arrive at the equation g = g0 + g0V g, g ≡ G(r0, r0) = g0 1 − V g0 where One can also write g0 ≡ G0(r0, r0) (76) (77) G(r1, r2) = G0(r1, r2) + G0(r1, r0)VrenG0(r0, r2) (78) with the renormalized scattering amplitude V Vren = 1 − V g0 (79) First of all we have to find the single-site g0 ≡ GE(0, 0). For our nontrivial topology one can write in the semi- classical approximation eπinΦ/Φ0GE(2πnR), (80) ∞(cid:88) g0 = n=−∞ GE(r) being the retarded Green function in an infinite two-dimensional metal. For n (cid:54)= 0 one can use the semi- classical approximation: GE(r) ≈ ei(pr+π/4). (81) (cid:114) 2 πpr (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)2 ν(ε) ν0 For the n = 0 term we have GE(0) = − im∗ 2 + C 10 (82) where C is a formally infinite real constant. This diver- gency is well known -- it means that the perturbation theory does not work well in spatial dimensions d ≥ 2 when applied to point-like impurities. This phenomenon is not specific for the cylinder geometry -- it is present in an infinite two-dimensional metal as well. Special meth- ods to deal with this divergence were developed already long ago. It was shown that in the case of isotropic scat- tering, accurate calculations lead to the substitution of the bare coupling constant λ by the exact complex ampli- tude Λ of scattering by the same impurity in the infinite two-dimensional metal. Thus, for the fully renormalized scattering amplitude Λ(ren) in the case of cylinder we get Λ(ren)(ε) = Λ 1 + Λg(ε)/πν0 , (83) where g(ε) ≡ g0 is given by the formulas (80), (81), (82) where the infinite constant C is discarded. As a result, we arrive at the expression (45). Consequently, the scat- tering rate is also renormalized: 1 τ (ε) = 2n π2 Λ 1 + Λg(ε)/πν0 . (84) For small λ the discussed renormalization is only essential in the vicinity of some Van Hove singularity so that we can use asymptotics g(ε)/πν0 ≈ π−1(−ε)−1/2 and for small λ (cid:28) 1 one can write Λ(ren)(ε) ≈ λ − iλ2 1 + (λ − iλ2)(−ε)−1/2/π . (85) The importance of the renormalization of the scattering matrix in the systems with the singularity in the density of states (e.g., superconductors) that can even lead to formation of bound states was discovered and explored in details already in 60-ies (see25 -- 29). for ε (cid:46) εnB, where It is clear that the non-Born effects first come into play εnB = (λ/π)2, (86) so that it is sometimes convenient to use the "normal- ized" energy:
≡ ε/εnB. (87) Note that for ε (cid:28) εnB the scattering amplitude for- mally vanishes: Λ(ren) ≈ π(−ε)1/2. It means, in partic- ular, that exactly at the van Hove singularity a quasi- one-dimensional system tends to become an ideal con- ductor with zero resistivity. In the following section we will demonstrate that for finite concentration of impuri- ties the resistivity remains finite, though very small: it is proportional not to n, but to n3. VI. SINGLE-IMPURITY NON-BORN EFFECTS or, in dimensional variables 11 (91) IN RESISTIVITY Physically the effect of renormalization is manifested in the scattering time τ (ε) in which λ should be replaced by Λ(ren)(ε). Similar to the Born case, for τ−1(ε) (cid:28) ε (88) the scattering is effectively weak (though non-Born!) so that only the single impurity effects should be taken into account and one can use the standard Drude formula with properly renormalized scattering time. In this Section we concentrate on this "weak non-Born scattering" regime. We will consider the cases of repulsing and attracting impurities separately. Certainly, there were some theoretical approaches to the non-Born effects in quasi-one-dimensional systems in the past. S. Hugle and R. Egger8 studied the smearing of Van Hove singularities within the self-consistent Born approximation similar to that described in Appendix A. In contrast with our work, instead of the quadratic spec- trum of electrons they considered more realistic linear spectrum, characteristic of carbon nanotubes. This dif- ference, however, is not essential, as far as one is in- terested only in the shape of the Van Hove singulari- ties: it may actually be reduced to redefinition of some constants. What is much more important, instead of considering individual impurities S. Hugle and R. Eg- ger introduced the disorder in the form of gaussian white noise. Such an approach does not allow to find the single- impurity non-Born effects that, as we have seen, are cru- cial at low concentrations n2 (cid:28) n(c) 2 . So, their results are applicable to impurities only at high concentration n2 (cid:29) n(c) 2 . A. Repulsing impurities For weak repulsive impurities (λ > 0, λ (cid:28) 1) the imaginary part of Λ can be neglected and we get (cid:18) (cid:19) ρ(ε) ρ0 = τ0 τ = =  Λ(ren)2 λ2 1 + √ 1 ε π θ(ε) = 1 1
1/2 +
−1/2 λ (1 +
−1/2)2 1 , , for
> 0, for
< 0, (89) This dependence is plotted in Fig. 2: So, for ε > 0 both the scattering rate and the resistiv- ity have smooth maxima at ε = εnB) with the value at maximum 1 τ (+) min = 2n2 m∗ λ, ρ(+) max ρ0 = 1 λ (cid:29) 1. For ε < 0 the scattering rate grows monotonically with growing ε and saturates at τ−1 = τ−1 for ε (cid:29) εnB . The non-Born effects somewhat suppress the resistiv- ity, compared to the Born results. For repulsing impu- rities this is true for all ε but the strongest effect is ex- pected for ε (cid:46) εnB. 0 B. Attracting impurities For attracting impurities the renormalized scattering amplitude has a pole in the complex plane of ε at ε = εnB(−1 + 2iλ), (92) close to the real axis. This fact indicates the existence of a quasistationary state. We have to take into account the imaginary part of Λ that keeps trace of the decay of this state: otherwise the pole would move to the real axis and there will be nonphysical divergence of amplitude. However, this is only necessary in the narrow vicinity of the resonance at ε = εnB. So we can write  = 1 λ 1
1/2 +
−1/2 1 (1 −
−1/2)2 , , 4 (1 −
)2 + 4λ2 , ρ(ε) ρ0 = τ0 τ = for
> 0, for
< 0, 1 −
(cid:29) λ, (93) for
< 0, 1 −
(cid:46) λ, This result is plotted Fig. 3. Thus, for ε > 0 (and also for ε < 0 but ε (cid:28) εnB) the behaviour of the renormalized scattering rate for at- tracting impurities is identical to that of repulsing ones. Their behaviours are very different, however, for ε < 0 (and not small ε (cid:38) εnB). While for repulsive impuri- ties both the rate τ−1 and the resistivity ρ smoothly and monotonically increase with ε, for attracting impurities they first grow, reach sharp maxima at ε = −εnB and only then decrease, saturating at τ−1 = τ−1 and ρ = ρ0 for ε (cid:29) εnB. The maximum has a Lorenzian shape: 0 ρ(ε) = ρ(−) max πΓhom 2 L(ε + εnB, Γhom), (94) L(x, γ) ≡ γ/2 π (x2 + (γ/2)2) . (95) The width of maximum (homogeneous broadening) 1 τ (+) min = 1 2λτ0 = nλ π2 , (90) Γhom ∼ 4λεnB = 4λ3 π2 (cid:28) εnB, (96) is relatively small. This decay is due to small probability of scattering to the bands other than the N -band. The height of the maximum is universal -- it does not depend on the strength of impurities λ. In dimensional variables: ρ(−) max = 4n2 e2m∗RE . (97) The scattering rate at maximum is even more universal: 1 τ (−) min = 1 λ2τ0 = 2n π2 = 4n2 m∗ . (98) VII. MULTI-IMPURITY EFFECTS. THE CENTRAL DIP IN RESISTIVITY. In the previous section we have implicitly assumed the concentration of impurities n to be so low that scattering amplitude at certain impurity could not be affected by the presence of all the others: τ−1(ε) (cid:28) ε. Let us first derive the condition that would justify this assumption. We have found that the non-Born effects are negligible for ε (cid:38) εnB. On the other hand, if one totally neglects the non-Born effects, then, as it follows from (59), the scat- tering effects lead to the saturation of both the density of states and the conductivity for ε (cid:46) εmin. These two facts taken together mean that for εnB (cid:28) εmin the non-Born effects do not have chance to show up at all. On the contrary, for εmin (cid:28) εnB the scattering only comes into play at ε (cid:28) εnB where the non-Born effects are already huge. Thus, looking at the expressions (61) for εmin and (86) for εnB we conclude that the non-Born effects are relevant for n < nc, where nc ∼ λ, (99) while for n > nc the Born approximation is justified for all ε and the results of section IV are applicable. In this Section we are going to study the effect of scat- tering at low concentration n (cid:28) nc but also at very low ε at the same time. We will show that the presence of other impurities ultimately becomes essential in the narrow vicinity of the Van Hove singularity -- at certain energy scale ε(nB) min . In the case of developed non-Born regime, for ε (cid:28) εnB we have Λg (cid:29) 1, so that (cid:12)(cid:12)(cid:12)Λ(ren)(ε) (cid:12)(cid:12)(cid:12)2 ≈ π2ε. (cid:18) √ 1 1 + π τ−1(ε) = 2εn (cid:19) We see that the rate 1/τ ceases to depend on λ and be- comes universal: independent on the characteristics of impurities: (100) 12 from either side and formally vanishes at ε = 0. Taken seriously, it would mean that exactly at singularity the system has zero residual resistivity. Of course, we ex- pect that taking scattering in account will remove this paradox. To demonstrate this, we have to incorporate the scat- tering in the result (101). Again, as in Section IV A we notice that the above calculations only make sense for τ−1(ε) (cid:28) ε, so that the dip in the resistivity predicted by (101) will be rounded at certain ε ∼ ε(nB) min , where ε(nB) min , however, is not given by (60) any more because the expression for the scattering time (101) differs from it has been changed by the non-Born effects. So, (56): the self-consistency condition τ−1(ε) ∼ ε for ε(nB) min reads τ−1(cid:16) (cid:17) (cid:113) ε = +ε(nB) min = 2n π min ∼ ε(nB) ε(nB) min , (102) from where immediately follows ε(nB) min = (n/π)2 . (103) Comparing (103) to (86) we see that, indeed, the scat- tering effects bring the renormalization of the amplitude Λ(ren)(ε) to stop at some small, but nonzero value. The results (100) and (103) were obtained under the assumption ε > 0 so we need yet to discuss the scattering effects for ε < 0. Here we get τ−1(ε) = 2nε (cid:28) ε, (104) which formally means that for negative ε the scattering does not affect the result (101) for all values of ε, down to ε = 0! This is, of course, not quite true because, due to scattering effects, the discontinuity in the density of states at ε = 0 should be smoothed and 1/τ (ε) should remain of the order 1/τmax also for ε < 0 in the range ε (cid:46) ε(nB) min . Thus, in the strongly non-Born domain n (cid:28) n(nB) we encounter the similar paradox as in the Born case at n (cid:29) n(nB). Namely, the above consideration gives non- matching estimates on the opposite sides of the interval ε (cid:46) ε(nB) min : τ−1 ∼ n2, n3, for ε > 0, ε ∼ n2, for ε < 0, ε ∼ n2. (105) (cid:40) The resolution of this paradox is also similar to that in bi ≡ the Born case: there is a quasifold at certain ε = ε(nB) min , (with qbi < 0, qbi ∼ 1) where the scattering qbiε(nB) rate undergoes a dramatic drop, so that  2nε, n2, √ 2n ε/π, bi for ε < ε(nB) for ε > ε(nB) bi for ε (cid:29) ε(nB) min , , , ε (cid:46) ε(nB) min , (106) It should be stressed that the scattering rate decreases as the Fermi level approaches the Van Hove singularity θ(ε) . ε (101) τ−1(ε) ∼ 13 aloc = (2εnB)−1/2 = πλ−1 is the radius of the localized state. So, naloc ∼ n/λ ∼ n/nc and, under condition n (cid:28) nc, the influence of other impurities typically is ex- ponentially small. However, this influence may be large in some rare non-typical configurations and we will esti- mate their contribution. Due to a rare local fluctuation two impurities may oc- cur at non-typically small distance r (cid:46) aloc from each other, resulting in a considerable splitting ∆(r) ∼ εnB of a pair of initially degenerate localized states. It leads to inhomogeneous broadening Γinhom ∼ (naloc)εnB ∼ n nc εnB (111) that prevails in the intermediate range of concentrations: λ2 (cid:28) n (cid:28) λ, while for lowest n (cid:28) λ2 the homo- geneous broadening is stronger. We should stress that the inhomogeneous broadening (111) exists already in the system where all impurities are identical (have the same λ). Naturally, the systems with dispersion of λ demonstrate much stronger inhomogeneous broadening. We will briefly discuss such systems in Section IX. (110) IX. SYSTEMS WITH DIFFERENT SORTS OF IMPURITIES and the weakest scattering is realized at some ε = ε(nB) dip below ε(nB) : bi ≈ 1 τmax 1 τ (ε = ε(nB) dip ) ∼ nε(nB) min ∼ n3, or, in dimensional variables 1 τmax ∼ (2πR)4[n2]3 m∗ . (107) (108) This result is supported by the calculations within the "self-consistent non-Born approximation", given in Ap- pendix B. Thus, we conclude that the minimal value of the scattering rate and, consequently, the minimal value of resistivity is attained a little bit to the left from the exact position of the Van Hove singularity, at ε = ε(nB) dip ∼ −n2 and ρmin = 1 1 e2RE τmax ∼ (2πR)4[n2]3 e2m∗RE (109) This minimal value depends neither on sign, nor on mag- nitude of λ and is much less than the standard resistivity: (cid:19)2 (cid:28) 1. (cid:18) n nc ρmin ρ0 ∼ n2 λ2 = The dependence ρ(
) near the minimum is shown in Fig 8. FIG. 8: The energy-dependence of resistivity near the mini- mum. Dashed line -- for n → 0, solid line -- for finite n. VIII. INHOMOGENEOUS CONTRIBUTION TO BROADENING OF THE RESONANT PEAK One could expect that in the case of attracting im- purities the scattering would lead also to broadening of the narrow resonant peak at ε = −εnB, so that Γ → Γhom + τ−1. But this idea is wrong since the corresponding electrons are localized at resonant states of certain individual impurities and, at low concentra- tion, have no chance to be scattered by some other im- purity. This statement is justified if naloc (cid:28) 1, where In realistic physical systems the impurities are not nec- essarily identical. They may be of different types and they may be situated not directly in the wall of the tube, but at some distance from it. As a result the effective scattering amplitudes Λi of different impurities may be different and random, with some distribution function P (λ) for real parameter λi (see (71)). The most impor- tant characteristic of this distribution is λ ≡(cid:112)(cid:104)λ2(cid:105) (112) What may be the consequences of such disorder? In the Drude approximation the only dependence of the resis- tivity ρ(ε) on Λ comes from the factor τ−1(ε). Since the contributions of different impurities to the resistivity are additive, one can write (cid:28) 1 (cid:29) τ (ε, λ) λ (cid:90) (cid:12)(cid:12)(cid:12)(cid:12) ∝ (cid:104)ρ(ε)(cid:105) ∝ Λ 1 + Λg(ε)/πν0 P (λ)dλ (cid:12)(cid:12)(cid:12)(cid:12)2 (113) For n (cid:29) λ the expression (113) can be expanded in small Λ, the non-Born effects are small and we return to the results of Section IV where one should substitute λ → λ. For n (cid:28) λ the scattering rate does not depend on λ in the range ε (cid:28) εnB ≡ (λ/π)2, therefore all the re- sults of Section VII also apply to the case of random λ in this range. The case ε ∼ εnB is non-universal, here the result of averaging may depend on explicit shape of the function P (λ). In particular, the contribution of the inhomogeneously broadened resonant peak can be eval- uated with the help of expressions (96), (94). Assuming (cid:90) that the Lorenzian peak in (94) is much sharper than the distribution P (λ), we obtain (cid:104)ρ(res)(ε)(cid:105) = dλP (λ)ρ(−) maxπ3εP = ρ(−) (cid:16)−π(cid:112)ε(cid:17) maxπΓhom(λ)δ(ε + (λ/π)2) = for ε < 0. , (114) X. SUMMARY AND DISCUSSION In this paper we have found the shape of the Van Hove singularity manifested in the resistivity of a clean metallic tube of radius R with low concentration n2 of weak short- range impurities (either repulsing or attracting) per unit surface of the tube. We have shown that there is certain . For n2 (cid:29) crossover concentration n(c) n(c) the Van Hove singularities are smoothed peaks at E − EN ∼ ΓB with the width 2 2 = 1 (2πR)2 π (cid:17)2 (cid:16)λ (cid:19)2/3 (cid:18) R2n2λ2 π ΓB ∼ 2 2m∗R2 . (115) The smoothing happens due to interference of scatter- ing events at different impurities, while the amplitude of scattering at each individual impurity is not affected. The structure of the Van Hove singularity for n2 (cid:29) n(c) remains simple: "plateau -- maximum -- plateau" (see Fig. 7). In the most interesting regime at n2 (cid:28) n(c) 2 , the non- Born renormalization of individual scattering amplitudes happens already at E − EN ∼ EnB where the interfer- ence effects are still negligible: 2 (cid:18)λ (cid:19)2 (cid:29) ΓB. π EnB ∼ 2 2m∗R2 (116) Note that the energy scale EnB does not depend on the concentration of impurities. The interference of scatter- ing events at different impurities comes into play only at E − EN ∼ ΓnB, where the individual amplitudes are already strongly renormalized (suppressed) and take uni- versal value (cid:18) 2π2m∗R2E − EN (cid:19)1/2 λ → λ0(E) = (117) 2 which does not depend on the initial "bare" λ. As a result, instead of maximum ρ(E) demonstrates a deep and narrow minimum at E − EN = Edip < 0 with a width ΓnB ∼ Edip ∼ 2 2m∗R2 (118) The structure of the Van Hove singularity for n2 (cid:28) n(c) depends on the sign of the scattering amplitude: for re- pulsive interaction it is "plateau -- minimum -- maximum -- plateau" (see Fig. 2), while for attractive interaction 2 (cid:0)4πR2n2 (cid:1)2 (cid:28) EnB. 14 it is "plateau -- maximum -- minimum -- maximum -- plateau" (see Fig. 3). We should stress, however, that an asym- metric structure of the Van Hove singularity, similar to the Fano resonance shape, arises at low concentration of impurities for both signs of the scattering amplitude. In the leading approximation in small parameter n2/n(c) (that corresponds to independent scattering at 2 different impurities) the resistivity an minimum ρmin van- ishes, as it is shown in Figs. 2 and 3. The value of ρmin becomes nonzero (ρmin ∝ n3 2, see Fig. 8) only if one takes into account the interference of scattering events at different impurities. In the future publications we are going to discuss the structure of Van Hove singularity in a conducting strip. Here the "bare" (non-renormalized) effective scattering amplitudes for different impurities inevitably differ from each other because of the random position of impuritiy with respect to the nodes of the transverse wave func- tion in the resonant band. Since the dependence of the renormalized scattering amplitude on the bare one is non- monotonic, it can be shown that the leading contribution to the resistivity comes not from the "strongest" impuri- ties (those sitting in the antinodes of the wave-function), but from some optimal ones. It leads to a serious mod- ification of the results especially in the range of small E − EN (cid:28) EnB. In Conclusion, our study shows that at low concentra- tion of impurities the non-Born effects lead to splitting of the Van Hove singularities in resistivity of a tube (or, in general, any other quasi-one-dimensional conductor) and this effect can not be described in terms of the Fano formula (1). The character of the splitting depends on whether the impurities are attracting or repulsing. We are indebted to M.V.Feigel'man, L.I.Glazman for illuminating discussions. This work was supported by Basic Research Program of The Higher School of Eco- nomics. Appendix A: Self-consistent calculations: strong Born scattering Strictly speaking, the concept of the self energy is rele- vant only in the weak scattering domain where ε (cid:29) εmin (for both ε > 0 and ε < 0). However, using the pertur- bative expressions (58) and (53) also in the strong scat- tering domain ε (cid:28) εmin can be helpful for qualitative understanding of the behaviour of the density of states and resolving the paradox mentioned in the subsection IV A 2. For a qualitative description of the density of states at strong scattering the self-consistent Born approximation can be used. The selfconsistency equation for Σ reads (cid:32) (cid:33) π(cid:112)ε − Σ(ε) 1 Σ(ε) = − i 2τ0 1 + , (A1) so that ν(ε) ν0 Σ(ε) = − i 2τ0 = 1 + Re − εminY (cid:40) (cid:20) 1 (cid:41) π(cid:112)ε − Σ(ε) (cid:19)(cid:21) (cid:18) 1 ε + i 2τ0 εmin 15 For ε > εbi and ε − εbi (cid:29) Im q this imaginary part can be totally neglected and , , (A2) (A3) ν(ε) ν0 = 1 + √ 1 εmin π Σ(ε) ≈ −εminY (ε/εmin), (cid:112)(ε/εmin) + Y (ε/εmin) 1 (A8) . (A9) where Y (q) is the solution of cubic equation Y 2(Y + q) + 1 = 0. (A4) On the other side of the bifurcation point, for ε < εbi and ε− εbi (cid:29) Im q the Im q-term may be taken into account perturbatively: There is a bifurcation point q = qbi such that for real q < qbi all three solutions of (A4) are real while for q > qbi there is one purely real solution and two con- jugated complex solutions (only the latter ones are phys- ically relevant). Near the point q = qbi one can write Ybi = 21/3, A = 22/33−1/2, √ Y ≈ Ybi ± iA q − qbi qbi = −3 · 2−2/3. (A5) (A6) Thus, if the parameter q were purely real then Im Σ would vanish for ε < εbi ≡ qbi εmin. In our case, however, q has small but finite imaginary part √ Im q = π εmin (cid:28) 1. (A7) and Σ(ε) ≈ −εminY (ε/εmin) − i 2τ0 1 + Y (cid:48) (ε/εmin) [1 + Y (cid:48) (ε/εmin)] , (A10) , (A11) ν(ε) ν0 = 1 + 2 [(ε/εmin) + Y (ε/εmin)]3/2 where Y (cid:48)(q) ≡ dY (q)/dq. In the narrow vicinity of the bifurcation point, for ε− εbi (cid:46) Im q one should keep Im q but, on the other hand, one can use expansion (A6) for Y (q). As a result, in this range we obtain Re Σ(ε) ≈ −εminYbi, (A12) (cid:34)(cid:0)Q2 + 1(cid:1)1/2 Im q (cid:35)1/2 + Q ≈ − √ A Im q 2τ0 (cid:40) Im Σ(ε) ≈ − A 2τ0 A Im q ≈ ν(ε) ν0 √ A Im q π √ 1 εmin (cid:40) Im q 2 (qbi + Ybi)3/2 (2Q)1/2 , Q > 0, 1 (cid:28) Q (cid:28) (Im q)−1, −1/2 , Q < 0, 1 (cid:28) Q (cid:28) (Im q)−1. (−2Q) ≈ √ 1 εmin π 2 (qbi + Ybi)3/2 (2Q)1/2 , Q > 0, 1 (cid:28) Q (cid:28) (Im q)−1, −1/2 , Q < 0, 1 (cid:28) Q (cid:28) (Im q)−1. (−2Q) (cid:34)(cid:0)Q2 + 1(cid:1)1/2 (cid:35)1/2 + Q ≈ (A13) (A14) Here Q(ε) = 2τ0 (ε − εbi) . (A15) So, as it is easy to check, for Q (cid:29) 1 the asymptotics (A13) overlaps with the results (A9) and (B19). The result (A15) should not be taken too seriously: the self-consistency equation (A3) can not be justified rigor- ously. However, the qualitative behaviour of the decay rate and the density of states predicted by (A13) and (B19) gives us a reasonable pattern of matching conflict- ing results (61) and (64). Namely, there is a narrow in- terval ε − εbi (cid:46) 1/2τ0 around certain bifurcation point εbi (εbi < 0, εbi ∼ εmin) where both ν(ε) and τ−1(ε) in- crease with ε very rapidly, and just this increase explains the parametrically large difference between the results (61) and (64). Appendix B: Self-consistent calculations: strong non-Born scattering The general (with an account for the non-Born renor- malization of the scattering amplitude) self-consistency equation for the self-energy Σ reads (cid:12)(cid:12)(cid:12)Λ(ren)(ε − Σ(ε)) (cid:12)(cid:12)(cid:12)2 g(ε − Σ(ε)) πν0 Σ(ε) = n π2 (B1) where Λ(ren) is given by (85) and the density of states is determined by formula (A2). In the case of strong non- Born effect one can use an asymptotic expression (100) and get where we have used the formula Σ(ε) = −inε − Σ 1 + (cid:32) (cid:33) π(cid:112)ε − Σ(ε) 1 , (B2) 1√ a + ib = a2 + b2 + a 2(a2 + b2) − i (cid:115)√ 16 . (B12) (cid:115)√ a2 + b2 − a 2(a2 + b2) Let us first neglect the constant term in g, then we get or Σ(ε) = − in π √ ε − Σ∗, Σ = −ε(nB) q ≡ ε , min Y, ε(nB) min Y 2 + q + Y ∗ = 0. (B3) (B4) (B5) So, the approximate equation (B3) leads to the result 1/τ (ε) ≡ −2Im Σ(ε) = 0 for all ε < εbi. To determine finite scattering rate in this range we should go beyond and take into account the first term −inε − Σ on the right hand side of equation (B2). When doing so we can, however, substitute the found zero-approximation solution to this correction term. Then, instead of (B5), we arrive at For real q the general structure of solutions for equation (B5) is as follows: [Y + in(Y4(q) + q)]2 + q + Y ∗ = 0, (B13) (cid:114) 3 (cid:114) 1 4 For q > 1/4 there are two complex conjugated solu- tions: Y1,2 = 1 2 ∓ i For q < −3/4 there are two real solutions: + q (B6) Y3,4 = − 1 2 ∓ (B7) For −3/4 < q < 1/4 all four solutions Y1,2,3,4 are accept- able. 4 There is, however, always only one physically relevant − q solution: Y (q) = (cid:26) Y4, Y2, where we have noted that in the range q < qbi both q and Y4(q) are real, and also q − Y4(q) ≡ Y 2 4 (q) is real and negative so that we could write ε− Σ = ε(nB) min Y 2 4 (q). Then Σ = −ε(nB) min [inY 2 4 (q) + Y (q)], (B14) where Y (q) is the solution of (B5) with complex for q < qbi, for q > qbi qbi = −3/4. (B8) q ≡ q − inY 2 4 (q). (B15) Thus, the bifurcation energy min qbi, ε(nB) bi = ε(nB) (cid:114) bi < ε (cid:28) εnB we have = −2Im Σ(ε) = 2 ε(nB) min (cid:16) ε − ε(nB) bi (cid:17) and for ε(nB) 1 τ (ε) ν(ε) ν0 = 1 + Re  ,  = (cid:17) (cid:114) π (cid:16) 1 ε + ε(nB) min Y2 = 1 + (cid:113) (cid:16) π ε/ε(nB) min (cid:17) , (B11) ε − ε(nB) ε + ε(nB) min bi (B9) For q < qbi the imaginary part of q can be treated per- turbatively: (B10) Y (q) ≈ Y4(q) + in Y 2 4 (q) 2Y4(q) − 1 , (B16) and Σ(q) ≈ −ε(nB) min Y4(q) (cid:26) 1 + in Y 2 4 (q) Y4(q) − 1/2 (cid:27) (B17) 1 τ (ε) = −2Im Σ(q) = nε(nB) min ((cid:112)1 + 4q − 1)3((cid:112)1 + 4q + 2) 8(q − qbi) , q ≡ ε ε(nB) min < qbi ≡ − 3 4 (cid:113) 1 4 + q + 1 2(q − qbi) ν(ε) ν0 = 1 + 1 2[Y4(q) − 1/2] = 1 + 17 (B18) (B19) , , where we have used Y4(q) − 1/2 = (cid:114) 1 4 + q − 1 = (cid:113) 1 q − qbi 4 + q + 1 and the density of states grow: . (B20) Σ ≈ − ε(nB) min 2 1 + in 2 ε(nB) min bi − ε ε(nB) (cid:40) (cid:41) , ν(ε) ν0 ≈ ε(nB) min bi − ε ε(nB) . (B22) In particular, for q (cid:29) 1 the scattering rate grows with ε while ν(ε) saturates: Σ(
) ≈ − min ε − inε, ε(nB) ν(ε) ≈ ν0 (B21) (cid:113) which is in agreement with (104). When q approaches qbi (i.e., ε → ε(nB) from below), both the scattering rate bi 1 P.Y.Yu and M.Cardona Fundamentals of Semiconductors. Physics and Material Properties. Chap. 6.2, Springer, (2010). 2 A. A. Abrikosov, L. P. Gorkov, and I. Dzyaloshinskii, Quantum Field Theoretical Methods in Statistical Physics, (Pergamon, New York, 1965). 3 Elliott, R.I., Krumhansl, I.A., Leath, P.L.: Rev. Mod. Phys. 45, 465 (1974). 4 Lee, P.A., Ramakrishnan, T.V.: Rev. Mod. Phys. 57, 287 (1985). 5 P. A. Lee, Phys. Rev. Lett. 71, 1887 (1993) 6 M.J. Kearney and P.N. Butcher, J. Phys. C20, 47 (1987). 7 P. Vasilopoulos, F.M. Peeters, Revista Brasileira de Fysica, 19, no 3, (1989) 8 S. Hugle, R. Egger, Phys. Rev. B 66, 193311 (2002) 9 Z. Zhang, D. A. Dikin, R. S. Ruoff, and V. Chandrasekhar, Europhysics Letters, 68, 713 (2004). 10 B. Babi´c and C. Schonenberger, Phys. Rev. B 70, 195408 (2004) 11 J. Kim, J. R. Kim, Jeong-O Lee, J. W. Park, H. M. So, N. Kim, K. Kang, K. H. Yoo, and J. J. Kim, Phys. Rev. Lett. 90, 166403 (2003). 12 W. Yi, L. Lu, H. Hu, Z. W. Pan, and S. S. Xie, Phys. Rev. Lett. 91, 076801 (2003). 13 U. Fano, Phys. Rev. 124, 1866 (1961). 14 A. E. Miroshnichenko, S. Flach, Y.S.Kivshar, Rev. Mod. Phys. 82, 2257 (2010) 15 N. B. Brandt et al, Zh. Eksp. Teor. Fiz 72, 2332 (1977) [Sov. Phys. JETP, 45, 1226 (1977)]; So, the scattering rate reaches its minimum at some ε = dip ≡ ε(nB) ε(nB) min qdip, where qdip = − 21 16 , 1 τ (εdip) = 27 8 nε(nB) min (B23) 16 N. B. Brandt et al, Fiz. Nizk. Temp. 8, 718 (1982) [Sov. J. Low Temp. Phys., 8, 358 (1982)] 17 A. Nikolaeva et al, Phys. Rev. B 77, 075332 (2008) 18 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959) 19 B. L. Altshuler, A. G. Aronov, and B. Z. Spivak, Pis'ma Zh. Eksp. Teor. Fiz., 33, 101 (1981) [JETP Letters, 33, 94 (1981) 20 B. L. Altshuler, A. G. Aronov, B. Z. Spivak, D. Yu. Sharvin, and Yu. V. Sharvin, Pis'ma Zh. Eksp. Teor. Fiz., 35, 476 (1982) [JETP Letters, 35, 588 (1982)] 21 A. G. Aronov and Yu. V. Sharvin, Reviews of Modern Physics, 59, 755 (1987) 22 A. S. Ioselevich, JETP Letters, 101, 358 (2015) 23 The quadratic spectrum is taken only for simplicity. Since the effects of interest are dominated by narrow vicinity of the Fermi level, the obtained results can be easily reformu- lated for arbitrary spectrum. 24 Landau and E.M. Lifshitz, Course in Theoretical Physics (Pergamon, Oxford, 1981), Vol. 3 (Quantum mechanics. Nonrelativistic theory.) 25 A. L. Fetter, Phys. Rev. 140, A1921 -- A1936 (1965). 26 K. Machida and F. Shibata, Prog. Theor. Phys. 47, 1817 (1972). 27 H. Shiba, Prog. Theor. Phys. 50, 50 (1973). 28 T. Soda, T. Matsuura, and Y. Nagaoka, Progress of The- oretical Physics 38, 551 (1967). 29 H. Shiba, Progress of Theoretical Physics 40, 435 (1968). 

1406.7771

2

1406

Raman spectroscopy as probe of nanometer-scale strain variations in graphene

Confocal Raman spectroscopy is a versatile, non-invasive investigation tool and a major workhorse for graphene characterization. Here we show that the experimentally observed Raman 2D line width is a measure of nanometer-scale strain variations in graphene. By investigating the relation between the G and 2D line at high magnetic fields we find that the 2D line width contains valuable information on nanometer-scale flatness and lattice deformations of graphene, making it a good quantity for classifying the structural quality of graphene even at zero magnetic field.

cond-mat.mes-hall

cond-mat

Raman spectroscopy as probe of nanometer-scale strain variations in graphene C. Neumann1,2, S. Reichardt1, P. Venezuela3, M. Drogeler1, L. Banszerus1, M. Schmitz1, K. Watanabe4, T. Taniguchi4, F. Mauri5, B. Beschoten1, S. V. Rotkin1,6, and C. Stampfer1,2 1 JARA-FIT and 2nd Institute of Physics, RWTH Aachen University, 52074 Aachen, Germany 2 Peter Grunberg Institute (PGI-9), Forschungszentrum Julich, 52425 Julich, Germany 3 Instituto de F´ısica, Universidade Federal Fluminense, 24210-346 Niter´oi, RJ, Brazil 4 National Institute for Materials Science,1-1 Namiki, Tsukuba, 305-0044, Japan 5 IMPMC, UMR CNRS 7590, Sorbonne Universit´es UPMC Univ. Paris 06, 6 Department of Physics and Center for Advanced Materials and Nanotechnology, MNHN, IRD, 4 Place Jussieu, 75005 Paris, France Lehigh University, Bethlehem, Pennsylvania 18015, USA (Dated: October 6, 2018) Confocal Raman spectroscopy is a versatile, non-invasive investigation tool and a major workhorse for graphene characterization. Here we show that the experimentally observed Raman 2D line width is a measure of nanometer-scale strain variations in graphene. By investigating the relation between the G and 2D line at high magnetic fields we find that the 2D line width contains valuable information on nanometer-scale flatness and lattice deformations of graphene, making it a good quantity for classifying the structural quality of graphene even at zero magnetic field. Graphene combines several highly interesting material properties in a unique way, promising unprecedented ma- terial functionalities. This makes graphene increasingly attractive for industrial applications1 but, at the same time, stresses the need for non-invasive characterization techniques. In recent years, Raman spectroscopy has proven to be highly useful as a non-invasive method not only to identify graphene2,3, but also to extract infor- mation on local doping4 -- 7, strain8,9 and lattice tempera- ture10,11. Even more insights can be gained when utiliz- ing confocal, scanning Raman spectroscopy to study spa- tially resolved doping domains7,12, edge effects3,13 and position dependent mechanical lattice deformations, in- cluding strain14 -- 16. The spatial resolution of so-called Raman maps is on the order of the laser spot size (which for confocal systems is typically on the order of 500 nm) and the extracted quantities (such as doping or strain) are in general averaged over the spot size. It is there- fore important to distinguish between length scales sig- nificantly larger or smaller than the laser spot size. In particular, we will distinguish between strain variations on a micrometer scale, which can be extracted from spa- tially resolved Raman maps, and nanometer-scale strain variations, which are on sub-spot-size length scales and cannot be directly observed by Raman imaging, but are considered as important sources of scattering for elec- tronic transport17. Here we show that the experimentally observed Ra- man 2D line width is a measure of nanometer-scale strain variations in graphene on insulating substrates, i.e. it contains valuable information on local (i.e. nanometer- scale) flatness, lattice deformations and crystal quality of graphene. To prove that the the experimentally observed 2D line width depends on sub-spot size strain variations and lattice deformations we employ the following strat- egy: (i) We start by showing that by combining Raman spectroscopy with magnetic fields, electronic broadening contributions for the Raman G line width can be strongly suppressed. Since in perpendicular magnetic fields the electronic states in graphene condense into Landau levels (LLs), the interaction between electronic excitations and lattice vibrations becomes B field dependent. In agree- ment with existing theory18 -- 21 and experiments22,23, we demonstrate that by applying a perpendicular B field of around 8 T, the G line does as good as not depend on electronic properties such as charge carrier doping, screening or electronic broadening. (ii) We observe that, under these conditions, the G line width nevertheless exhibits strong variations across graphene flakes. In particular, we show that the G line width is significantly increased in regions where the graphene flake features bubbles and folds, i.e. in cor- respondence with increased structural deformations. (iii) Finally, we show that at 8 T there is a (nearly) lin- ear dependence between the G line width and the 2D line width, implying that there is a common source of line broadening. According to points (i) and (ii) the broad- ening must be related to structural lattice deformations. This finding is further supported by a detailed analysis of the relation between the area of the 2D peak and its line width. By analyzing the relation between the G and 2D line width, we find that nm-scale strain variations constitute a dominant contribution to the observed line broadenings. Importantly, the 2D line has been shown to be only very weakly dependent on the B field24, meaning that no magnetic field is required to extract information on nm-scale strain variations from the 2D line width, which makes this quantity interesting for practical appli- cations. For the low temperature Raman measurements, we em- ploy a commercially available confocal Raman setup that allows us to perform spatially-resolved experiments at a temperature of 4.2 K and magnetic fields of up to 9 T. We use an excitation laser wavelength of 532 nm with a spot diameter on the sample of around 500 nm. For detection, 2 FIG. 1. (color online) (a) Schematic cross section of the investigated sample highlighting the different regions I and II. (b) Optical image of a Gr-hBN heterostructure resting partly on hBN and SiO2. (c) and (d) Raman spectrum taken on the SiO2-Gr-hBN (c) and hBN-Gr-hBN (d) areas. The positions where the spectra were taken are marked by a blue and a red star, respectively, in panel (b). (e) Raman map of the intensity of the hBN peak. The dashed lines mark the regions I and II. (f) ΓGversus ωG recorded on various spots on regions I (blue) and II (red) of the sample. (g) Γ2D versus ω2D recorded on various spots on regions I (blue) and II (red) of the sample. (h) Histograms of Γ2D recorded on various spots on regions I (blue) and II (red) of the sample. (i) ω2D versus ωG recorded on various spots on regions I (blue) and II (red) of the sample. we use a single mode optical fiber and a CCD spectrom- eter with a grating of 1200 lines/mm. All measurements are performed with linear laser polarization and a ×100 objective. The investigated graphene (Gr) sheet is partly encap- sulated in hexagonal boron nitride (hBN) and partly sandwiched between SiO2 and hBN as illustrated in Fig- ure 1a. An optical image of our sample is shown in Fig- ure 1b. In contrast to graphene encapsulated in hBN, graphene flakes supported by SiO2 usually feature lower carrier mobilities of around 103-104 cm2/(Vs), indicating a detrimental influence of SiO2 on the electronic prop- erties of graphene. In this regard, our structure gives us the invaluable capability of probing a single graphene sheet exposed to two different substrates (region I and II in Figures 1a and 1b). The sample is fabricated with a dry and resist-free transfer process following refs. 25 and 26, where we pick up an exfoliated graphene flake with an hBN flake and deposit it onto the hBN-SiO2 transi- tion area of the substrate. A typical Raman spectrum of graphene supported by SiO2 and covered by hBN, taken at the position of the red star in Figure 1b, is shown in Figure 1c. The characteristic hBN line as well as the graphene G and 2D lines can be clearly identified. At first glance, the spectra recorded in the hBN-Gr-hBN area look similar (see Figure 1d, taken at the position marked by the blue star in Figure 1b). However, it is ev- ident that the ratio between the 2D and G line intensity is higher in this case. Furthermore, the full width at half maximum (FWHM) of the 2D line, Γ2D, is significantly smaller. The confocal nature of our Raman setup enables us to do spatially resolved measurements. An example of a Raman map is shown in Figure 1e, where the spatially resolved intensity of the hBN line is depicted. The hBN and SiO2 areas can be clearly distinguished in the map (see highlighted regions I and II). When analyzing the Raman spectra of every point on the map, one finds that the G lines recorded in the hBN-encapsulated area are broader than in the SiO2 supported area (compare red and blue data points in Figure 1f). This is a clear indica- tion of reduced charge carrier doping induced by the hBN substrate compared to SiO2. In fact, at low charge carrier doping, the phonon mode can decay into electron-hole pairs, which results in a broadening of the G peak5,27. For the 2D line, in contrast, the Γ2D recorded in the hBN-encapsulated area is mostly between 16 cm−1 and 20 cm−1, while it is above 22 cm−1 in the SiO2 area (see blue and red curves in the histogram of Figure 1h, re- spectively). Note that both Γ2D and ΓG do not show a dependence on the respective frequencies ω2D and ωG (Figures 1f and 1g). In Figure 1i the position of the G and 2D lines for every spectrum obtained on the investi- gated graphene sheet are displayed. For both substrates, the data points scatter along a line with a slope of 2.2. This slope coincides with the ratio of strain induced shifts (i.e. of the related Gruneisen parameters) of the Raman G and 2D modes28. This indicates that there are sig- 2670268026902700271027201580159016002.2hBN (I)SiO (II)210 µm1200160024002800-1Raman shift (cm)(c)(d)152535203010counts (a.u.)(g)(h)(i)hBN intensity (a.u.)26802690270027100102030405051015202501580158515901595-1Γ (cm)G-1Γ (cm)2D(f)graphene-1Γ (cm)2D-1ω (cm)2D-1ω (cm)G-1ω (cm)2DSiO2hBNIIII(e)(b)IIIISiO-Gr-hBN2hBN-Gr-hBNIIIIIntensity (a.u.)d(a)SiO-Gr-hBN222.7Intensity (a.u.)1200160024002800IIIIII-1ω (cm)GhBN-Gr-hBN17.2hBN G 2D 3 matches the G mode phonon, the position of the G line is shifted and its line width increases. An example for the evolution of the Raman G peak with magnetic field, taken on the hBN sandwich area, is shown in Figure 2a. The individual spectra are offset for clarity. For a detailed analysis, single Lorentzians are fitted to every spectrum. The resulting frequency, ωG, and FWHM, ΓG, are dis- played in Figures 2b and 2c, respectively. The arrow at B = 3.7 T (Figure 2c) indicates a value of the magnetic field where a LL transition is energetically matched with the phonon, leading to a broadening of the G line. How- ever, at a magnetic field of about 8 T, no LL transition is close to the G mode, as illustrated in Figure 2d, where the energies of the relevant LL transitions as a function of magnetic field are compared to the energy of the G mode phonon. Consequently at this high magnetic field the influence of the electronic system on the position and width of the G line is minimized. Note that this effect is independent of the charge carrier density and the exact values of the broadening of the LL transitions assuming that the latter are within a reasonable range as found by other studies23,34. Thus, the residual broadening of the G line is most likely determined by phonon-phonon scat- tering and averaging effects over different strain values that vary on a nanometer scale (i.e. sub-spot size length scale see also Supplementary Information). To show that this applies to the entire sample, we first show that the broadening of the electronic states is low enough on the entire hBN-Gr-hBN area. In Figures 3a and 3b, we show maps of ΓG at B = 0 T and 3.8 T, respectively. On the hBN part, the width of the G line shows the resonant behavior depicted in Figure 2c (see also histogram in Figure 3d). This effect happens on all spots on the hBN area, independent of the local dop- ing and strain values and independent of possible local folds and bubbles. The suppression of magneto-phonon resonances on the SiO2 substrate can be attributed to the higher charge carrier density. At higher charge car- rier density the needed LL transitions are blocked by the Pauli principle. In a next step, we tune the magnetic field to 8 T, where the electronic influences on the Ra- man G line are at a minimum. A map of ΓG over the entire flake at a magnetic field of 8 T is shown in Fig- ure 3c. Distinct features across the whole sample are visible as regions with increased line width. A compari- son with a scanning force microscope image of the sample (Figure 3e) reveals that many of these regions can be as- sociated with folds and bubbles most likely induced dur- ing the fabrication process, some of which even cross the border between the underlying hBN and SiO2 substrate regions. As electronic broadening effects are suppressed at 8 T, the increased line width of the G line in the vicinity of these lattice deformations arises from enhanced phonon- phonon scattering and/or an averaging effect over vary- ing nm-scale strain conditions. Interestingly, the same features can also be identified in a Γ2D map recorded at B = 0 T, shown in Figure 3f. This FIG. 2. (color online) (a) Raman spectra recorded as a func- tion of magnetic field, ranging from 0 T (bottom spectrum) to 8.9 T (top spectrum). The spectra are vertically offset for clarity. (b) and (c) Frequency, ωG, and FWHM, ΓG, of the G peak as a function of magnetic field as obtained from Lorentzian fits to the data shown in panel (a). The arrow in panel (c) showcases a value of the magnetic field at which the phonon is energetically matched to a LL transition. (d) Evo- lution of the energies of LL transitions with magnetic field. The full lines represent inter-band transitions in which the LL index changes by one. The dashed lines represent inter-band transitions in which the LL index does not change. The red line represents the G mode phonon frequency at zero B field. The circled region in (c) and (d) highlights the region in which no LL transitions energetically match the G mode phonon. nificant strain variations on both substrates across the entire graphene layer. Assuming the strain to be of biax- ial nature, the spread of the data points translates into a maximum, micrometer-scale strain variation of about 0.14%28. The offset of the SiO2 and hBN data points can be understood in terms of the higher charge carrier dop- ing induced by the SiO2 substrate, which shifts the data 5, and differences in points toward higher values of ωG the dielectric screening of hBN and SiO2 that effectively shift the 2D line position29. Since the data stems from a single graphene flake that has undergone identical fab- rication steps for both substrate regions, the difference in charge carrier doping is unambiguously due the two different substrate materials. For a more refined comparison of the Raman spectra on both substrates, we seek to suppress the effects on the G line, arising from these differences in charge carrier dop- ing. We therefore minimize the influence of the electronic system on the Raman G line by applying a perpendicular magnetic field. In the presence of a perpendicular mag- netic field, the electronic states in graphene condense into Landau levels (LLs). The coupling of these Landau levels to the G mode is well understood18,19 and experimentally confirmed21 -- 23,30 -- 35. When a LL transition energetically 102030158015851590-1ω (cm)G-1Γ (cm)GIntensity (a. u.)15801600(a)(b)(c)-1Raman shift (cm)B (T)100200300Energy (meV)0 024680(d)ωph246804 FIG. 3. (color online) (a), (b), and (c) Raman maps of the FWHM of the G peak, ΓG, taken at different magnetic fields, i.e. B = 0 T (a), 3.8 T (b), and 8 T (c), respectively. The different regions I and II (labeled in Figure 1a) can be well distinguished in all three panels. (d) Histograms of ΓG for the different magnetic fields, B = 0 T (blue), 3.8 T (red), B = 8 T (gray) and the two substrate substrates hBN (top panel) and SiO2 (bottom panel). (e) Scanning force microscope (SFM) image of the investigated sample. (f) Raman map of Γ2D recorded at 0 T. The arrows highlight mechanical folds visible in the SFM image as well as in the Raman maps (see panels (c), (e), and (f)). strongly suggests that the lattice deformations identified at 8 T in ΓG also cause a broadening of the 2D mode. The same trend is highlighted in Fig. 4a, where we show the relation of ΓG and Γ2D for all recorded Raman spec- tra at 8 T. The additional teal data points stem from a graphene-on-SiO2 sample and the orange star originates from a different hBN-Gr-hBN sandwich structure with all data having been obtained at 8 T. Notably, the points from all substrate regions lie on one common line. From this linear relation between Γ2D and ΓG (Figure 4a), we conclude that there must be a common source of line broadening, which is connected to structural deforma- tions. This is mainly due to the fact that at 8 T the G-line broadening is only very weakly affected by electronic con- tributions (see above). The range of the presented scat- ter plot can be extended by including data recorded on low-quality graphene samples with significant doping, as shown in Figure 4b. Here, no magnetic field but high dop- ing (corresponding to Fermi energies much higher than half of the phonon energy ¯hωph/2 ≈ 100 meV) is used to suppress Landau damping of the G mode, leaving ΓG unaffected from electronic contributions. The col- ored data points are from Raman maps (B = 0 T) of CVD (chemical vapor deposition)-grown graphene flakes that were transferred onto SiO2 by a wet chemistry-based transfer. These graphene sheets contain doping values of nel > 3 × 1012 cm−2, which corresponds to Fermi ener- gies, EF > 200 meV (see suppl. material). The data points show the same trend as the values obtained at 8 T (gray data points in Figure 4b) and even extend the total range of the dependence to higher values of Γ2D. While the linear relation between ΓG and Γ2D in Fig- ure 4a and 4b shows that structural deformations also broaden the 2D line, it is less straightforward to identify the actual mechanism of broadening. It is, in principle, possible that the high values of Γ2D around folds and bubbles are due to a combination of increased phonon- phonon scattering, averaging effects over different strain values within the laser spot and reduced electronic life times. However, interestingly the slopes in Figures 4a and 4b are around 2.2 (see black lines). This is a remark- able resemblance to the strain induced frequency shifts of both modes (compare Figure 1i). This provides very strong indication that averaging over different strain val- ues, which vary on a nanometer scale (see Fig. 4c), play an important role in the broadening of the experimentally observed 2D line. This averaging effect broadens the G and 2D line by the same ratio as their peak positions shift for fixed average strain values explaining the slope of 2.2 between ΓG and Γ2D (see Supplementary Information). We are aware that the low charge carrier densities in the hBN encapsulated area might result in a narrowing of the 2D mode by three to four wave numbers36. However, the large differences of Γ2D on the order of 20-30 cm−1 on both substrates cannot be explained by the differences in charge carrier doping7,36,37. hBN0T1020305152535 8T3.8Tcounts (a.u.)SiO20355B=0T1015202530-1Γ (cm)G-1Γ (cm)G(b)301020Γ (cm )2D-15 μmhBNHeight (nm)0206040B=3.8TB=8T(a)(c)(e)(f)(d)SiO25 FIG. 4. (color online) (a) ΓG versus Γ2D recorded on various points on the hBN part (blue) and SiO2 (red) of the sample at a magnetic field of 8 T. Additional data points from a graphene-on-SiO2 sample (teal) and a second hBN-Gr-hBN sample (orange star) are shown. (b) The data points of panel (a) are depicted in gray. The colored data are recorded on four different CVD graphene flakes on SiO2 substrate at 0 T. All four samples have doping values of nel > 3 × 1012, such that Landau damping of the G line is suppressed. The dashed and dotted lines in panels (a) and (b) indicate the calculated values of Γ2D from DFT calculations including electron-phonon and phonon-phonon broadening (dotted line) and electron-phonon, electron-eletron and phonon-phonon broadening (dashed line). large variations, bottom: small variations). (d) Γ2D versus the integrated area of the 2D peak as obtained from single Lorentzian fits for the hBN part (blue) and SiO2 (red) measured at 8 T. Both data clouds are scaled to an average area2D value of one. (e) Similar plot as in panel (d) but for 0 T. The solid black line is the calculated dependence of Γ2D and area2D for varying electronic broadening from the DFT calculations, specified in the text and in ref. 37. The dashed and dotted black lines are the same as in panel (a) and (b). (c) Two schematic illustrations of nanometer-scale strain variations (top: Interestingly, the lowest Γ2D observed in our experi- ments are very close to the value that we compute from first-principles as in ref. 37 (see Supplementary Informa- tion for details) assuming an undoped, defect-free and stress-free sample of graphene (horizontal dashed and dotted lines in Figures 4a and 4b). In such an approach, the width of the 2D peak is determined by the anhar- monic decay rate of the two phonons involved (5.3 cm-1 according to ref. 38), and, indirectly, by the broadening of the electron and hole, denoted as γ in ref. 37, (see also ref. 39). According to ref. 37, the electron-phonon contribution to γ is 81.9 meV for electronic states in res- onance with the 2.33 eV laser-light. With such a value of γ we obtain a Γ2D of 12.1 cm1 (dotted lines in Figures 4a, 4b and 4e). If, following ref. 40, we double such value of γ to account for the electron-electron scattering, we obtain a Γ2D of 17.9 cm−1 (dashed lines in Figures 4a, 4b and 4e), in close agreement with the lowest measured values. In principle, the observed increase of Γ2D with respect to its minimum value could be attributed to an increase of the electronic broadening γ, due to doping (increas- ing the electron-electron scattering) or to the presence of defects (increasing the electron-defect scattering)37,39,41. By investigating the relation between Γ2D and the in- tegrated area of the 2D peak (area2D) we can exclude such a hypothesis. In Figures 4d and 4e we show scat- ter plots of Γ2D versus the region-normalized area2D for both B = 8 T and 0 T, highlighting the very weak B- field dependence of Γ2D. More importantly, we observe that the area of the 2D peak does not depend on Γ2D, contrary to what is expected in presence of a variation of the electronic broadening γ37,39,41. In particular the measured data does not follow the calculated dependence of Γ2D on area2D, reported in Fig. 4e, obtained in the calculation by varying electronic broadening γ. This dis- misses differences in the electronic broadening as a main mechanism for the observed variations of Γ2D. Finally, our finding that the 2D line depends on nanometer-scale strain inhomogeneities is also in good agreement with high resolution scanning tunneling mi- croscopy measurements, which reveal that graphene on SiO2 forms short-ranged corrugations, while graphene on hBN features significantly more flat areas42. In summary, we showed that by using a magnetic field of 8 T to strongly suppress the influence of the electronic contributions on the Raman G line width, the latter can be used as a measure for the amount of nm-scale strain variations. Most importantly, we observed a nearly linear dependence between the G and 2D line widths at 8 T independent of the substrate material, indicating that the dominating source of the spread of the broadening of both peaks is the same. From the slope ∆Γ2D/∆ΓG of around 2.2, we deduce that averaging effects over nanometer- scale strain variations make a major contribution to this B=8T203040505101520510152020304050(a)(b)-1Γ (cm)G-1Γ (cm)G-1Γ (cm)2D-1Γ (cm)2D00(c)2.2152025303540-1Γ (cm)2D-1Γ (cm)2D(d)(e)8T0T101000.511.52area (a.u.)2D00.511.52area (a.u.)2D1015202530354010trend. Since the 2D line width shows only a very weak dependence on the B field, this quantity can even be used without a magnetic field to gain information on the local strain homogeneity and thus on the structural quality of graphene. These insights can be potentially very valuable for monitoring graphene fabrication and growth processes in research and industrial applications, where a fast and non-invasive control of graphene lattice deformations is 6 of great interest. ACKNOWLEDGMENT We thank T. Khodkov for support during the measure- ments. Support by the Helmholtz Nanoelectronic Facility (HNF), the DFG, the ERC (GA-Nr. 280140) and the EU project Graphene Flagship (contract no. NECT-ICT- 604391), are gratefully acknowledged. P.V. acknowledges financial support from the Capes-Cofecub agreement. 1 K. S. Novoselov, V. Fal'ko, L. Colombo, P. Gellert, M. Schwab, and K. Kim, Nature 490, 192 (2012). 2 A. Ferrari, J. Meyer, V. Scardaci, C. Casiraghi, M. Lazzeri, F. Mauri, S. Piscanec, D. Jiang, K. Novoselov, S. Roth, and A. Geim, Phys. Rev. Lett. 97, 187401 (2006). 3 D. Graf, F. Molitor, K. Ensslin, C. Stampfer, A. Jungen, C. Hierold, and L. Wirtz, Nano Lett. 7, 238 (2007). 4 A. C. Ferrari, Solid State Commun. 143, 47 (2007). 5 J. Yan, Y. Zhang, P. Kim, and A. Pinczuk, Phys. Rev. Lett. 98, 166802 (2007). 6 S. Pisana, M. Lazzeri, C. Casiraghi, K. S. Novoselov, A. K. Geim, A. C. Ferrari, and F. Mauri, Nat. Mater. 6, 198 (2007). 7 C. Stampfer, F. Molitor, D. Graf, K. Ensslin, A. Jungen, C. Hierold, and L. Wirtz, Appl. Phys. Lett. 91, 241907 (2007). 8 M. Mohr, J. Maultzsch, and C. Thomsen, Phys. Rev. B 82, 201409 (2010). 9 M. Huang, H. Yan, T. F. Heinz, and J. Hone, Nano Lett. 10, 4074 (2010). 21 C. Qiu, X. Shen, B. Cao, C. Cong, R. Saito, J. Yu, M. S. Dresselhaus, and T. Yu, Phys. Rev. B 88, 165407 (2013). 22 J. Yan, S. Goler, T. D. Rhone, M. Han, R. He, P. Kim, and A. Pinczuk, Phys. Rev. Lett. 105, V. Pellegrini, 227401 (2010). 23 C. Neumann, S. Reichardt, M. Drogeler, B. Terr´es, K. Watanabe, T. Taniguchi, B. Beschoten, S. V. Rotkin, and C. Stampfer, Nano Lett. 15, 1547 (2015). 24 C. Faugeras, P. Kossacki, D. Basko, M. Amado, M. Sprin- kle, C. Berger, W. A. de Heer, and M. Potemski, Phys. Rev. B 81, 155436 (2010). 25 L. Wang, I. Meric, P. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi, K. Watanabe, L. Campos, D. Muller, J. Guo, P. Kim, J. Hone, K. Shepard, and C. Dean, Science 342, 614 (2013). 26 S. Engels, B. Terr´es, F. Klein, S. Reichardt, M. Goldsche, S. Kuhlen, K. Watanabe, T. Taniguchi, and C. Stampfer, physica status solidi (b) 251, 2545 (2014). 27 C. Casiraghi, S. Pisana, K. Novoselov, A. Geim, and A. Ferrari, Appl. Phys. Lett. 91, 233108 (2007). 10 I. Calizo, A. Balandin, W. Bao, F. Miao, and C. Lau, 28 J. E. Lee, G. Ahn, J. Shim, Y. S. Lee, and S. Ryu, Nat. Nano Lett. 7, 2645 (2007). 11 A. A. Balandin, S. Ghosh, W. Bao, I. Calizo, D. Tewelde- brhan, F. Miao, and C. N. Lau, Nano Lett. 8, 902 (2008). 12 M. Drogeler, F. Volmer, M. Wolter, B. Terr´es, K. Watan- and abe, T. Taniguchi, G. Guntherodt, C. Stampfer, B. Beschoten, Nano Lett. 14, 6050 (2014). 13 C. Casiraghi, A. Hartschuh, H. Qian, S. Piscanec, C. Georgi, A. Fasoli, K. Novoselov, D. Basko, and A. Fer- rari, Nano Letters 9, 1433 (2009). 14 T. Mohiuddin, A. Lombardo, R. Nair, A. Bonetti, G. Savini, R. Jalil, N. Bonini, D. Basko, C. Galiotis, N. Marzari, et al., Phys. Rev. B 79, 205433 (2009). 15 J. Zabel, R. R. Nair, A. Ott, T. Georgiou, A. K. Geim, K. S. Novoselov, and C. Casiraghi, Nano Lett. 12, 617 (2012). 16 D. Yoon, Y.-W. Son, and H. Cheong, Phys. Rev. Lett. 106, 155502 (2011). 17 N. J. Couto, D. Costanzo, S. Engels, D.-K. Ki, K. Watan- abe, T. Taniguchi, C. Stampfer, F. Guinea, and A. F. Morpurgo, Phys. Rev. X 4, 041019 (2014). 18 T. Ando, J. Phys. Soc. Jpn. 76, 024712 (2007). 19 M. Goerbig, J.-N. Fuchs, K. Kechedzhi, and V. I. Fal'ko, Phys. Rev. Lett. 99, 087402 (2007). 20 O. Kashuba and V. I. Fal'ko, New J. Phys. 14, 105016 (2012). Commun. 3, 1024 (2012). 29 F. Forster, A. Molina-Sanchez, S. Engels, A. Epping, K. Watanabe, T. Taniguchi, L. Wirtz, and C. Stampfer, Phys. Rev. B 88, 085419 (2013). 30 C. Faugeras, M. Amado, P. Kossacki, M. Orlita, M. Kuhne, A. A. Nicolet, Y. I. Latyshev, and M. Potemski, Phys. Rev. Lett. 107, 036807 (2011). 31 C. Faugeras, P. Kossacki, A. Nicolet, M. Orlita, M. Potem- ski, A. Mahmood, and D. Basko, New J. Phys. 14, 095007 (2012). 32 C. Faugeras, M. Amado, P. Kossacki, M. Orlita, M. Sprin- kle, C. Berger, W. A. De Heer, and M. Potemski, Phys. Rev. Lett. 103, 186803 (2009). 33 P. Kossacki, C. Faugeras, M. Kuhne, M. Orlita, A. Mah- mood, E. Dujardin, R. Nair, A. Geim, and M. Potemski, Phys. Rev. B 86, 205431 (2012). 34 Y. Kim, J. Poumirol, A. Lombardo, N. Kalugin, T. Geor- giou, Y. Kim, K. Novoselov, A. Ferrari, J. Kono, O. Kashuba, V. Fal'ko, and D. Smirnov, Phys. Rev. Lett. 110, 227402 (2013). 35 P. Leszczynski, Z. Han, A. A. Nicolet, B. A. Piot, P. Kos- sacki, M. Orlita, V. Bouchiat, D. M. Basko, M. Potemski, and C. Faugeras, Nano Lett. 14, 1460 (2014). 36 S. Berciaud, X. Li, H. Htoon, L. E. Brus, S. K. Doorn, and T. F. Heinz, Nano Lett. 13, 3517 (2013). 37 P. Venezuela, M. Lazzeri, and F. Mauri, Phys. Rev. B 84, 035433 (2011). 38 L. Paulatto, F. Mauri, and M. Lazzeri, Phys. Rev. B 87, 214303 (2013). 40 F. Herziger, M. Calandra, P. Gava, P. May, M. Lazzeri, F. Mauri, and J. Maultzsch, Phys. Rev. Lett. 113, 187401 (2014). 41 D. M. Basko, S. Piscanec, and A. Ferrari, Phys. Rev. B 39 D. M. Basko, Phys. Rev. B 78, 125418 (2008). 80, 165413 (2009). 42 C.-P. Lu, G. Li, K. Watanabe, T. Taniguchi, and E. Y. Andrei, Phys. Rev. Lett. 113, 156804 (2014). 7 

1110.5702

1

1110

Quasiparticle bandgap engineering of graphene and graphone on hexagonal boron nitride substrate

Graphene holds great promise for post-silicon electronics, however, it faces two main challenges: opening up a bandgap and finding a suitable substrate material. In principle, graphene on hexagonal boron nitride (hBN) substrate provides potential system to overcome these challenges. Recent theoretical and experimental studies have provided conflicting results: while theoretical studies suggested a possibility of a finite bandgap of graphene on hBN, recent experimental studies find no bandgap. Using the first-principles density functional method and the many-body perturbation theory, we have studied graphene on hBN substrate. A Bernal stacked graphene on hBN has a bandgap on the order of 0.1 eV, which disappears when graphene is misaligned with respect to hBN. The latter is the likely scenario in realistic devices. In contrast, if graphene supported on hBN is hydrogenated, the resulting system (graphone) exhibits bandgaps larger than 2.5 eV. While the bandgap opening in graphene/hBN is due to symmetry breaking and is vulnerable to slight perturbation such as misalignment, the graphone bandgap is due to chemical functionalization and is robust in the presence of misalignment. The bandgap of graphone reduces by about 1 eV when it is supported on hBN due to the polarization effects at the graphone/hBN interface. The band offsets at graphone/hBN interface indicate that hBN can be used not only as a substrate but also as a dielectric in the field effect devices employing graphone as a channel material. Our study could open up new way of bandgap engineering in graphene based nanostructures.

cond-mat.mes-hall

cond-mat

Quasiparticle bandgap engineering of graphene and graphone on hexagonal boron nitride substrate Neerav Kharche∗,†,¶ and Saroj K. Nayak∗,‡ Computational Center for Nanotechnology Innovations, Rensselaer Polytechnic Institute, Troy, NY 12180, USA, and Department of Physics, Applied Physics and Astronomy, Rensselaer Polytechnic Institute, Troy, NY 12180, USA E-mail: [email protected]; [email protected] Abstract Graphene holds great promise for post-silicon electronics, however, it faces two main challenges: opening up a bandgap and finding a suitable sub- strate material. In principle, graphene on hexago- nal boron nitride (hBN) substrate provides poten- tial system to overcome these challenges. Recent theoretical and experimental studies have pro- vided conflicting results: while theoretical stud- ies suggested a possibility of a finite bandgap of graphene on hBN, recent experimental studies find no bandgap. Using the first-principles density functional method and the many-body perturba- tion theory, we have studied graphene on hBN substrate. A Bernal stacked graphene on hBN has a bandgap on the order of 0.1 eV, which disap- pears when graphene is misaligned with respect to hBN. The latter is the likely scenario in real- In contrast, if graphene supported istic devices. on hBN is hydrogenated, the resulting system (graphone) exhibits bandgaps larger than 2.5 eV. While the bandgap opening in graphene/hBN is due to symmetry breaking and is vulnerable to slight perturbation such as misalignment, the gra- ∗To whom correspondence should be addressed †Computational Center for Nanotechnology Innovations, Rensselaer Polytechnic Institute, Troy, NY 12180, USA ‡Department of Physics, Applied Physics and Astron- omy, Rensselaer Polytechnic Institute, Troy, NY 12180, USA ¶Department of Physics, Applied Physics and Astron- omy, Rensselaer Polytechnic Institute, Troy, NY 12180, USA phone bandgap is due to chemical functionaliza- tion and is robust in the presence of misalignment. The bandgap of graphone reduces by about 1 eV when it is supported on hBN due to the polariza- tion effects at the graphone/hBN interface. The band offsets at graphone/hBN interface indicate that hBN can be used not only as a substrate but also as a dielectric in the field effect devices em- ploying graphone as a channel material. Our study could open up new way of bandgap engineering in graphene based nanostructures. KEYWORDS: graphene, hexagonal boron nitride, GW, polarization, non- local screening, bandgap renormalization Functionalized Graphene exhibits remarkable electronic proper- ties compared to the conventional materials such as Si and III-Vs making it an attractive material for next generation electronic devices.1 For practi- cle applications, devices made from an atomically thin material such as graphene should be supported on a substrate. Typically graphene devices are fab- ricated on SiO2 substrate, however, carrier mobil- ity in graphene on SiO2 reduces due to charged surface states, surface roughness, and surface opti- cal phonons in SiO2.2,3 Several other oxide-based substrates have been investigated so far, how- ever, none yields significant improvement over SiO2.3 Recently, graphene supported on a hexago- nal boron nitride (hBN) substrate was found to ex- hibit much higher mobility compared to any other 1 substrate.4,5 High mobility of graphene on hBN is enabled by extremely flat surface of hBN and ab- scence of dangling bonds at the graphene/hBN in- terface.5 Another important challenge for the use of graphene in devices is the lack of controllable bandgap.1 A bandgap can be opened through quantum confinement by patterning graphene into the so-called graphene nano-ribbons (GNRs).6,7 However, it is difficult to control the bandgap in GNRs due to its sensitivity to the width and edge geometry.7 Alternatively, the bandgap can be opened by chemical functionalization of graphene with a variety of species such as H, F, OH, etc.8 The hybridization of the functionalized C atom changes from sp2 to sp3, which opens up a bandgap. The bandgap opened by this mecha- nism is expected to be more robust in the presence of disorder compared to the bandgap opened by the quantum confinement.9 The bandgap opening by H-functionalization/hydrogenation of graphene has been a subject of several recent experimen- tal and theoretical studies, which show that the bandgap of graphene can be tuned by controlling the degree of hydrogenation.10 -- 14 Here we report the electronic structure of graphene and single-sided hydrogenated graphene (graphone) supported on the hBN substrate calcu- lated using the first-principles density functional method and the many-body perturbation theory in the GW approximation, the state-of-the-art method for accurate predictions of the electronic structure. Theoretical studies have suggested a possibility of inducing a bandgap in graphene when supported on the hBN substrate,15 how- ever, recent experimental studies find no bandgap in this system.4,5 Earlier calculations based on the tight-binding model ascribe this discrepancy to the random stacking arrangement of graphene on hBN.5 The tight-binding model can not be reliably used for quantitative predictions involv- ing novel materials such as graphene on hBN especially because the interlater hopping param- eters between graphene and hBN are not known. First-principles calculations are therefore required for accurate prediction of electronic structure of such novel material systems. Our calculations show that slight misalignment of graphene closes the bandgap induced by hBN in Bernal stacked graphene. We consider hydrogenation as an alter- native to induce bandgap in graphene supported on hBN. Hydrogenation of a substrate supported graphene results in 50% hydrogen coverage16 and the resulting material is called as graphone.8,13 Free standing graphone exhibits a bandgap larger than 2.5 eV. Interestingly, the bandgap of gra- phone reduces by about 1 eV due to the substrate induced polarization effects17 -- 20 when it is sup- ported on hBN. However, unlike graphene, the bandgap of graphone is unaffected by the mis- alignment with respect to hBN. The band offsets at graphone/hBN interface suggest that hBN can be used not only as a substrate but also as a dielec- tric in the field effect devices employing graphone as a channel material. The electronic structure calculations are per- formed in the framework of density functional the- ory (DFT) within the local density approximation (LDA) as implemented in the ABINIT code.21 The Trouiller-Martins norm-conserving pseudopoten- tials22 and the Teter-Pade parameterization for the exchange-correlation functional23 are used. To en- sure negligible interaction between periodic im- ages, a large value (10 Å) of the vacuum re- gion is used. The Brillouin zone is sampled us- ing Monkhorst-Pack meshes of different size de- pending on the size of the unit cell: 18 × 18 × 1 for Bernal stacked graphene on hBN, 6 × 6 × 1 for misaligned graphene on hBN, 18× 18× 1 for chair-graphone, and 8 × 8× 1 for boat-graphone. For the plane wave expansion of the wavefunc- tion, a 30 Ha kinetic energy cut-off is used. The quasiparticle corrections to the LDA bandstructure are calculated within the G0W0 approximation and the screening is calculated using the plasmon-pole model.24 The Vienna ab initio simulation package (VASP),25 which provides well-tested implemen- tation of van der Waals interactions (vdW),26 is used to compute the equilibrium distance be- tween graphene (or graphone) and hBN. The PAW pseudopotentials,27 the PBE exchange-correlation functional in the GGA approximation,28 and the DFT-D2 method of Grimme29 are used. Same values of energy cutoff, vacuum region, and k- point grid as in the ABINIT calculations are used in VASP calculations. The optimized geometries calculated using VASP and ABINIT without in- 2 graphene supported on hBN show no evidence of bandgap.4,5 The perfect Bernal stacking of graphene on hBN is difficult to achieve in the ex- periments and random orientation is more proba- ble. To mimic the random orientation, we simu- late larger supercells where the stacking between the graphene layer and the underlying hBN sub- strate deviates from the ideal Bernal stacking. To reduce the computational requirements only one layer of hBN is included. In the Bernal stacking, the graphene layer is rotated by an angle θ = 30◦ with respect to hBN. We consider three rotation angles 21.8◦, 32.2◦, and 13.2◦. The commensu- rate supercells for these rotations contain 28 (14 C, 7 B, 7 N), 52 (26 C, 13 B, 13 N), and 76 (38 C, 19 B, 19 N) atoms respectively. The super- cell for 21.8◦ rotation is depicted in the inset of Figure [figure][1][]1(b). The remaining two su- percells and the Brillouin zones of all three su- percells are shown in the supplementary material. The commensuration conditions derived in30 are used to generate the supercells. In these supercells, the sublattice asymmetry induced by the hBN sub- strate on the graphene layer is significantly re- duced compared to the Bernal stacking. The LDA density of states of the misaligned graphene (θ = 21.8◦) is compared with that of the Bernal stacked graphene in Figure [fig- ure][1][]1(b). The Bernal stacked graphene has a finite bandgap, which closes due to the slight misalignment. Since LDA is known to under- estimate the bandgap, we have calculated the GW corrections to the bandgap of misaligned graphene. The GW corrected bandgap of the Bernal stacked graphene increases from ELDA g = 68 meV to EGW g = 145 meV while the GW cor- rected bandgap of the misaligned graphene re- mains 0. The GW bandgaps for rotations 32.2◦ and 13.2◦ are also 0. The Moiré patterns larger than those shown in Figure [figure][1][]1(b) were recently observed in graphene supported on hBN.5 These graphene samples indeed showed 0 bandgap similar to our calculations. The above calculations indicate that due to mis- alignment hBN substrate can not reliably induce a bandgap in graphene and alternative approaches are required to open up a sizable bandgap. Recent experimental measurements and theoretical calcu- lations show that the bandgaps on the order of 1 eV Figure 1: (a) The LDA and GW bandstructures of Bernal stacked graphene on hBN substrate. The inset shows a small bandgap opening at the K point. (b) Bandgap closing due to misalignment illusrated by the LDA density of states and GW bandgaps of the heterogeneous bilayer of graphene and hBN in perfect Bernal and misaligned stack- ing arrangements. The inset shows atomistic schematic of a commensurate unit cell of the mis- aligned bilayer. cluding vdW interactions are found to agree well with each other. Figure [figure][1][]1(a) shows LDA and GW bandstructures of Bernal stacked graphene on hBN substrate such that half of the C atoms in graphene are positioned exactly above the B atoms. This stacking arrangement has been found to be low- est energy configuration.15 The equlibrium dis- tance between graphene and hBN is 3.14 Å. Three monolayers of the semi-infinite hBN substrate are included in the simulation domain to ensure that the GW bandgap is converged. The bands con- tributed by hBN are identified separately. The weak interlayer interaction between graphene and hBN allows graphene to retain its linear bandstruc- ture near the K point. Underlying hBN substrate induces sublattice asymmetry on the graphene lat- tice opening up a small bandgap at the K point as shown in the inset of Figure [figure][1][]1(a). The LDA bandgap is smaller by about 0.1 eV com- pared to the GW corrected bandgap. Contrary to the above conclusion and earlier the- oretical studies, recent transport measurements on 3 −1−0.500.5100.010.020.030.04DOS (/eV.Atom)E−EF (eV)MisalignedBernalEg (LDA) = 32 meV Eg (GW) = 102 meVEg (LDA/GW) = 0(b)BNCKΓMK−5−4−3−2−1012345E−EF (eV)−101 K ΓEg (GW)= 166 meVGWLDA(a)orbing hydrogen from one side of graphane.31,32 The hybridization of C atom changes from sp2 to sp3 upon hydrogenation and the planar structure of graphene becomes non-planar as depicted in Fig- ure [figure][2][]2. Graphene is weakely bonded to hBN by van der Waals interaction and the frontier orbitals of graphene, which take part in bonding with hydro- gen are virtually unaffected by hBN. Therefore the stable configurations of graphone are likely to be unaffected by hBN. Indeed, the optimized atomic structures of graphone on hBN are virtu- ally identical to earlier studies,8,13,16 which did not include any substrate. The situation is differ- ent with the substrates, which have stronger inter- action with graphene. For example, graphene on Ir(111) substrate exhibits specific hydrogenation patterns based on the local orientation of graphene with respect to Ir(111) surface.10 Hydrogen atoms may desorb or diffuse through graphone into the hBN substrate at elevated temperatures thereby af- fecting the local electronic struture of graphone. However, such phenomena require in depth analy- sis using methods such as the molecular dynamics and are out of the scope of the present work. Chair and boat-graphone exhibit very different electronic and magnetic properties as evident from their bandstructures (Figure [figure][2][]2). The ground state of chair-graphone is ferromagnetic while boat-graphone has a nonmagnetic ground state. The bandstructure of chair-graphone is highly spin polarized with a spin splitting of 2.79 eV, which is also its bandgap. This fea- ture makes chair-graphone an attractive material for spintronics.8 Boat-graphone has a spin de- generate bandstructure and behaves as an insu- lator with a large bandgap of 5.15 eV. Hydro- gen atoms may migrate over the graphone lattice resulting in the combination of chair- and boat- configurations. In such structures magnetic order- ing of chair-graphone may not be preserved. The bandgap is, however, expected to persist.9 The electronic structure of an atomically thin material such as graphone, when it is used in de- vices, is expected to be highly dependent on the surrounding materials. The atomistic schemat- ics of graphone supported on hBN substrate and calculated bandstructures of graphone-hBN super- cells are shown in Figure [figure][3][]3. The gra- Figure 2: Atomistic schematics and bandstructures of graphone in (a) chair and (b) boat conforma- tions. Unit cells are depicted by the dotted lines in the top views of atomistic schematics while Bril- louin zones are shown in the insets of bandstruc- ture plots. can be opened up in hydrogenated graphene.10 -- 13 Furthermore, the bandgap can be tuned by con- trolling the degree of hydrogenation. Now, the question arises, what is the effect of hBN sub- strate on the electronic structure of hydrogenated- graphene? To address this issue, we consider single-sided semi-hydrogenated graphene, which is referred to as graphone.13 Recent experimental and theoretical study has shown that hydrogena- tion of a substrate supported graphene results in 50% hydrogen coverage,16 which we have used for the present study. Graphone has two distinct configurations chair (Figure [figure][2][]2(a)) and boat (Figure [figure][2][]2(b)), depending on the relative placement of hydrogenated C atoms. Both chair and boat configurations are energetically sta- ble8,13,16 and they can be synthesized by differ- ent hydrogenation processes. A direct single-sided hydrogenation of graphene is likely to result in boat-graphone, which is more stable compared to chair-graphone.8,16 On the other hand, chair- graphone can be synthesized by selectively des- 4 KΓMK−5−4−3−2−1012345E−EF (eV) ΓM1RΓM2 −5−4−3−2−1012345E−EF (eV)2.79 eV5.15 eVGWGW LDA LDAGW LDA(b)(a)ГMKГM1M2RCHexpected in hydrogenated graphene with different hydrogen coverage. The bandgaps of GNRs are also reduced when they are supported on hBN sub- strate.33 Non-local polarization effects can not be modeled in the LDA. The GW approach includes non-local polarization effects through the screened Coulomb interaction, however, it tends to be com- putationally extremely demanding. j; f ree + ∆Pj where EGW The bandgap reduction due to hBN substrate polarization can be estimated by using com- putationally much less demanding image-charge model.20 In this model, the quasiparticle en- ergy of a substrate-supported layer is given by EQP j;supported = EGW j; f ree is the quasiparticle GW energy of state j(cid:105) of free stand- ing layer, graphone in this case, and ∆Pj is the cor- rection due to substrate polarization. The quasi- particle bandgaps of chair and boat-graphone esti- mated using the image-charge model are 2.03 eV and 4.36 eV respectively, which compare well with the bandgaps obtained using computationally extensive GW calculations including hBN sub- strate (Figure [figure][3][]3). The details of image- charge model calculations and underlying assump- tions are discussed in the supplemental material. In addition to the bandgaps, the band offsets are of crucial importance for the successful use of a heterostructure in the electronic devices. As an illustration, we analyse the conduction and valence band offsets, denoted by ∆Ec and ∆Ev, respectively, at the chair-graphone/hBN interface when hBN is placed on both the hydrogenated and non-hydrogenated sides of graphone. Fig- ure [figure][4][]4(b) shows a schematic depict- ing the calculated band offsets in the hBN/chair- graphone/hBN heterostructure. Three monolayers of hBN are included on either sides of graphone. The band offsets do not change when more than 3 monolayers of hBN are included. The bandgap of chair-graphone on hBN reduces from 1.93 eV (Figure [figure][3][]3) to 1.74 eV in the presence of hBN on the hydrogenated side of graphone. This is due to the additional reduction of screened Coulomb interaction induced by the polarization of hBN layers on the hydrogenated side. The band offsets at chair-graphone/hBN interfaces are asymmetric such that the bands of hBN on the hydrogenated side of graphone are lower in energy compared to the bands of Figure 3: Atomistic schematics and bandstructures of (a) chair-graphone and (b) boat-graphone sup- ported on hBN substrate. The substrate polar- ization induced renormalization of energy levels reduces the bandgap of substrate-supported gra- phone compared to the free-standing graphone. phone LDA bands in the vicinity of the valence- band maximum and the conduction-band mini- mum remain unaffected by the presence of hBN The shape of GW-corrected bands substrate. remains more or less unaffected, however, the bandgap of chair and boat-graphone when sup- ported on hBN substrate reduces by 0.86 eV and 0.72 eV respectively compared to the bandgap of their free-standing counterparts. The reduction in GW bandgap is attributed to the polarization effects at the graphone/hBN in- terface. Similar reductions have been found in experiments and GW calculations for several dif- ferent molecules adsorbed on metals, semicon- ductors, and insulators.17 -- 20 Careful inspection of Figures [figure][2][]2 and [figure][3][]3 indicates that the GW corrections to the LDA bands are smaller in graphone on hBN compared to those in free standing graphone. This is because of the fact that the polarization of hBN substrate reduces the screened Coulomb potential, W , which in turn re- duces the GW bandgap. Similar bandgap reduc- tion due to the polarization of hBN substrate is 5 −5−4−3−2−1012345E−EF (eV)−5−4−3−2−1012345E−EF (eV)1.93 eV4.43 eV(b)(a)KΓMKΓM1RΓM2CHBNdC-BN = 3.13 ÅdC-BN = 3.14 ÅGWGW LDA LDAdegrading the performance of the field-effect de- vice.35 On the contrary, when hBN is placed on the hydrogenated side of graphone, it has suffi- ciently high barriers for both electrons (∆Ec) and holes (∆Ev) to prevent their thermionic emission into the hBN dielectric layer. that In summary, our first-principles calculations suggest in a Bernal stacked graphene on hBN substrate a bandgap on the order of 0.1 eV opens up due to the sublattice asymmetry in graphene induced by hBN. This bandgap closes in the presence of slight misorientation from the Bernal stacking. Hydrogenation of graphene pro- vides a promising approach to open up bandgaps larger than 2.5 eV. The polarization effects due to surrounding hBN dielectrics, however, reduce bandgaps of graphone by about 1 eV. Thus, accu- rate electronic structure calculations of the atom- ically thin materials such as graphone should al- ways take into account the surrounding materi- als. The calculated band offsets suggest that in the field effect devices employing chair-graphone as a channel material, hBN can be used as a sub- strate as well as a dielectric layer separating the graphone channel and the gate electrode. Bandgap engineering of graphene by chemical functional- ization is currently an extremely active area of re- search and this work provides theoretical instruc- tion for analyzing the experimental observations such as the electronic structure modulation by the substrate. Acknowledgement We thank Prof. Timothy Boykin, Prof. Mathieu Luisier, and Prof. Gerhard Klimeck for helpful discussions. This work is supported partly by the Interconnect Focus Center funded by the MARCO program of SRC and State of New York, NSF PetaApps grant number 0749140, and an anony- mous gift from Rensselaer. Computing resources of the Computational Center for Nanotechnology Innovations at Rensselaer partly funded by State of New York and of nanoHUB.org funded by the National Science Foundation have been used for this work. Figure 4: Band offsets of chair-graphone em- bedded in hBN layers. (a) A supercell contain- ing 3 monolayers of hBN on both sides of chair- graphone. The direction of electric field induced by the dipoles in graphone layer is indicated by an arrow. (b) Band offsets at the graphone/hBN inter- faces. hBN on the non-hydrogenated side. To investi- gate the origin of this asymmetry, we carried out Bader charge analysis of a free standing chair- graphone. In chair-graphone, the hydrogenated C-atoms aquire a slight net positive charge while the non-hydrogenated C-atoms aquire a slight net negative charge creating a dipole layer with an electric field depicted by a thick grey arrow in Figure [figure][4][]4(a). This electric field causes asymmetry in the band offsets at graphone/hBN interfaces. Similar modifications of the band off- sets have been found in conventional semiconduc- tor homo- and heterojunctions when a dipole layer is inserted at the interface.34 In field effect devices employing chair-graphone as channel material, hBN can be used as a sub- strate as well as a dielectric layer separating the gate electrode and the graphone channel. The band diagrams in Figure [figure][4][]4(b) indicate that hBN can be placed on either side of graphone when used as a substrate. When hBN is used as a dielectric layer, however, it should be placed on the hydrogenated side and not on the non- hydrogenated side of graphone. This is because of the fact that when hBN is placed on the non- hydrogenated side of graphone, ∆Ev at the inter- face is too low (< 1 eV) to prevent the thermionic emission of holes into the hBN dielectric layer, which consequently results in gate leakage current 6 SubstrateDielectric{{−3−2−1012345E-EF (eV)Graphoneh−BNSubstrateh−BNDielectric1.741.700.852.162.92(b)Edipole(a)References (1) Schwierz, F. Graphene transistors. Nature Nanotech. 2010, 5, 487 -- 496. (2) Chen, J.-H.; Jang, C.; Xiao, S.; Ishigami, M.; Fuhrer, M. S. Intrinsic and extrinsic perfor- mance limits of graphene devices on SiO2. Nature Nanotech. 2008, 3, 206 -- 209. (3) Ponomarenko, L. A. et al. Effect of a High- kappa Environment on Charge Carrier Mo- bility in Graphene. Phys. Rev. Lett. 2009, 102, 206603. (4) Dean, C. R. et al. Boron nitride substrates for high-quality graphene electronics. Nature Nanotech. 2010, 5, 722 -- 726. (5) Xue, J. et al. Scanning tunnelling microscopy and spectroscopy of ultra-flat graphene on hexagonal boron nitride. Nature Mater. 2011, 10, 282 -- 285. (6) Han, M. Y.; Oezyilmaz, B.; Zhang, Y.; Kim, P. Energy band-gap engineering of graphene nanoribbons. Phys. Rev. Lett. 2007, 98, 206805. (7) Son, Y.-W.; Cohen, M. L.; Louie, S. G. En- ergy gaps in graphene nanoribbons. Phys. Rev. Lett. 2006, 97, 216803. (8) Li, L. et al. Functionalized Graphene for High-Performance Two-Dimensional Spin- tronics Devices. ACS Nano 2011, 5, 2601 -- 2610. (9) Abanin, D. A.; Shytov, A. V.; Levitov, L. S. Peierls-Type Instability and Tunable Band Gap in Functionalized Graphene. Phys. Rev. Lett. 2010, 105, 086802. (10) Balog, R. et al. Bandgap opening in graphene induced by patterned hydrogen adsorption. Nature Mater. 2010, 9, 315 -- 319. (11) Haberer, D. et al. Tunable Band Gap in Hy- drogenated Quasi-Free-standing Graphene. Nano Lett. 2010, 10, 3360 -- 3366. (12) Elias, D. C. et al. Control of Graphene's Properties by Reversible Hydrogenation: Ev- idence for Graphane. Science 2009, 323, 610 -- 613. (13) Zhou, J. et al. Ferromagnetism in Semihydro- genated Graphene Sheet. Nano Lett. 2009, 9, 3867 -- 3870. (14) Fiori, G. et al. Simulation of hydrogenated graphene field-effect transistors through a multiscale approach. Phys. Rev. B 2010, 82, 153404. (15) Giovannetti, G.; Khomyakov, P. A.; Brocks, G.; Kelly, P. J.; van den Brink, J. Substrate-induced band gap in graphene on hexagonal boron nitride: Ab initio density functional calculations. Phys. Rev. B 2007, 76, 073103. (16) Subrahmanyam, K. S. et al. Chemical stor- age of hydrogen in few-layer graphene. Proc. Natl. Acad. Sci. USA 2011, 108, 2674 -- 2677. (17) Neaton, J. B.; Hybertsen, M. S.; Louie, S. G. Renormalization of molecular electronic lev- els at metal-molecule interfaces. Phys. Rev. Lett. 2006, 97, 216405. (18) Thygesen, K. S.; Rubio, A. Renormalization of Molecular Quasiparticle Levels at Metal- Molecule Interfaces: Trends across Binding Regimes. Phys. Rev. Lett. 2009, 102, 046802. (19) Freysoldt, C.; Rinke, P.; Scheffler, M. Con- trolling Polarization at Insulating Surfaces: Quasiparticle Calculations for Molecules Adsorbed on Insulator Films. Phys. Rev. Lett. 2009, 103, 056803. (20) Li, Y.; Lu, D.; Galli, G. Calculation of Quasi-Particle Energies of Aromatic Self- Assembled Monolayers on Au(111). J. Chem. Theory Comput. 2009, 5, 881 -- 886. (21) Gonze, X. et al. ABINIT: First-principles approach to material and nanosystem prop- erties. Computer Physics Communications 2009, 180, 2582 -- 2615. 7 (33) Jiang, X.; Kharche, N.; Nayak, S. K. (to be published). (34) Pan, M. et al. Modification of band offsets by a ZnSe intralayer at the Si/Ge(111) interface. Appl. Phys. Lett. 1998, 72, 2707 -- 2709. (35) Robertson, J. Band offsets of wide-band-gap oxides and implications for future electronic devices. J. Vac. Sci. Technol. B 2000, 18, 1785 -- 1791. (22) Troullier, N.; Martins, J. L. Efficient pseu- dopotentials for plane-wave calculations. Phys. Rev. B 1991, 43, 1993 -- 2006. (23) Goedecker, S.; Teter, M.; Hutter, J. Separable dual-space Gaussian pseudopotentials. Phys. Rev. B 1996, 54, 1703 -- 1710. (24) Hybertsen, M. S.; Louie, S. Electron correla- tion in semiconductors and insulators: Band gaps and quasiparticle energies. Phys. Rev. B 1986, 34, 5390 -- 5413. (25) Kresse, G.; Furthmüller, J. Efficiency of ab- initio total energy calculations for metals and semiconductors using a plane-wave basis set. Computational Materials Science 1996, 6, 15 -- 50. (26) Bucko, T.; Hafner, J.; Lebegue, S.; Angyan, J. G. Improved Description of the Structure of Molecular and Layered Crystals: Ab Initio DFT Calculations with van der Waals Corrections. Journal of Physi- cal Chemistry A 2010, 114, 11814 -- 11824. (27) Kresse, G.; Joubert, D. From ultrasoft pseu- dopotentials to the projector augmented- wave method. Phys. Rev. B 1999, 59, 1758 -- 1775. (28) Perdew, J.; Burke, K.; Ernzerhof, M. Gener- alized gradient approximation made simple. Phys. Rev. Lett. 1996, 77, 3865 -- 3868. (29) Grimme, S. Semiempirical GGA-type den- sity functional constructed with a long-range dispersion correction. Journal of Computa- tional Chemistry 2006, 27, 1787 -- 1799. (30) Shallcross, S.; Sharma, S.; Pankratov, O. A. Quantum interference at the twist bound- ary in graphene. Phys. Rev. Lett. 2008, 101, 056803. (31) Sofo, J. O.; Chaudhari, A. S.; Barber, G. D. Graphane: A two-dimensional hydrocarbon. Phys. Rev. B 2007, 75, 153401. (32) Zhou, J.; Wu, M. M.; Zhou, X.; Sun, Q. Tuning electronic and magnetic properties of graphene by surface modification. Appl. Phys. Lett. 2009, 95, 103108. 8 

1712.03238

2

1712

Fibonacci Topological Superconductor

We introduce a model of interacting Majorana fermions that describes a superconducting phase with a topological order characterized by the Fibonacci topological field theory. Our theory, which is based on a $SO(7)_1/(G_2)_1$ coset factorization, leads to a solvable one dimensional model that is extended to two dimensions using a network construction. In addition to providing a description of the Fibonacci phase without parafermions, our theory predicts a closely related "anti-Fibonacci" phase, whose topological order is characterized by the tricritical Ising model. We show that Majorana fermions can split into a pair of Fibonacci anyons, and propose an interferometer that generalizes the $Z_2$ Majorana interferometer and directly probes the Fibonacci non-Abelian statistics.

cond-mat.mes-hall

cond-mat

Fibonacci Topological Superconductor Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104 Yichen Hu and C. L. Kane We introduce a model of interacting Majorana fermions that describes a superconducting phase with a topological order characterized by the Fibonacci topological field theory. Our theory, which is based on a SO(7)1/(G2)1 coset factorization, leads to a solvable one dimensional model that is extended to two dimensions using a network construction. In addition to providing a description of the Fibonacci phase without parafermions, our theory predicts a closely related "anti-Fibonacci" phase, whose topological order is characterized by the tricritical Ising model. We show that Majorana fermions can split into a pair of Fibonacci anyons, and propose an interferometer that generalizes the Z2 Majorana interferometer and directly probes the Fibonacci non-Abelian statistics. Current interest in topological quantum phases is heightened by the proposal to use them for quantum information processing[1, 2] and by prospects for real- izing them in experimentally viable electronic systems. There is growing evidence that the fractional quantum Hall (QH) state at filling ν = 5/2 is a non-Abelian state[3–7] with Ising topological order. A simpler form of Ising order is predicted in topological superconductors (T-SC)[8, 9] and in SC proximity effect devices[10–14]. In these systems the Ising σ particle is not dynamical, but is associated with domain walls or vortices that host gapless Majorana fermion modes. Recent experiments have found promising evidence for Majorana fermions in 1D and 2D SC systems[15–17]. Ising topological order is insufficient for universal quantum computation, but the richer Fibonacci topolog- ical order is sufficient[18]. Fibonacci order arises in the Z3 parafermion state introduced by Read and Rezayi[19], which is a candidate for the fractional QH state at ν = 12/5. Parafermions can also be realized by com- bining SC with the fractional QH effect[20–24]. This line of inquiry culminated in the tour de force works[25, 26] that showed a ν = 2/3 QH state, appropriately proximi- tized, could exhibit a Fibonacci phase. In this paper we introduce a different formulation of the Fibonacci phase based on a model of interacting Ma- jorana fermions. Our starting point is a system of chiral Majorana edge states, which can in principle be realized in SC proximity effect structures. We show that a par- ticular four fermion interaction leads to an essentially exactly solvable model that realizes the Fibonacci phase. In addition to providing a direct route to the Fibonacci phase without parafermions, our theory reveals a distinct but closely related "anti-Fibonacci" state that is a kind of particle-hole conjugate to the Fibonacci state with a topological order that combines Ising and Fibonacci. Our formulation also suggests a method for experimentally probing the Fibonacci state. We introduce a generaliza- tion of the interferometer introduced earlier for Majorana states[27, 28], and argue that it provides a method for unambiguously detecting Fibonacci order. hibit a Fibonacci phase is foreshadowed by Rahmani, et al. [29](RZFA), who showed that a 1D Majorana chain with strong interactions can be tuned to the tricritical Ising (TCI) critical point. The same critical point arises in the 1D "golden chain" model of coupled Fibonacci anyons[30], as well as at interfaces connecting Ising and Fibonacci order in the QH effect[31]. There is a sense in which the TCI point of the RZFA model is like a Fi- bonacci chain, but it is not clear how to extend it to 2D. Our theory provides a method for accomplishing that. Mong et al. [25] formulated the Fibonacci phase using a "trench" construction that began with 1D strips of ν = 2/3 QH states coupled along trenches in the presence of a SC. A single trench mapped to the 3 state clock model, with a critical point described by the Z3 parafermion con- formal field theory (CFT). The resulting 1D states were coupled to create a gapped 2D phase. This is similar to the coupled wire construction[32] for the Read Rezayi state introduced in Ref. 33, but differs in an important way. That model was based on the coset construction[34– 36], which allows a simple CFT ([SU (2)1]3 with cen- tral charge c = 3) to be factored into less trivial CFTs (SU (2)3 + SU (2)3 1/SU (2)3 with c = 9/5 + 6/5). This ex- act factorization identifies a solvable coupled wire Hamil- tonian, where counter-propagating modes of the two fac- tors pair up differently, resulting in a non-trivial unpaired chiral edge mode[33, 37]. The construction in this paper is based on the coset SO(7)1/(G2)1[38]. SO(7)1 describes 7 free chiral Majo- rana modes with c = 7/2. G2 is a Lie group that sits inside SO(7). (G2)1, with c = 14/5, is the Fibonacci CFT[25, 39]. The quotient is a CFT with c = 7/2 − 14/5 = 7/10, (1) which can be identified with the TCI model. Thus, the edge states of a non-interacting T-SC with Chern number n = 7 factor into a (G2)1 Fibonacci (FIB) sector and a SO(7)1/(G2)1 TCI sector. In the following we will design an interaction that separates the factors and leads to 2D topological phases with either c = 14/5 (Fibonacci) or c = 7/10 (anti-Fibonacci) edge states. The fact that interacting Majorana fermions can ex- We begin with some facts about G2, which is well known in mathematical physics[38, 40]. G2 is the sim- plest exceptional Lie group. Its relation to SO(7) involves (cid:80)7 the mathematics of the octonion division algebra[41]. An octonion is specified by 8 real numbers: q = q0 + a=1 qaea, where ea are 7 square roots of −1 that sat- isfy the non-associative multiplication rule eaeb = −δab + Cabcec. (2) Cabc is a totally antisymmetric tensor. It is not unique, but can be chosen to satisfy[41] Ca+1b+1c+1 = Cabc, C124 = 1, (3) where the indices are defined mod 7. Eq. 3 along with antisymmetry specifies all the non-zero elements of Cabc. ea define a set of 7 unit vectors that transform under SO(7). However, not all SO(7) rotations preserve (2). G2 is the automorphism group of the octonions: the sub- group of SO(7) that preserves Cabc. The 21 generators of SO(7) can be represented by 7 × 7 skew symmetric matrices T m,n of the form T m,n ab = i(δmaδnb − δmbδna). There are 14 combinations that pre- serve Cabc, which can be written[40] T A,A+2−T A+1,A+5 M A = T A,A+2+T A+1,A+5−2T A+3,A+4 √ √ 2 6 1 ≤ A ≤ 7 8 ≤ A ≤ 14. (4) (cid:88) These matrices are normalized by Tr[M AM B] = 2δAB and represent the generators of G2 in the 7D fundamental representation, analogous to the Pauli matrices of SU (2). In what follows, it will be useful to express the quadratic Casimir operator as 2 3 (5) where ∗Cabcd =
abcdef gCef g/6 is the dual of Cabc whose non-zero elements follow from ∗C3567 = −1, as in (3). (δadδbc − δacδbd) − 1 3 ∗ Cabcd abM A cd = M A A We now consider the coset factorization of a 1D system of 7 free chiral Majorana fermions described by (cid:40) 7(cid:88) a=1 H0 = − iv 2 γa∂xγa. (6) We adopt a Hamiltonian formalism[42] with Majorana operators satisfying {γa(x), γb(x(cid:48))} = δ(x − x(cid:48))δab. H0 describes a SO(7)1 Wess Zumino Witten (WZW) model with c = 7/2. The coset construction allows this to be written H0 = HFIB + HTCI. The FIB sector is expressed in terms of (G2)1 currents in Sugawara form [36][43], HFIB = πvJ AJ A k + g , J A = 1 2 M A abγaγb, (7) with k = 1, g = 4. Using (5), the operator product gives HFIB = − 2iv 5 HTCI = − iv 10 ∗Cabcdγaγbγcγd. (8) γa∂xγa − πv 60 ∗Cabcdγaγbγcγd, γa∂xγa + πv 60 a A (cid:88) (cid:88) (cid:88) a ab (cid:88) (cid:88) (cid:88) abcd abcd 2 non-chiral system with interaction λ(cid:80) FIG. 1. (a) 7 chiral Majorana edge modes factor into FIB (b) A 1D and TCI sectors with c = 14/5 + 7/10 = 7/2. L transmits the TCI sector, but reflects the FIB sector. The bottom panels show network constructions for the Fibonacci phase (c) and the anti-Fibonacci phase (d). A J A R J A is (cid:104)Hα(x)Hβ(x(cid:48))(cid:105) = The correlator of Hα=FIB,TCI v2δαβcα/8π2(x− x(cid:48))4, with cFIB = 14/5 and cTCI = 7/10 [44]. This shows that H0 decouples into two independent sectors, as depicted in Fig. 1a. HFIB describes a (G2)1 WZW model, with two primary fields 1, τ of dimension h = 0, 2/5. τ transforms under the 7D representation of G2 and obeys the Fibonacci fu- sion algebra τ × τ = 1 + τ . HTCI describes the M (5, 4) minimal CFT with 6 primary fields 1,
,
(cid:48),
(cid:48)(cid:48), σ, σ(cid:48), with h = 0, 1/10, 3/5, 3/2, 3/80, 7/16[36]. The Majo- rana fermion operator γa factors into the product γa = τa ×
(9) with h = 2/5 + 1/10 = 1/2. The 21 bilinears iγaγb decompose into 14 J A's, along with 7 operators τa ×
(cid:48) with h = 2/5 + 3/5 = 1. J A act only in the FIB sector: [J A, HTCI] = 0. The trilinear combination Cabcγaγbγc is
(cid:48)(cid:48) with h = 3/2 and acts only in the TCI sector. We now introduce a 1D model of 7 non-chiral Majo- rana fermions γaR/Lwith an interaction that gaps the FIB sector, leaving the TCI sector gapless. Consider H = − iv 2 (γaR∂xγaR−γaL∂xγaL)+λ (cid:88) (cid:88) L , (10) J A R J A a A where J A R/L are given in (7). The λ term commutes with HTCI, so it operates only in the FIB sector. A per- turbative renormalization group analysis gives dλ/d(cid:96) = −2λ2/πv, so λ < 0 is marginally relevant. When λ flows to strong coupling it is natural to expect that it leads to a gap ∆ ∝ e−πv/2λ in the FIB sector and a gapless TCI critical point. This is similar to the RZFA model, except the G2 symmetry locates the critical point exactly. The exact factorization allows the two sectors to be separated. Consider the 1D system in Fig. 1b, with FIBA-FIBc=7/2c=14/5 (FIB)c=7/10 (TCI)=(a)(b)(c)(d)0Lγετλ(x) (cid:54)= 0 for 0 < x < L. Provided L (cid:29) ξ = v/∆, the gap in the FIB sector leads to an exponential suppression of transmission. The FIB sector will be perfectly reflected, while the TCI sector will be perfectly transmitted. In- terestingly, this means an incident Majorana fermion γa splits, with τa reflected and
transmitted. This forms the basis for the interferometer to be discussed below. We wish to use (10) to construct a 2D gapped topo- logical phase. One approach is to adapt the coupled wire model[32]. This requires coupling right movers of the TCI sector on wire i to left movers of the TCI sector on wire i + 1. If this gaps the TCI sector, then we will have a 2D gapped phase with TCI edge states. This is prob- lematic, however, because the simplest tunneling term that can be built from local operators and does not cou- ple to the gapped FIB sector is the trilinear Cabcγaγbγc. The resulting tunneling term u
(cid:48)(cid:48) i+1L, with dimension 3, is perturbatively irrelevant. This does not preclude the possibility of a gapped phase for large u, but a non- perturbative analysis would be necessary to establish it. Fortunately, however, the exact factorization of the coset model allows for an alternative network construction, in- spired by the Chalker Coddington model[45]. iR
(cid:48)(cid:48) Fig. 1c shows a network of n = 7 T-SC islands in which each island has 7 chiral Majorana modes. In the absence of coupling the Majorana modes are localized on each island, so the system is a trivial SC. If the is- lands are strongly coupled by single particle tunneling they will merge, and the system is a n = 7 T-SC. In the absence of interactions, the transition between these phases will have 7 gapless 2 + 1D Majorana modes. For strong interactions intermediate topological phases can arise. We turn off the single particle tunneling and cou- ple the neighboring islands with the interaction term in (10). Provided the contact length L (cid:29) ξ, the excita- tions in the FIB sector will be reflected from the contact, which means they are transmitted to the next island. Ex- citations in the TCI sector, however, are transmitted by the contact, so they remain localized on the same island. From Fig. 1c, it can be seen that both the TCI and the FIB sectors are localized in the interior of the network. The TCI states are localized on the islands, while the FIB states are localized on the dual lattice of voids be- tween the islands. Since all bulk states are localized in finite, lattice scale regions, there will be a bulk excita- tion gap. The perimeter of the network, however has a gapless FIB edge state with c = 14/5. We emphasize that though fine tuning is required to achieve the exactly solvable Hamiltonian (10), the tuning does not need to be perfect. This gapped Fibonacci phase will be robust to finite single particle tunneling and other interactions. Fig. 1d shows a similar network that is surrounded by a n = 7 chiral Majorana edge state. This leads to a distinct phase that also has a bulk gap, but has TCI edge states with c = 7/10. This state can be viewed as a Fibonacci phase sitting inside a n = 7 T-SC, with c = 1 τ 1 1
(cid:48) ψ
(cid:48)(cid:48)
3 σi σ(cid:48) σ TABLE I. The 6 quasiparticles of the TCI model can be iden- tified with combinations of Ising and Fibonacci quasiparticles. 7/2 − 14/5. We call this the "anti-Fibonacci" in analogy with the "anti-pfaffian" [46, 47], which is the pfaffian sitting inside a ν = 1 QH state. The anti-Fibonacci has a topological order associated with the TCI CFT. However, the 6 TCI quasiparticles can also be understood as a combination of 1, τ Fibonacci quasiparticles with the 1, ψ, σi Ising quasiparticles. The TCI fusion rules[36] of the quasiparticles identified in Table I are reproduced by the simpler Fibonacci and Ising fusion rules (e.g. σi × σi = 1 + ψ). Similar fusion rule decompositions have been identified for other theories[31, 39]. As in the T-SC σ and σ(cid:48) are not dynamical quasiparticles, but they will be associated with h/2e vortices in the SC. Depending on the energetics, a SC vortex in the anti-Fibonacci phase will bind either a σ or σ(cid:48). If it is σ, then the vortex binds a Fibonacci anyon. Likewise in the Fibonacci phase, a vortex could bind 1 or τ [25]. The above considerations suggest a possible route to- wards realizing the Fibonacci phase is to start with a sys- tem close to a multi-component T-SC - trivial SC tran- sition. This could be achieved by introducing SC via the proximity effect into a 2D electron gas in the vicinity of a quantum Hall plateau transition with degenerate Lan- dau levels. Progress in this direction has recently been re- ported in a quantum anomalous Hall insulator coupled to a SC, where a plateau observed in the two terminal con- ductance was attributed to T-SC[17]. Another promising venue is graphene, which has a four-fold degenerate ze- roth Landau level. Coexistence of SC with the quantum Hall effect in these systems appears feasible[48, 49]. If the Fibonacci and/or the anti-Fibonacci T-SC can be realized, then it will be important to develop ex- perimental protocols for probing them. One approach is to measure the thermal Hall conductance, which di- rectly probes the central charge c of the edge states: κxy = cπ2T k2 B/3h. This has proven to be a power- ful method for identifying the topological order of QH states[7, 50, 51], but it does not directly probe the non- Abelian quasiparticle statistics. In the QH effect, Fabry Perot[52–54] and Mach Zehnder[55, 56] interferometers have been proposed for this purpose. Here we introduce a distinct interferometer that generalizes the Majorana fermion interferometer[27, 28]. Fig. 2 shows a Hall bar with 4 Ohmic contacts (C1-4) where the electron density is adjusted so that adjacent regions have QH filling factors ν = 1 and ν = 4. The middle is coupled to a SC that leads to a n = 1 T-SC 4 cled by anyon b is given by the monodromy matrix[54] Mab = SabS11/Sa1Sb1, which depends the topological data in the modular S-matrix Sab. We therefore predict I2 = e2 h t11MabV1, (11) where a and b are labels for the transmitted and localized quasiparticles. Provided quasiparticles can be introduced to the island without modifying t11, (which depends on the local Hamiltonian near the edges) the ratios of the conductances for different localized quasiparticles will be universal (note Ma1 = 1). Other proposed interferomet- ric measurements of Fibonacci statistics have challenges similar to controlling t11[39, 54]. A possible (albeit more complicated) way to overcome that is to include a con- tact inside the island that allows quasiparticles to come and go, leading to telegraph noise[58]. For the FIB phase, where the transmitted quasiparticle is τ the universal ratio is determined by M FIB √ τ τ = −1/ϕ2, 5)/2 is the golden mean. (12)
b ψbi ψσi In the where ϕ = (1 + A-FIB phase, the ratios are determined by M TCI for b = 1,
,
(cid:48),
(cid:48)(cid:48), σ, σ(cid:48). These can be evaluated from the 6 × 6 TCI S-matrix[29]. However, the same results are obtained by treating the A-FIB as the FIB sitting inside Ising. Then, M TCI , where bi(f ) are the Ising (Fibonacci) decomposition of particle b from Table = −1 (which is I. The non-trivial Ising term is M I probed in the Majorana interferometer). In the A-FIB state, if a vortex binds σ, the extra quasiparticle can be controlled with a magnetic flux, and M TCI
b = M I M FIB τ bf
σ = +1/ϕ2. In this paper we have introduced a theory of the Fi- bonacci phase based on Majorana fermions near a multi- component topological critical point with strong interac- tions. While this phase has the same topological struc- ture as the parafermion based Fibonacci states, our the- ory clarifies the relation between the Fibonacci and ant- Fibonacci phases and shows the way in which Majorana fermions can fractionalize into Fibonacci anyons. It also points to a promising direction in the broader problem of searching for exotic topological phases in strongly in- teracting systems with massless single-particle Dirac or Majorana fermions. It is a pleasure to thank Jonathan Heckman, Abhay Pasupathy, Ady Stern and Jeffrey Teo for helpful discus- sions. This work was supported by a Simons Investigator grant from the Simons Foundation. [1] A. Kitaev, Annals of Physics 321, 2 (2006). [2] C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, Rev. Mod. Phys. 80, 1083 (2008). [3] G. Moore and N. Read, Nuclear Physics B 360, 362 (1991). FIG. 2. A Fibonacci interferometer in a Hall bar with Ohmic contacts C1-4 and SC in the shaded region. Dirac (Majorana) edge states are indicated by solid (dashed) lines. The c = 7/2 edge splits into FIB and TCI edges around the Fibonacci (a) or anti-Fibonacci (b) island. A quasiparticle adds a branch cut (dotted line) that modifies transmission from C1 to C2. † inψin−ψ region and a trivial n = 8 SC region. We assume that at the boundary between the n = 1 and n = 8 SCs there is an island of either Fibonacci (Fig. 2a) or anti-Fibonacci (Fig. 2b). This leads to the pattern of edge states shown. Suppose contact C1 is at voltage V1, and that the SC and the other 3 contacts are grounded. We use a Landauer-Buttiker formalism[57] to compute the current † in C2, given by I2 = vF(cid:104)ψ outψout(cid:105), where ψin(out) describe the ν = 1 chiral fermions entering (leaving) C2. V1 only affects ψin, which in the SC decomposes into 1(cid:105). γ0 comes directly γ0 + iγ(cid:48) from C1, but γ(cid:48) 1 comes from the region where τ and
split and then recombine. First suppose there are no quasi- particles on the island. γ(cid:48) 1 will be a linear combination j=1 t1jγj of the incident Majorana modes, where tij is a real orthogonal scattering matrix and γ2−7 are associ- ated with the c = 3 edge. Ignoring the contributions from the grounded contact C3, iγ0γ(cid:48) 1 = t11iγ0γ1. This relates I2 to the current coming out of C1, I2 = t11(e2/h)V1. 1[27, 28]. Thus I2 ∝ (cid:104)iγ0γ(cid:48) (cid:80)7 Quasiparticles localized on the island will modify this result. The transmitted particles will encounter a branch cut due to non-Abelian statistics that can modify the state of the localized quasiparticle. Provided the local Hamiltonian near the edge is not modified by the pres- ence of the extra quasiparticle, this will be purely of topological origin. The expectation value of the cur- rent will only be non-zero if the localized quasiparti- cle returns to its original state. The probability ampli- tude that anyon a returns to its original state when cir- ν = 1ν = 1ν = 4ν = 4c=1c=3c=1/2c=1/2c=1/2c=14/5c=7/10n=8 SCFIBc=1c=3n=1 T-SCc=4c=4τC2C4C3C1c=7/2τεγ1γ0ν = 1ν = 1ν = 4ν = 4c=1c=3c=1/2c=1/2c=1/2c=14/5c=7/10c=1c=3c=4c=4C2C4C3C1c=7/2τεγ1γ0A-FIBσγ'1γ'1n=8 SCn=1 T-SC(b)(a)[4] I. P. Radu, J. B. Miller, C. M. Marcus, M. A. Kastner, L. N. Pfeiffer, and K. W. West, Science 320, 899 (2008). and [5] M. Dolev, M. Heiblum, V. Umansky, A. Stern, 076803 (2009). [32] C. L. Kane, R. Mukhopadhyay, and T. C. Lubensky, Phys. Rev. Lett. 88, 036401 (2002). D. Mahalu, Nature 452, 829 (2008). [33] J. C. Y. Teo and C. L. Kane, Phys. Rev. B 89, 085101 5 [6] R. L. Willett, L. N. Pfeiffer, and K. W. West, Pro- ceedings of the National Academy of Sciences 106, 8853 (2009). [7] M. Banerjee, M. Heiblum, V. Umansky, D. Feldman, Y. Oreg, and A. Stern, arXiv:1710.00492. [8] N. Read and D. Green, Phys. Rev. B 61, 10267 (2000). [9] A. Y. Kitaev, Physics-Uspekhi 44, 131 (2001). [10] L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). [11] L. Fu and C. L. Kane, Phys. Rev. B 79, 161408 (2009). [12] R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010). [13] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). [14] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 82, 184516 (2010). [15] V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science 336, 1003 (2012). [16] S. Nadj-Perge, I. K. Drozdov, J. Li, H. Chen, S. Jeon, J. Seo, A. H. MacDonald, B. A. Bernevig, and A. Yaz- dani, Science 346, 602 (2014). [17] Q. L. He, L. Pan, A. L. Stern, E. C. Burks, X. Che, G. Yin, J. Wang, B. Lian, Q. Zhou, E. S. Choi, K. Mu- rata, X. Kou, Z. Chen, T. Nie, Q. Shao, Y. Fan, S.-C. Zhang, K. Liu, J. Xia, and K. L. Wang, Science 357, 294 (2017). [18] M. H. Freedman, M. Larsen, and Z. Wang, Communi- cations in Mathematical Physics 227, 605 (2002). [19] N. Read and E. Rezayi, Phys. Rev. B 59, 8084 (1999). [20] M. Barkeshli and X.-L. Qi, Phys. Rev. X 2, 031013 (2012). [21] N. H. Lindner, E. Berg, G. Refael, and A. Stern, Phys. Rev. X 2, 041002 (2012). [22] M. Cheng, Phys. Rev. B 86, 195126 (2012). [23] D. J Clarke, J. Alicea, and K. Shtengel, Nature commu- nications 4, 1348 (2013). [24] A. Vaezi, Phys. Rev. B 87, 035132 (2013). [25] R. S. K. Mong, D. J. Clarke, J. Alicea, N. H. Lindner, P. Fendley, C. Nayak, Y. Oreg, A. Stern, E. Berg, K. Sht- engel, and M. P. A. Fisher, Phys. Rev. X 4, 011036 (2014). [26] A. Vaezi, Phys. Rev. X 4, 031009 (2014). [27] A. R. Akhmerov, J. Nilsson, and C. W. J. Beenakker, Phys. Rev. Lett. 102, 216404 (2009). [28] L. Fu and C. L. Kane, Phys. Rev. Lett. 102, 216403 (2009). [29] A. Rahmani, X. Zhu, M. Franz, and I. Affleck, Phys. Rev. Lett. 115, 166401 (2015). [30] A. Feiguin, S. Trebst, A. W. W. Ludwig, M. Troyer, A. Kitaev, Z. Wang, and M. H. Freedman, Phys. Rev. Lett. 98, 160409 (2007). [31] E. Grosfeld and K. Schoutens, Phys. Rev. Lett. 103, (2014). [34] P. Goddard, A. Kent, and D. Olive, Physics Letters B 152, 88 (1985). [35] E. Fradkin, C. Nayak, and K. Schoutens, Nuclear Physics B 546, 711 (1999). [36] P. Di Francesco, P. Mathieu, and D. S´en´echal, Confor- mal field theory, Graduate texts in contemporary physics (Springer, New York, NY, 1997). [37] S. Sahoo, Z. Zhang, and J. C. Y. Teo, Phys. Rev. B 94, 165142 (2016). [38] S. Shatashvili and C. Vafa, Selecta Mathematica 1, 347 (1995). [39] P. Bonderson, Non-Abelian Anyons and Interferometry, Ph.D. thesis, California Institute of Technology (2007). [40] M. Gnaydin and S. V. Ketov, Nuclear Physics B 467, 215 (1996). [41] J. C. Baez, Bull. Amer. Math. Soc. 39, 145 (2002). [42] The Hamiltonian density is equivalent to the CFT energy momentum tensor on a cylinder: H0 = vTcyl/2π. [43] In addition to [42], J A differs in normalization from the [44] Note that (cid:104)γa(x)γb(0)(cid:105) = δab/2πix and(cid:80)∗C 2 WZW current defined in Ref. [36], which is 2πJ A. abcd = 168. [45] J. T. Chalker and P. D. Coddington, Journal of Physics C: Solid State Physics 21, 2665 (1988). [46] M. Levin, B. I. Halperin, and B. Rosenow, Phys. Rev. Lett. 99, 236806 (2007). [47] S.-S. Lee, S. Ryu, C. Nayak, and M. P. A. Fisher, Phys. Rev. Lett. 99, 236807 (2007). [48] F. Amet, C. T. Ke, I. V. Borzenets, J. Wang, K. Watan- abe, T. Taniguchi, R. S. Deacon, M. Yamamoto, Y. Bomze, S. Tarucha, and G. Finkelstein, Science 352, 966 (2016). [49] G.-H. Lee, K.-F. Huang, D. K. Efetov, D. S. Wei, S. Hart, T. Taniguchi, K. Watanabe, A. Yacoby, and P. Kim, Nature Physics 13, 693698 (2017). [50] C. L. Kane and M. P. A. Fisher, Phys. Rev. B 55, 15832 (1997). [51] M. Banerjee, M. Heiblum, A. Rosenblatt, Y. Oreg, D. E. Feldman, A. Stern, and V. Umansky, Nature 545, 75 (2016). [52] A. Stern and B. I. Halperin, Phys. Rev. Lett. 96, 016802 (2006). [53] P. Bonderson, A. Kitaev, and K. Shtengel, Phys. Rev. Lett. 96, 016803 (2006). [54] P. Bonderson, K. Shtengel, and J. K. Slingerland, Phys. Rev. Lett. 97, 016401 (2006). [55] D. E. Feldman and A. Kitaev, Phys. Rev. Lett. 97, 186803 (2006). [56] K. T. Law, Phys. Rev. B 77, 205310 (2008). [57] M. Buttiker, Phys. Rev. B 38, 9375 (1988). [58] C. L. Kane, Phys. Rev. Lett. 90, 226802 (2003). 

1605.06584

1

1605

Amplitude dependence of image quality in atomically-resolved bimodal atomic microscopy

In bimodal FM-AFM, two flexural modes are excited simultaneously. The total vertical oscillation deflection range of the tip is the sum of the peak-to-peak amplitudes of both flexural modes (sum amplitude). We show atomically resolved images of KBr(100) in ambient conditions in bimodal AFM that display a strong correlation between image quality and sum amplitude. When the sum amplitude becomes larger than about 200 pm, the signal-to-noise ratio (SNR) is drastically decreased. We propose this is caused by the temporary presence of one or more water layers in the tip-sample gap. These water layers screen the short range interaction and must be displaced with each oscillation cycle. Further decreasing the sum amplitude, however, causes a decrease in SNR. Therefore, the highest SNR in ambient conditions is achieved when the sum amplitude is slightly less than the thickness of the primary hydration layer.

cond-mat.mes-hall

cond-mat

Amplitude dependence of image quality in atomically-resolved bimodal atomic microscopy Hiroaki Ooe,1,2 a) Dominik Kirpal,1 Daniel S. Wastl,1 Alfred J. Weymouth,1 Toyoko Arai,2 and Franz J. Giessibl1. 1Institute of Experimental and Applied Physics, University of Regensburg D-93053 Regensburg, Germany 2Natural Science and Technology, Kanazawa University, Kanazawa, 920-1192 Ishikawa, Japan a electronic mail: [email protected] In bimodal FM-AFM, two flexural modes are excited simultaneously. The total vertical oscillation deflection range of the tip is the sum of the peak-to-peak amplitudes of both flexural modes (sum amplitude). We show atomically resolved images of KBr(100) in ambient conditions in bimodal AFM that display a strong correlation between image quality and sum amplitude. When the sum amplitude becomes larger than about 200 pm, the signal-to-noise ratio (SNR) is drastically decreased. We propose this is caused by the temporary presence of one or more water layers in the tip-sample gap. These water layers screen the short range interaction and must be displaced with each oscillation cycle. Further decreasing the sum amplitude, however, causes a decrease in SNR. Therefore, the highest SNR in ambient conditions is achieved when the sum amplitude is slightly less than the thickness of the primary hydration layer. 1 Frequency modulation atomic force microscopy (FM-AFM)1 is a powerful tool for investigating atomic-scale phenomena. The interaction between a tip at the end of an oscillating cantilever and a sample is measured via the frequency shift of the oscillation. If the oscillation amplitude is much larger than the decay length of the short range interaction, the tip spends little time within the short range interaction region, leading to a small contribution of the short range interaction to the detectable signal. It has been shown that in vacuum FM-AFM, the maximum signal-to-noise ratio (SNR) is achieved with an oscillation amplitude slightly larger than this decay length.2 When investigating short range forces that decay at lengths comparable to interatomic distances, the highest SNR is achieved with amplitudes in the range from several tens to a few hundred picometers (small amplitudes). Soft cantilevers that have a spring constant of k < 100 N/m (which is typical for commercial silicon cantilevers) require large amplitudes to prevent the tip from crashing into the surface at close distance (so- called "jump-to-contact").3 For this reason, atomic resolution measurements with soft cantilevers require the use of large amplitudes from one nanometer to tens of nanometers.4-6 One way to achieve controllable small amplitudes with soft cantilevers is to use a higher flexural mode which provides a much higher effective stiffness than the fundamental mode.7 Theoretically, the effective stiffness of the second flexural mode is about 40 times higher than in the first flexural mode and the resonance frequency is about 6.2 times higher.8 This can be implemented with bimodal AFM,9,10 in which the first flexural mode is excited at a large amplitude and the second flexural mode at a small amplitude to detect short range interactions. Bimodal AFM has been very successful in ambient and vacuum environments.10-12 In ambient environments or liquids, the cantilever must oscillate through the liquid or through water condensation layers.12-15 It was shown that the small oscillation of the second flexural mode could be used to increase sensitivity to materials properties.16,17 Several groups have applied this technique to biological samples, including antibodies12 and proteines.10 Schwenk and coworkers used bimodal AFM to increase MFM contrast of magnetic samples.20,21 Kawai and coworkers explicitly demonstrated the advantage of a higher flexural mode oscillating at smaller amplitudes (amplitudes less than 100 pm) with a standard Si cantilever on a KBr(100) surface in UHV.22 Moreno and coworkers used this to achieve intramolecular resolution in UHV conditions at low temperature.11 More recently, Santos and coworkers have started to consider the advantages of small oscillations in both flexural modes.17 It is expected for atomic resolution that the ideal bimodal measurement should be acquired with small amplitudes in both the first and second flexural modes.17 This requires the use of a much stiffer sensor. In this Letter, we present data acquired with a qPlus sensor with a spring constant of k = 1800 N/m.24 This high stiffness allows oscillation amplitudes of the first flexural mode smaller than one angstrom.14-15 We performed measurements in ambient condition on KBr(100). The amplitude of the first flexural mode, A1 and of the second flexural mode, A2, were independently set. These amplitudes were calibrated with a thermal spectrum and the ratio of the deflection sensitivity of two flexural modes.8,24-26 The frequency shifts of the first flexural mode, ∆f1, and of the second flexural mode, ∆f2, were recorded in quasi-constant 2 height mode using low gain integral feedback to compensate for thermal drift. We used a sensor with a resonance frequency of the first flexural mode of f1 = 32596.7 Hz and quality factor Q1 = 2906. The second flexural mode had a resonance frequency of f2 = 194858.2 Hz and a quality factor of Q2 = 1848. In ambient conditions, water condenses on all surfaces. Near the surface, it forms ordered hydration layers with a thickness of ~ 200 - 310 pm.13-15,29-33 In previous work, the ideal amplitude of oscillation was determined for single-mode FM-AFM measurements in ambient conditions.13-15 On the KBr(100) surface, the highest SNR was observed with an amplitude of A ~ 75pm.14 With smaller amplitudes, the signal becomes noisier due to instrumental noise.1,26,27 With larger amplitudes, the SNR suffers for two reasons: The average tip-sample distance becomes larger, reducing the signal, and water molecules come between the tip and sample. The tip then needs to penetrate the hydration layer during each oscillation cycle and the water molecules screen the short range interaction.14 Because of these effects, the SNR is enhanced when the peak-to-peak amplitude is slightly smaller than the thickness of a single hydration layer.15 Figure 1 shows single-mode FM-AFM and bimodal FM-AFM measurements. The oscillation models of the first and second flexural modes are shown in Figure 1 (a) and (b). First we collected single-mode images, exciting either the first or the second flexural mode. Figure 1(c) is a ∆f1 image taken with only the first flexural mode excited at A1 = 75 pm and Figure 1(d) is a ∆f2 image taken with only the second flexural mode excited at A2 = 75 pm. Atomic resolution can clearly be seen in both images. We then investigated if the two modes influence each other. To do this, we first acquired ∆f1 data with only the first flexural mode excited, then also excited the second mode. Figures 1 (e) and (f) show simultaneously acquired ∆f1 and ∆f2. The slow scan direction of these imaging was downward. Down to line A, only the first flexural mode was excited at A1 = 75 pm, and atomic resolution can clearly be seen in ∆f1. From line A down, the second flexural mode is also excited at A2 = 75 pm and the ∆f2 controller was turned on from line B. With both modes excited, the ∆f1 image becomes much weaker. Next we acquired data with both first and second flexural modes excited at the same amplitude. Figure 2 shows images of ∆f1 and ∆f2 with both flexural modes excited at amplitudes of 75 pm, 53 pm and 40 pm. When A1 = A2 = 75 pm, the (a) ∆f1 image and (b) ∆f2 image show faint atomic contrast similar to that in Figure 1 (e). The images improve when the amplitudes are decreased, as can be seen in Figure 2 (c) and (d), for which A1 = A2 = 53 pm. Very clear images are obtained when A1 = A2 = 40 pm, shown in Figure 2 (e) and (f). Similar to previous findings in vacuum,2 we find an optimal SNR for amplitudes in the sub-Angstrom level. However, empirically we find a notable difference to the decrease of SNR when increasing the amplitude beyond its optimal value Aopt. In vacuum, SNR decreases quite shallow at a rate of approximately2 (Aopt/A)0.5. In ambient environments with a liquid adsorption layer, we find a much stronger decay of image quality with sum amplitude. The sum amplitude is the vertical range that is covered by the oscillating cantilever: zp-p= 2 (A1 + A2). We propose that using sum amplitudes greater than half the thickness of the first hydration layer (approx. 200 pm) allows water molecules to penetrate the gap between 3 the tip apex and the sample, reducing the image quality. We then varied A1 and A2, keeping zp-p less than the thickness of a single hydration layer. In Figure 3(a) and (b), atomic contrast can be seen in both ∆f1 and ∆f2. A1 = 60 pm > A2 = 15 pm. Correspondingly, ∆f1 shows higher SNR. In Figure 3(c) and (d), A1 = A2 and the SNR of the two images are similar. Finally, in Figure (e) and (f), A1 = 15 pm < A2 = 60 pm, and the corresponding ∆f2 image has a higher SNR. These results show that SNR is higher with larger amplitudes. Instrumental noise decreases with increasing amplitude. Therefore, for each mode, larger amplitudes correspond to lower noise, as expected when considering the noise contributions in FM-AFM.1,27,28 In vacuum, the optimal amplitude is given by the balance of a more precise frequency measurement for larger amplitudes at the cost of a smaller frequency shift signal for larger amplitudes, resulting in an optimal amplitude that is approximately given by the decay length of the short-range interaction.2 In ambient conditions, the noise in frequency measurement also decreases for larger amplitudes, but the frequency shift signal induced by short-range interactions drops rapidly once the sum amplitude is large enough to admit water molecules in the tip-sample gap. The result is that the ideal amplitudes for bimodal FM-AFM follow the same pattern as for single-mode FM-AFM measurements. In ambient conditions, the sum amplitude must be smaller than the thickness of a hydration layer to ensure that the tip does not leave and re-penetrate a hydration layer with each cycle. At the same time, the amplitude has to be as large as possible to reduce the noise. The resolution of each mode can be increased by increasing its amplitude up to the ideal sum amplitude. In this study, we investigated the effect of the amplitude of the first and second flexural modes on the image quality in bimodal FM-AFM with small amplitudes in ambient conditions. Two orthogonal flexural modes can have a strong influence on each other. This is due to the hydration layer of sample surface. We showed that for this system, maximizing the SNR for both ∆f1 and ∆f2 results in the requirement that A1 = A2. Our results supporting that conventional bimodal AFM might also benefit from stiffer cantilevers that enable a smaller fundamental amplitude. Acknowledgements Funding was provided by Deutsche Forschungsgemeinschaft under GRK 1570 and by "Strategic Young Researcher Overseas Visits Program for Accelerating Brain Circulation" from the Japan Society for the Promotion of Science. The authors gratefully acknowledge the support for this study provided by Kanazawa University SAKIGAKE Project. References [1] T. R. Albrecht, P. Grütter, D. Horne, and D. Rugar, J. Appl. Phys. 69, 668 (1991). [2] F. J. Giessibl, H. Bielefeldt, S. Hembacher, and J. Mannhart, Appl. Surf. Sci. 140, 352 (1999). [3] F. J. Giessibl, Phys. Rev. B 56, 16010 (1997), 4 [4] F. J. Giessibl, Science 267, 68 (1995). [5] S. Kitamura, and M. Iwatsuki, Jpn. J. Appl. Phys. 34, L145 (1995). [6] K. Fukui, H. Onishi, and Y. Iwasawa, Phys. Rev. Lett. 79, 4202 (1997). [7] S. Kawai, S. Kitamura, D. Kobayashi, S. Meguro, and H. Kawakatsu, Appl. Phys. Lett., 86, 193107 (2005). [8] S. Rast, C. Wattinger, U. Gysin, and E. Meyer, Rev. Sci. Instrum. 71, 2772 (2000). [9] T. R. Rodriguez, and R. Garcia, Appl. Phys. Lett. 84, 449 (2004). [10] R. Garcia, E. T. Herruzo, Nature Nanotechnology 7, 217 (2012). [11] C. Moreno, O. Stetsovych, T. K. Shimizu, and O. Custance, Nano Lett. 15, 2257 (2015). [12] N. F. Martinez, J. R. Lozano, E. T. Herruzo, F. Garcia, C. Richter, T. Sulzbach, and R. Garcia, Nanotechnology, 19, 384011 (2008). [13] D. S. Wastl, M. Judmann, A. J. Weymouth, and F. J. Giessibl, ACS Nano, 9, 3858 (2015). [14] D. S. Wastl, A. J. Weymouth, and F. J. Giessibl, Phys. Rev. B 87, 245415 (2013). [15] D. S. Wastl, A. J. Weymouth, and F. J. Giessibl ACS Nano, 8, 5233 (2014) [16] N. F. Martinez, S. Patil, J. R. Lozano, R. Garcia, Appl. Phys. Lett. 89, 153115 (2006). [17] S. Santos, Appl. Phys. Lett. 103, 231603 (2013). [18] S. D. Solares, and G. Chawla, J. Appl. Phys. 108, 054901 (2010). [19] G. Chawla, and S. D. Solares, Appl. Phys. Lett. 99, 074103 (2011). [20] J. Schwenk, M. Marioni, S. Romer, N. R. Joshi, and H. J. Hug, Appl. Phys. Lett. 104 112412 (2014). [21] J. Schwenk, X Zhao, M. Baconi, M. A. Manioni, S. Romer, and H. J. Hug, Appl. Phys. Lett. 107, 132407 (2015) [22] S. Kawai, T. Glatzel, S. Koch, B. Such, A. Baratoff, and E. Meyer, Phys. Rev. Lett. 103, 220801 (2009). [23] S. Santos, V. Barcons, H. K. Christenson, D. J. Billingsley, W. A. Bonass, J. Font, and N. H. Thomson, Appl. Phys. Lett. 103, 063702 (2013). [24] F. J. Giessibl, Appl. Phys. Lett. 76, 1470 (2000). [25] J. Welker, F. F. Elsner, and F. J. Giessibl, Appl. Phys. Lett. 99, 084102 (2011). [26] F. J. Giessibl, Rev. Mod. Phys. 75, 949 (2003). [27] K. Kobayashi, H. Yamada, and K. Matsushige, Rev. Sci. Instrum. 82, 0337025 (2011). [28] M. Luna, F. Rieutord, N. A. Melman, Q. Dai, and M. Salmeron, J. Phys. Chem. A 102, 6793 (1998). [29] S. Jeffery, P. M. Hoffmann, J. B. Pethica, C. Ramanujan, H. Ozgur Ozer, and A. Oral, Phys. Rev. B 70, 054114 (2004). [30] T. Fukuma, M. J. Higgins, and S. P. Jarvis, Biophys. J. 92, 3603 (2007). [31] K. Kimura, S. Ido, N. Oyabu, K. Kobayashi, and Y. Hirata, J. Chem. Phys. 132, 194705 (2010). [32] J. N. Israelachvili and R. M. Pashley, Nature (London) 306, 249 (1983). [33] J. I. Kilpatrick, S. Loh, and S. P. Jarvis, J. Am. Chem. Soc. 135, 2628 (2013). 5  FIG. 1 Schematics of (a) first and (b) second flexural mode. (c) ∆f1 image with only the first flexural mode excited at A1 = 75 pm. (d) ∆f2 with only the second flexural mode excited at A2 = 75 pm. (e) ∆f1 and (f) ∆f2 images simultaneously acquired. Up to line A, only the first mode was excited at A1 = 75 pm. Past line B, both modes were excited at A1 = A2 = 75 pm. For clarity, all images were line-flattened. The raw data with scale of frequency shift are shown in Figure 1 of supplementary material. 6 FIG. 2 Bimodal FM-AFM images taken in which A1 = A2. (a and b) A1 = A2 = 75 pm (c and d) A1 = A2 = 53 pm (e and f) A1 = A2 = 40 pm. Images are line-flattened for clarity. The raw data with scale of frequency shift are shown in Figure 2 of supplementary material. 7  FIG.3 A survey of images taken with different A1 and A2 values in A1 + A2 ~ 80 pm. (a) ∆f1 image with A1 = 60 pm, and (b) ∆f2 image with A2 = 15 pm. (c) ∆f1 image with A1 = 40 pm, and (d) ∆f2 image with A2 = 40 pm. (e) ∆f1 image with A1 = 15 pm, and (f) ∆f2 image with A2 = 60 pm. All images were line flattened for clarity. The raw data with scale of frequency shift are shown in Figure 3 of supplementary material. 8 For Supplementary Material. Figure 1 Raw data 9 Figure 2 Raw data 10 Figure 3 Raw data 11 

1108.1666

2

1108

An electrical probe for mechanical vibrations in suspended carbon nanotubes

The transport properties of a suspended carbon nanotube probed by means of a STM tip are investigated. A microscopic theory of the coupling between electrons and mechanical vibrations is developed. It predicts a position-dependent coupling constant, sizeable only in the region where the vibron is located. This fact has profound consequences on the transport properties, which allow to extract information on the location and size of the vibrating portions of the nanotube.

cond-mat.mes-hall

cond-mat

An electrical probe for mechanical vibrations in suspended carbon nanotubes N. Traverso Ziani1, G. Piovano1,2, F. Cavaliere1,2 and M. Sassetti1,2 1 Dipartimento di Fisica, Universit`a di Genova, Via Dodecaneso 33, 16146, Genova, Italy. 2 CNR-SPIN, Via Dodecaneso 33, 16146, Genova, Italy. (Dated: March 8, 2021) and The transport properties of a suspended carbon nanotube probed by means of a STM tip are investigated. A microscopic theory of the coupling between electrons and mechanical vibrations is developed. It predicts a position-dependent coupling constant, sizeable only in the region where the vibron is located. This fact has profound consequences on the transport properties, which allow to extract information on the location and size of the vibrating portions of the nanotube. PACS numbers: 85.85.+j, 73.63.Kv I. INTRODUCTION Carbon nanotubes (CNTs)1 are extremely versatile systems with metallic or semiconducting behavior de- pending on their wrapping orientation.2,3 Deposing them on an insulating substrate and tunnel-coupling it to bi- ased electrodes it is possible to create a single-electron transistor, in which the nanotube behaves as a quan- tum dot.4 -- 6 Alternately one can embed a quantum dot into the nanotube via geometrical defects or external gates, thus building a nanotube dot tunnel-coupled to contacts.7 Recent improvements in manipulation techniques have allowed to suspend nanotubes between two contacts. In this case nanotubes behave as mechanical resonators,8 with possible applications ranging from ultra-sensitive mass sensing to displacement sensors.9 Among the differ- ent mechanical vibrations10 the radial breathing mode is the highest in energy11 -- 13 followed by the twist and the stretching ones. The latter have received a lot of ex- perimental attention 14 -- 17 also in view of the peculiar features induced on transport, such as negative differen- tial conductance15 or Franck-Condon blockade.17 Bend- ing modes have usually energies lower than the experi- mental temperature requiring external AC drivings.18,19 Transport experiments have been employed to analyze the structure of nanotubes exploiting a scanning tun- neling microscope (STM) tip.20 Effects such as spin- charge separation were observed studying the differential conductance as a function of the tip position.21 Super- conducting probes have been used to extract the non- equilibrium electron energy distribution function.22 Also, the effects of chemical or magnetic impurities adsorbed along the nanotube were considered.23 -- 28 Scanning tunnel microscopy experiments have also been performed on suspended nanotubes. In particular, it has been shown how electrons injected from a tunnel micro- scope tip can excite, detect and control a specific vibra- tional mode.11 -- 13 Owing to their small waist (of the order of some nm) nanotubes behave as a one-dimensional interacting elec- tronic system.2,3 Typically, correlated quantum systems are studied by means of numerical techniques.29 -- 32 How- ever, due to their inherently one-dimensional nature, car- bon nanotubes are described in terms of a Luttinger model.33,34 In this context, transport from a tunneling tip to a static nanotube has been recently considered.28,35 -- 39 The coupling between the electrons and vibrational modes has been extensively studied in literature.40 -- 47 In most cases the simple Anderson-Holstein model has been employed,48,49 in which the vibron couples only to the total charge neglecting the spatial modulation of the charge density. The Anderson-Holstein interaction yields position-independent Franck-Condon factors50,51 with visible effects in transport properties.52 -- 57 Recently, a microscopic theory involving the coupling with spatial fluctuations of the nanotube electronic den- sity has also been developed58 in order to explain anoma- lous transport behaviors.58 In this paper we investigate the possibility of creating an electrical probe for the stretching vibrational modes of a suspended carbon nanotube, by means of a scanning tunnel microscope tip. Building on the theory outlined in Ref. 58, we describe the coupling between vibrons and total charge as well as the spatial charge density modu- lations. This coupling gives rise to a position-dependent, electron-vibron coupling which strongly affects the trans- port properties. Position-dependent tunneling rates and conductance arise. This allows to obtain precise infor- mations about the vibrational mode of the nanotube. Effects are visible in metallic nanotubes and are more pronounced in semiconducting ones. The paper is structured as follows. In Sec. II A a Lut- tinger liquid model for a carbon nanotube with open boundary conditions is introduced. In Secs. II B-II C the lattice vibrations, the electron-vibron coupling and its diagonalization are discussed. Section II D is devoted to the transport properties. Our results are illustrated and commented in Sec. III. Conclusions are drawn in Sec. IV. 1 1 0 2 t c O 7 ] l l a h - s e m . t a m - d n o c [ 2 v 6 6 6 1 . 8 0 1 1 : v i X r a 2 II. MODEL AND METHODS A. Modeling a carbon nanotube quantum dot The electronic properties of the CNT are characterized by the wrapping vector wn,m = na+1 + ma−1, where a±1 represent the basis vectors of the graphene lattice.2 Due to the wrapped nature of the system, the energy spec- trum is composed of subbands corresponding to trans- verse excitations along the waist of the tubule. In a typ- ical experiment only the lowest-lying subband is occu- pied,2,3 giving a one dimensional character to the CNT. In this regime both n-doped semiconducting CNTs (away from the band gap) and metallic CNTs can be described in the low energy sector as Luttinger liquids with four branches,59 -- 63 labeled by α = ±1, stemming from the two Dirac valleys of the graphene, and by s = ±1, denot- ing the z component of the electron spin (units /2). FIG. 1: (Color online) Schematic setup of a CNT, suspended between the two substrates S1 and S2 at positions x1 and x2. A quantum dot with ends at x = 0 and x = L is embedded in the CNT. The system under investigation is schematically depicted in Fig. 1, it consists of a CNT, suspended between two substrates at x1 and x2 and free to vibrate. Embedded in the CNT there is a quantum dot of length L with ends at x = 0 and x = L. We assume x1 ≤ 0 and x2 ≥ L in order to mimic a quantum dot inside the CNT, due to geometrical defect or external gates, or to treat the CNT itself as the quantum dot. In the following, we focus on the description of the quantum dot with open boundary conditions. The bosonized hamiltonian is H =(cid:80) Hj with j Hj = 1 2 Ej N 2 j + † ωj(q)b j(q)bj(q) , ( = 1) (1) where j ∈ {ρ+, ρ−, σ+, σ−} are the four linear combina- tions of states in the α, s branches that diagonalize the Coulomb interaction. Here, Nj represent the zero modes counting the excess electrons in the j sector (cid:88) q (cid:88) (cid:88) α,s (cid:88) (cid:88) α,s Nρ+ = Nα,s Nσ+ = s Nα,s ; ; Nρ− = Nσ− = α Nα,s ; αs Nα,s . α,s α,s The bosonic operators bj(q) trigger collective excitations of the electron system with momentum q = πn/L with FIG. 2: (Color online) Fermi velocity vF of a semiconducting CNT of length L = 400 with N0 electrons and different CNT waist lengths wn,m: blue (solid) 1 nm - such as for (3,2), purple (dashed) 2 nm - such as for (5,4), yellow (dotted) 3 nm - such as for (8,6), green (dash-dotted) 4 nm - such as for (11,7). n ∈ N∗. They are connected to the α, s modes by a Bogoljubov transformation.59,60 Note that the mode j = ρ+ represents the total charge of the system. The collective modes propagates with velocities vj = vF/gj with gρ+ = g and gj = 1 ∀j (cid:54)= ρ+. Here g parame- terizes the strength of electron scattering, with g < 1 for repulsive interactions. Note that only the velocity of the total charge mode is renormalized. The corresponding energies are ωj(q) = vjq and Ej = πvj/4gjL. The value of the Fermi velocity vF depends on the prop- erties of the CNT. For a metallic CNT (n − m/3 ∈ N ) the dispersion relation is linear with vF = v0 = 8·105 m/s around the degenerate point ¯k. In the following, we will assume an effective n-doping with a shift of EF towards higher energy values and new Fermi points k(±) F = ¯k± qF with qF = EF/vF. In a semiconducting CNT the conduction and valence bands are separated at momentum ¯k(cid:48) by the direct gap2 ∆ = 4πv0 3wn,m , (2) √ with wn,m = a n2 + nm + m2 the waist length of the CNT. We will consider n-doped semiconducting nan- otubes with EF > ∆/2, having chosen the energy ref- erence to lie in the middle of the band gap. Doping gives rise to two new Fermi points k(±) F = ¯k(cid:48) ± qF where qF = 2m∗√ 0 the effective mass.2 Around the new Fermi points, the Fermi veloc- ity is EF − ∆ with m∗ = ∆/2v2 vF = 3wn,mN0 8L , (3) with N0 the total number of excess electrons in the quan- tum dot. The dependence of vF as a function of N0 is shown in Fig. 2. One observes a lower velocity with re- spect to the metallic case. The electron field operator Ψs(r) ≡ Ψs(x, y) has to sat- isfy open-boundaries conditions Ψs(0, y) = Ψs(L, y) = 0. It can be written in the bosonized form Ψs(r) = fr,α(r)eirqFx ψ+1,rα,s(rx) (4) (cid:88) (cid:88) r=±1 α=±1 in terms of the right-movers field operators ψ+1,α,s(x) = where φj(x) = e−iθα,s ei πx 4L ( Nρ+ +α Nρ− +s Nσ+ +αs Nσ− ) · ηα,s√ 2πa 2 [ φρ+ (x)+α φρ− (x)+s φσ+ (x)+αs φσ− (x)] , i e (cid:114) π (cid:88) qL (cid:26) 1√ (cid:104)bj(q) − b gj cos (qx) (cid:104)bj(q) + b (cid:105)(cid:111) . † j(q) (cid:105) † j(q) q √ + i gj sin(qx) (5) (6) Here ηα,s are Majorana fermions, [θα,s, Nα,s] = i, and a is the length cutoff. The functions fr,α(r) in Eq. (4) consist of a superposition of wavefunctions for pz orbitals, peaked around the positions of atoms in the CNT and oscillating with a typical wave vector K0 ∝ a−1 where a ≈ 2.5· 10−10 m.59,60 Their specific form depends on the type of nanotube under consideration and will be not discussed here. (cid:90) (cid:90) x> x< (cid:104) 3 The coupling between electrons and vibrations can be microscopically derived starting from the tight-binding theory of a distorted CNT lattice.42,64 For the typical experimental situations one has L (cid:38) 100 nm then a con- tinuum elastic model is appropriate with40 (9) dr ρ(r)∂x up(r) , Hd−v = c Here40 c ≈ 30 eV and ρ(r) =(cid:80) s Ψ† s(r) Ψs(r) is the elec- tronic density operator, with Ψs(r) given by Eq. (4). It consists of two components: a long wavelength part ρLW(r) and an oscillatory contribution ρSW(r) fluctu- 0 ∝ a. This latter compo- ating on a length scale K−1 nent, does not make sizeable contributions since q0 (cid:28) K0. Under the realistic assumption of strongly local- ized atomic orbitals2,3 with negligible overlapping, the electron-vibron coupling becomes Hd−v = c dx ρLW(x)∂x up(x) , (10) with x< = max{0, xv0}, x> = min{xv1, L} and ρLW = Nρ+ L + 1 2π φρ+(x) + x → −x ∂x , (11) written here directly in its bosonized form. (cid:105) B. Lattice vibrations and electron-vibron coupling C. Diagonalizing the electron-vibron coupling We consider the case of a vibrating portion of the CNT (the vibron), of size L, located between xv0 and xv1. As confirmed in a recent experiment,58 the vibrating part can be different from the CNT dot. Thus, we will formu- late the theory in this most general case. We focus the description on the stretching mode. The p-th mode has energy ω0 = pπvs/L with vs ≈ 2.4 · 104 m/s the veloc- ity of the stretching modes which is approximately non- dispersive. In most experiments the fundamental mode with p = 1 is observed.15,17 For these reasons, we will concentrate to the case of small p ≤ 3. The p-th mode is described as a harmonic oscillator 2 P0 2M M ω2 0 X 2 0 , Hv = + (7) where M = 2πwn,mLρ0 is the vibron mass with ρ0 ≈ 6.7 · 10−7 Kg/m2 the graphene density and X0 is the amplitude operator of the strain field 2 (cid:20) (cid:21) √ 2 X0 sin up(r) = π(x − xv0) L p . (8) The latter represents a standing wave with momentum ±q0 with q0 = pπ/L. ρ+ + H (pl) h = H (0) and H (pl) ρ+ =(cid:80) The relevant terms of the electron-vibron coupling are ρ+ + Hv + Hd−v with H (0) /2 q ωρ+(q)bρ+(q)bρ+(q). Introducing Bµ = µ we have ibρ+(πµ/L) and(cid:112)2ωρ+(πµ/L) Xµ = Bµ + B† (cid:32) P 2 ρ+ = Eρ+ N 2 ρ+ (cid:33) (cid:88) M ω2 0 h = N 2 ρ+ + + ω2 µ X 2 µ 2 µ 2 1 Eρ+ 2 √ P 2 0 2M + √ + M C0 X0 Nρ+ + M X0 2 0 + µ≥1 Cµ Xµ , X 2 (cid:88) µ≥1 with [ Xµ, Pν] = iδµ,ν, ωµ = µω1/g, ω1 = πvF/L. terms of these new variables the density operator is (cid:114) 2πvF (cid:88) L3 µ≥1 (cid:16) πµx (cid:17) Xµ . L µ cos (13) ρLW(x) = Nρ+ L + The last term in Eq. (12) describes a central harmonic os- cillator (vibron) linearly coupled to a infinity of harmonic oscillators (plasmon modes of the dot). Note that, for reasonable experimental parameters and considering the lowest stretching modes, one always has ωµ > ω0 both for metallic and semiconducting CNTs. Additionally, the vibron is also coupled to the total average charge Nρ+ on (12) In 4 the quantum dot in analogy to the Anderson-Holstein model. The coupling coefficients are √ C0 = 2λmω3/2 0 J0 where (κ ≥ 0) (cid:90) x> Jκ = 1 L and dx cos x< λm = √ ω1Jµ (14) , (15) (cid:105) 0 ; Cµ≥1 = 2λmω3/2 L cos (cid:104) κπx (cid:104) pπ (cid:105) L (x − xv0) (cid:112)ρ0πwn,mvs c . vs Taking as a reference2,3 a waist length of about 2 nm, one has λm ≈ 2. Note that a recent experiment65 reports a larger c which leads to a larger λm. One finds Jκ = J (0) κ θ(x1−L−0+) , (17) κ θ(−0+−xv0)+J (+) κ +J (−) with θ(x) the Heavyside step function and κδ2 {sin (πκξ0) − (−1)p sin [πκ(ξ0 + δ)]} π(p2 − κ2δ2) pδ sin (πpξ0/δ) − κδ2 sin (πκξ0) π(p2 − κ2δ2) , , J (0) κ = J (−) κ = J (+) κ = (−1)pκδ2 sin [πκ(ξ0 + δ)] − pδ(−1)κ sin [πpδ(1 − ξ0)/δ] π(p2 − κ2δ2) π(p2 − κ2δ2) where δ = L/L and ξ0 = xv0/L. X0 Nρ+ with (cid:96)−1 Let us now comment the general features of the cou- pling depending on the relative size and position of dot and vibron. When the vibron is much larger than the dot (δ (cid:29) 1) and the latter is embedded into it one has Jκ ≈ δκ,0, namely in the large vibron limit the coupling to the charge density fluctuations vanishes and only the con- ventional Anderson-Holstein coupling survives. In this limit, the electron-vibron coupling reduces to the stan- dard form66 ω0λm(cid:96)−1 M ω0, and the coupling constant is λm. The most interesting case occurs when the vibron is smaller than the dot and embedded into it. In this regime, one finds J0 ≡ 0, while Jκ≥1 (cid:54)= 0. This fact signals a radical departure from the Anderson-Holstein model: the coupling between electrons and vibrons oc- curs only via the spatial fluctuations of the electron den- sity. We now turn to the diagonalization of the electron-vibron coupling. The terms linear in Xµ in Eq. (12) can be exactly diagonalized,67 leading to 0 = √ 0 (cid:33) ¯X 2 µ 2 (cid:32) ¯P 2  Nρ+ , + Ω2 µ µ 2 ¯Xν (cid:88) µ≥0 k0ν Eρ+ N 2 ρ+ 1 2 √ h = + M C0 + (cid:88) ν≥0 z2 = ω2 0 + , with √ with ην = (16) with (cid:16) Xµ, Pµ (cid:17) = (cid:88) ν≥0 (cid:16) ¯Xν, ¯Pν (cid:17) . kµν (19) In order to diagonalize the term ∝ Nρ+ a Lang-Firsov canonical transformation is used (cid:80) U = e −i Nρ+ √ ν≥0 ην ¯Pν , (20) which finally casts the hamiltonian into the diagonal form M C0k0ν/(Ω2 ν M ). This leads to a shift ¯Xν → ¯Xν − ην Nρ+ , (21) (cid:1) N 2 + ρ+ (cid:32) ¯P 2 µ 2 (cid:88) µ≥0 (cid:33) + Ω2 µ ¯X 2 µ 2 , (22) h = 1 2 (cid:0)Eρ+ − ∆Eρ+ (cid:80) with ∆Eρ+ = C 2 0 ν≥0(k2 0ν/Ω2 ν). The energies Ωµ of the new eigenmodes are the roots of the secular equation (cid:88) ν≥1 C 2 ν z2 − ν2ω2 1 ; (23) kµν = k0ν = Cµ ν − µ2ω2 (cid:88) Ω2 1 + 1 (with µ ≥ 1) , k0ν (24) C 2 µ ν − µ2ω2 1)2 (Ω2 µ≥1 . (25) −1/2 As can be clearly seen, the modes with µ ≥ 1 have FIG. 3: Energy of the lowest eigenmode. (a) Plot of Ω0/ω0 as a function of α; (b) Plot of Ω1/ω1 as a function of α. In all figures, δ = 1, xv0 = 0, p = 1, g = 1 and λm = 2. energies Ωµ (cid:38) µω1 and represent blue-shifted dressed plasmons. The lowest-lying solution, on the other hand, has Ω0 < ω0 and represents a dressed vibronic mode red-shifted by the electron-vibron interaction. By inspecting Eq. (23) one always obtains a real solution (18) 1 > (cid:80) 0ω2 µ≥1 C 2 for Ωµ with µ ≥ 1. On the other hand, the existence of a real solution for Ω0 requires ω2 µ/µ2. When this condition is not fulfilled, the Wentzel-Bardeen instability occurs.41 In our calculations we have always checked that for realistic parameters the system does not exhibit this instability. The energy of the dressed vibronic mode is very sensitive to the ratio α = vF/vs between the Fermi and the sound velocity. While for α = 32, corresponding to the case of a metallic CNT, one has Ω0 ≈ ω0, for lower values of α, typical of a semiconducting CNT, a suppression of Ω0 occurs, see Fig. 3(a). Note that the dressed plasmons are almost insensitive to the ratio α (cf. Fig. 3(b)). The transformations in Eq. (19) and (20) affects the elec- tronic field operator of Eq. (5). Up to an irrelevant phase constant, the field φρ+(x) is (cid:88) (cid:88) µ≥0 ν≥1 φρ+(x) = αµ(x) ¯Xµ + βµ(x) ¯Pµ, (26) where αµ(x) = βµ(x) = √ √ 2ω1 M ηµ + (cid:17) (cid:16) πνx (cid:88) L kνµ ν kνµ sin (cid:114) 2 ω1 ν≥1 (cid:16) πνx (cid:17) L (27) . (28) cos 5 to the contacts through the CNT. Additionally, a back gate is capacitively coupled to tune the effective charge on the CNT-dot. From now on, we will focus on the most interesting regime, namely that of a quantum dot along all the CNT with a vibron imbedded into it. The hamiltonian for the vibrating CNT is then (cid:16) Nρ+ − Ng (cid:17)2 + (cid:88) µ≥0 HCNT = 1 2 Eρ+ ¯P 2 µ 2 + Ω2 µ 2 ¯X 2 µ+ (cid:88) j(cid:54)=ρ+ Hj , where Ng represents the charge induced by the back gate voltage Vg. The dot is laterally coupled to the two Fermi contacts, via the tunneling hamiltonian58 HT,L = t0 ψ+1,α,s(xj)cj,s(q) + h.c. , (29) (cid:88) (cid:88) j=1,2 α,s,q where t0 is the tunneling amplitude, cj,s(q) are the op- erators for an electron with momentum q and spin s in the non-interacting lead j and x1 = 0, x2 = L are the position of the tunneling contacts. The STM tip is modeled as a semi-infinite Fermi con- tact, placed above the CNT at a position x along it. The tunnel coupling is expressed in terms of the Fermi field operator for the forward modes of the tip28,35 ψs,F(z) (z is the coordinate along the tip with z = 0 at the vertex) (cid:34)(cid:90) L (cid:88) (cid:35) HT,T= dy τ0(y − x) † ψ +1,α,s(y) ψs,F(0+) + h.c. . 0 α,s D. Modeling transport FIG. 4: Schematic setup of a STM transport experiment per- formed on a suspended CNT. The quantum dot is the nan- otube itself, biased with respect to the tip and two lateral contacts. Electrons can flow through the system via the STM tip at position (x) and at the contacts (x1,2). A back gate allows to tune the effective charge on the dot. Our task is to model an STM tunneling tip on a CNT.11 -- 13 The setup is sketched in Fig. 4 and is com- posed of a STM tip and two lateral contacts tunnel- coupled to a suspended CNT. Tip and contacts are biased in such a way that, for V > 0, electrons flow from the tip The function τ0(x) = τ0ϕ(x) describes the geometry of the tip, with 0 ≤ ϕ(x) ≤ 1 peaked around x = 0, where ϕ(0) = 1. In the following we will assume a typical tip, with an effective width of a few atomic cells of the CNT. A voltage −V /2 is applied to the STM tip, while the lat- eral contacts are kept at the same voltage V /2. Voltage drops are assumed to occur symmetrically on the CNT. More general potential distributions do not affect the re- sults at a qualitative level. We will consider the sequential tunneling regime, treat- ing the tunnel couplings to the lowest perturbative order. Since we are interested into the low-energy transport regime (eV kBT ≈ Ω0) we disregard the dynamics of the modes µ ≥ 1. Furthermore, we will consider the relevant situation of a damped vibronic mode, with a thermal equilibrium distribution at temperature T . The eigenstates of the suspended CNT can be expressed by {Nα,s}(cid:105) specifying the distribution of excess elec- trons in the channel α with spin s. We will consider the resonance between the state of a closed shell with N0 = 4κ, (κ integer) electrons and zero excess charges, denoted as 0(cid:105) = 0, 0, 0, 0(cid:105), and N0 + 1 electrons in the state α,s. Note that no qualitative difference in our re- sults would occur, for a different value of N0. There are four states with N0 + 1, electrons, denoted by α, s(cid:105), each with one extra electron in the state α, s. They are all degenerate, with energy Eρ+(1 − Ng)2/2 + 3ω1/8. 6 We set up a master equation for the reduced density matrix, obtained tracing out the leads and vibron de- grees of freedom - thus neglecting coherences among vi- brational states.66,68,69 Upon the assumption of a STM tip width of some unit cells, coherence effects between states α, s(cid:105) and α(cid:48), s(cid:105) (α (cid:54)= α(cid:48)) are vanishing. Co- herence between different spin states is also absent in view of the absence of spin correlations in the contacts. Therefore, the master equation reduces to a standard rate equation for the occupation probability of the quantum α,s Pα,s(cid:105)(t). The steady-state current is then written, to lowest order, in terms of tunneling rates dot PN0 (t) = P0(cid:105)(t) ; PN0+1(t) = (cid:80) I(x) = e Γ(C) outΓ(T) in (x) − Γ(C) Γin(x) + Γout(x) in Γ(T) out(x) Γin/out(x) = Γ(T) in/out(x) + Γ(C) in/out , (30) (31) with and (cid:88) (cid:88) (cid:88) α,s Γ(T) in (x) = Γ(T) 0(cid:105)→α,s(cid:105)(x) ; Γ(T) out(x) = Γ(T) α,s(cid:105)→0(cid:105)(x) , Γ(C) in = Γ(j)0(cid:105)→α,s(cid:105) ; Γ(C) out = Γ(j)α,s(cid:105)→0(cid:105) , j α,s j (cid:88) of the Anderson-Holstein model,52,54 -- 56 appropriate for large vibrons. In that case, as discussed above, cou- pling to the density fluctuations would vanish leading to α0(x) = β0(x) = 0 and to a coupling simply given by λmax which is clearly independent of the tunneling posi- tion. III. RESULTS A. Local electron-vibron coupling Let us analyze in details the space dependence of the electron-vibron coupling λ(x) which determines the cur- rent behavior. The coupling λ(x) depends both on ge- ometrical parameters xv0 (the vibron origin), δ = L/L (vibron length) and on physical ones α = vF/vs, λm de- fined in Eq. (16) and the electronic interaction parameter g. We remind that the parameter α is affected by the metal- lic or semiconducting nature of the CNT. In a metallic CNT one finds α = 32. On the other hand, for a semicon- ducting CNT lower values of α are possible (see later). Figure 5 shows λ(x) for different vibron configurations the rate associated to the tip (T) and to the contacts (C). We quote here explicitly the expressions for the tunnel-in processes, the tunnel-out rates are similar. The lateral contact and tip rates are respectively Γ(j)0(cid:105)→α,s(cid:105) = Γ0 Bl(xj)f (∆E + lΩ0 + eV /2) ,(32) Γ(T) 0(cid:105)→α,s(cid:105)(x) = Γ(T) 0 Bl(x)f (∆E + lΩ0 − eV /2) .(33) (cid:88) (cid:88) l≥0 l≥0 0 = 2πν0τ02, ν0 is the leads Here, Γ0 = 2πν0t02, Γ(T) density of state and f (E) the Fermi function. The rates are then a superposition of Fermi functions at energies (cid:19) (cid:18) 1 2 ∆E = Eρ+ − Ng + 3 8 ω1 , (34) shifted by the energy lΩ0, representing the contribution of a transport channel exciting l vibron quanta. The weights Bl(x) considered in the regime kBT (cid:28) Ω0 are52 where Bl(x) = λ2l(x) l! e−λ2(x) , λ2(x) = 1 2Ω0 α2 0(x) + Ω0 2 β2 0 (x) (35) (36) FIG. 5: (Color online) Local electron-vibron coupling λ(x) as a function of the tip position x for: (a) δ = 1, xv0 = 0, p = 1; (b) as above but p = 2; (c) δ = 0.4, xv0 = 0.15L, p = 1; (d) as above but p = 2. In all panels red (solid) lines denote a metallic CNT with α = 32, blue (dashed) lines denote a semiconducting CNT with α = 5. Other parameters: λm = 2 and g = 1. The shaded plots on top of the panels depict the amplitude of the strain field up(x). The arrow in Panel (c) denotes the tip position for the conductance shown in Fig. 8. represents the local electron-vibron couplingstrength,58 see Eqns. (27,28). This is in sharp contrast to the case and CNT types. It can be seen that the electron-vibron coupling strength for a metallic CNT (solid red lines) is smaller than that for a semiconducting CNT. Indeed, for a semiconducting CNT, the velocities of the electronic and vibronic subsystems are closer, which implies a more favorable interplay between them. In the rest of the pa- per, we will choose α = 5 to model a semiconducting CNT and α = 32 for the metallic one. The amplitude of the electron-vibron coupling is maxi- mal in the region where the strain field is maximum. In- deed, λ(x) closely follows the amplitude of up(x), which is sketched on top of the panels of Fig. 5. For δ < 1, this implies a particularly sizeable λ(x) only in the re- gion where the vibron sits, see Figs. 5(c,d). Coupling to higher vibronic modes produces more oscillations, as can be seen in Figs. 5(b,d). It also makes the electron-vibron coupling strength weaker. Figure 6 shows the comparison 7 neling rates. Figure 7 shows the ratio λL/λR as a func- tion of δ for a vibron with origin at xv0 = 0.1L. At small values of δ the vibron is asymmetrically located near the left tunnel barrier. As a consequence, λL > λR. This mechanism is at the origin of the systematic suppres- sion of conductance traces in a recent experiment.58 For increasing δ the situation evolves towards a more sym- metric setup and indeed for δ = 0.8, corresponding to a symmetric vibron with respect to the CNT, one recovers λL = λR. B. Transport properties Figure 8 shows the density plot of the differential conductance G = ∂I/∂V in the (V, Ng) plane for the situation depicted in Fig. 5(c), semiconducting case. The large white areas at small V are the Coulomb FIG. 6: (Color online) Local electron vibron coupling λ(x) as a function of x for δ = 1, xv0 = 0 and different vibron modes: red (solid) p = 1, green (dashed) p = 2, blue (dotted) p = 3. The thin lines are the maxima of λ(x). Other parameters: α = 5, λm = 2, g = 1. between the first three vibronic modes. The intensity of the electron-vibron coupling strength decreases with the √ increasing order of the vibronic mode. Denoting λp the maximum of λ(x) for the p mode, we find λp = λ1/ p. We now briefly comment on the value of the electron- FIG. 8: (Color online) Semiconducting CNT. Differential con- ductance G (units e2Γ(T) 0 /Ω0) as a function of Ng and V (units Ω0/e) for a vibron originating at xv0 = 0.15L with δ = 0.4, p = 1 and a tip at x = 0.35L. Other parameters: λm = 2, α = 5, g = 1, kBT = 0.15 Ω0 and A = Γ0/Γ(T) 0 = 100. blockade regions where transport is interdicted and the CNT is occupied by N0 or N0 + 1 electrons. Within the transport region, delimited by the two most intense conductance traces, a series of equally spaced lines are clearly visible. They correspond to the excitation of the vibronic mode at energy Ω0. We consider different tunneling amplitudes through the tip and the lateral contacts, introducing the asymmetry A = Γ0/Γ(T) . 0 FIG. 7: (Color online) Ratio of the coupling strengths λL/λR (see text) as a function of δ for a vibron with origin at xv0 = 0.1L and p = 1. The shaded plots schematically depict the amplitude of the strain field of the vibronic mode up(x). Other parameters: α = 5, λm = 2 and g = 1. We will concentrate the discussion on the regime kBT < Ω0, quoting for simplicity analytical expressions for A (cid:29) 1 only, realistic in an STM experiment. Exploiting the fact that λL,R (cid:28) 1 we assume Bl(xj) ≈ δl,0. In the linear regime (V → 0) the conductance is then vibron coupling at the position of the tunneling barriers λL ≡ λ(0) and λR ≡ λ(L), which govern the lateral tun- Glin ≈ cosh (37) (cid:104) β∆E 2βe2Γ(T) 2 − 1 2 ln(4) (cid:105) 0 B0(x) (cid:104) β∆E (cid:105) , 2 cosh 8 with ∆E in Eq. (34). The logarithmic factor ln(4) stems from the fourfold degeneracy of the state with N0 + 1 electrons. The amplitude of the linear conductance is modulated by the factor B0(x) = e−λ2(x) . (38) Therefore, the conductance is suppressed in the region where the electron-vibron coupling is large. In the nonlinear regime (V > kBT ) one finds (cid:88) (cid:104) Bl(x) (cid:105) . Gnonlin ≈ βe2Γ0 2 l≥0 cosh β ∆E+lΩ0−eV /2 2 (39) Equation (39) represents a fan of equally-spaced conduc- tance peak lines located at Eρ+(1 − 2Ng) + 3ω1/8Eρ+ + lΩ0 − eV /2 = 0, thus with negative slope in the (V, Ng) plane, see Fig. 8. They originate by the tunneling from the STM tip to the CNT, triggering vibronic excitations. We note that, because of the smallness of λL,R, the trig- gering process due to the tunneling on the contacts bar- riers is strongly suppressed (since Bl(xj) ≈ δl,0) with a corresponding absence of conductance lines with positive slope. Each of the above peaks is weighted by Bl(x) which in turn conveys informations on λ(x). This is par- ticularly clear for the l-th (l ≥ 1) nonlinear conductance peak. Indeed, for λ(x) < 1 (which is the case considered in our calculations, see Fig. 5) one finds that Bl(x) is a monotonically increasing function of λ(x) - see Eq. (35) - which implies an increase of the conductance for in- creasing coupling strength. This fact allows to directly map the spatial modulations of the nonlinear differential conductance into modulations of λ(x). To be more specific, let us consider the resonance case ∆E = 0 ⇐⇒ Ng = 1 2 + 3ω1 8Eρ+ (40) and study G. As is clear by inspecting Figs. 9(a-d), the differential conductance G exhibits position-dependent modulations in close agreement to the behavior of λ(x). The most striking features show up indeed near the vi- bron position, where G is suppressed for V ≈ 0, and is enhanced for eV ≈ 2lΩ0 (l = 1, 2, . . .), see Fig. 5. For the case of a metallic CNT, shown in Fig. 10, the spa- tial modulations of the conductance are less pronounced due to the decreased intensity of the electron-vibron cou- pling with respect to the semiconducting case. The close resemblance of the spatial modulations of the non linear G with λ(x) is supported studying the conduc- tance at fixed bias shown in Fig. 11 for eV = 2Ω0 and eV = 4Ω0. Clearly G is enhanced where λ(x) is large. This confirms that position-resolved conductance maps are source of valuable informations about the intensity of the strain field along the CNT and consequently on the location and size of the vibron mode. FIG. 9: (Color online) Semiconducting CNT: Plot of G (units e2Γ(T) 0 /Ω0) as a function of the tip position x and bias voltage V (units Ω0/e) at resonance Ng = (1/2) + (3ω1/8Eρ+ ) and (a) δ = 1, xv0 = 0 and p = 1; (b) same as in (a) but for p = 2; (c) δ = 0.4, xv0 = 0.15L and p = 1; (d) same as in (c) but for p = 2. Other parameters: λm = 2, α = 5, g = 1, kBT = 0.15 Ω0 and A = 100. (Color online) Metallic CNT: Plot of G (units FIG. 10: e2Γ(T) 0 /Ω0) as a function of the tip position x and bias voltage V (units Ω0/e) at resonance Ng = (1/2) + (3ω1/8Eρ+ ) and (a) δ = 1, xv0 = 0 and p = 1; (b) same as in (a) but for p = 2; (c) δ = 0.4, x0 = 0.15L and p = 1; (d) same as in (c) but for p = 2. Other parameters: λm = 2, α = 32, g = 1, kBT = 0.15 Ω0 and A = 100. 9 Figure 12(a) shows the position-resolved nonlinear con- ductance at eV = 2Ω0 for increasing Coulomb interaction strength (decreasing values of g). Clearly, conductance is suppressed in turns, signaling the suppression of λ(x). This fact can be explained in terms of an increase of the velocity of the charged mode vρ+ = vF/g which induces an effective parameter α higher than that of the nonin- teracting case. Concerning the role of the asymmetry between the tip rate and the leads rate, Fig. 12(b) shows that the amplitude of the conductance modulations is decreased when making the tunnel barriers more symmetric with a collapse when reversing the asymmetry A < 1 (i.e. making the contacts more opaque than the STM tip). This can be understood by observing that for A (cid:29) 1 the current is dominated by the slowest barrier which is the STM one. This implies that the space-dependent tunneling rate can be efficiently probed in this regime. IV. CONCLUSIONS In this paper we have shown how transport measure- ments, performed with a scanning tunnel microscope tip on a suspended carbon nanotube, can bring information about its vibrational stretching dynamics. This theory predicts a position-dependent coupling constant, which is larger in the region where the vibron is located. The position-dependent coupling constant strongly af- fects the tunneling rate of electrons through the tip and has relevant consequences in the transport spectra. In particular, we showed that conductance maps in the lin- ear and nonlinear regime, obtained sweeping the tip along the nanotube, are closely connected to the local coupling constant and allow to localize the position and size of the vibron. Effects can be more pronounced in semiconduct- ing nanotubes due to the reduced Fermi velocity which matches more closely the speed of the vibrational mode. The role of electronic interactions and of the asymmetry between tip and metal contact tunnel barriers have also been addressed. This work could inspire a new class of experiments which aim at studying the vibrational degrees of freedom of a nanotube by means of electrical measurements. Acknowledgments. The authors acknowledge stimulating discussions with V. Cataudella, E. Mariani, A. Nocera, E. Paladino and C. Stampfer. Financial support by the EU- FP7 via ITN-2008-234970 NANOCTM is also gratefully acknowledged. (Color online) Differential conductance G (units FIG. 11: e2Γ(T) 0 /Ω0) as a function of the tip position x for Ng = (1/2)+(3ω1/8Eρ+ ) and eV = 2Ω0 (red solid line) or eV = 4Ω0 (blue dashed line) for a metallic CNT: (a) δ = 1, xv0 = 0 and p = 1; (b) same as in (a) but for p = 2; (c) δ = 0.4, xv0 = 0.15L and p = 1; (d) same as in (c) but for p = 2. Other parameters: λm = 2, α = 32, g = 1, kBT = 0.15 Ω0 and A = 100. C. Interaction and asymmetry effects We close by briefly commenting about the role of electron-electron interactions and that of the barriers asymmetry. FIG. 12: (Color online) (a) Conductance G (units e2Γ(T) 0 /Ω0) as a function of the tip position x for decreasing values of g from g = 1 (noninteracting, lightest gray) to g = 0.2 (strongly interacting, black); (b) Plot of G for different asymmetries (see key) and g = 1. Other parameters: δ = 1, xv0 = 0, p = 1, λm = 2, α = 5, kBT = 0.15 Ω0 and eV = 2Ω0. 1 S. Iijima, Nature (London) 354, 56 (1991). 2 J.-C. Charlier, X. Blase, and S. Roche, Rev. Mod. Phys. 79, 677 (2007). 3 R. Saito, G. Dresselhaus, and M. S. Dresselhaus, 'Physical Properties of Carbon Nanotubes', Imperial College Press (1998). 10 4 D. H. Cobden and J. Nygard, Phys. Rev. Lett. 89, 046803 Phys. Rev. A 75, 015602 (2007). (2002). 5 M. Bockrath, D. H. Cobden, P. L. McEuen, N. G. Chopra, A. Zettl, A. Thess, and R. E. Smalley, Science 725, 1922 (1997). 6 S. J. Tans, M. H. Devoret, H. Dai, A. Thess, R. E. Smalley, 30 J. Qian, B. I. Halperin, and E. J. Heller, Phys. Rev. B 81, 125323 (2010). 31 U. De Giovannini, F. Cavaliere, R. Cenni, M. Sassetti, and B. Kramer, Phys. Rev. B 77, 035325 (2008). 32 F. Cavaliere, U. De Giovannini, M. Sassetti, and B. L. J. Geerligs, and C. Dekker, Nature 386, 474 (1997). Kramer, New J. Phys. 11, 123004 (2009). 7 H. W. C. Postma, T. Teepen, Z. Yao, M. Grifoni, and C. 33 T. Giamarchi, Quantum Physics in One Dimension, Ox- Dekker, Science 293, 76 (2001). 8 A. K. Huttel, G. A. Steele, B. Witkamp, M. Poot, L. P. Kouwenhoven, and H. S. J. van der Zant, Nano Lett. 9, 2547 (2009). 9 C. Stampfer, A. Jungen, R. Linderman, D. Obergfell, S. Roth, and C. Hierold, Nano Lett. 6, 1449 (2006). 10 M. Poot and H. S. J. van der Zant, to appear on Phys. ford Science Publications (2004). 34 R. Egger and A. O. Gogolin, Phys. Rev. Lett. 79, 5082 (1997). 35 D. Bercioux, G. Buchs, H. Grabert, and O. Groning, Phys. Rev. B 83, 165439 (2011). 36 S. Eggert, Phys. Rev. Lett. 84, 4413 (2000). 37 A. Cr´epieux, R. Guyon, P. Devillard, and T. Martin, Phys. Rep., arXiv:1106.2060v1. Rev. B 67, 205408 (2003). 11 B. J. Leroy, S. G. Lemay, J. Kong, and C. Dekker, Appl. 38 A. V. Lebedev, A. Cr´epieux, and T. Martin, Phys. Rev. B Phys. Lett. 84, 4280 (2004). 12 B. J. LeRoy, S. G. Lemay, J. Kong, and C. Dekker, Nature 432, 371 (2004). 13 B. J. LeRoy, I. Heller, V. K. Pahilwani, C. Dekker, and S. G. Lemay, Nano Lett. 7, 2937 (2007). 14 V. Sazonova, Y. Yaish, H. Ustunel, D. Roundy, T. A. Arias, and P. L. McEuen, Nature 431, 284 (2004). 15 S. Sapmaz, P. Jarillo-Herrero, Y. M. Blanter, C. Dekker, and H. S. J. van der Zant, Phys. Rev. Lett. 96, 026801 (2006). 16 A. K. Huttel, B. Witkamp, M. Leijnse, M. R. Wegewijs, and H. S. J. van der Zant, Phys. Rev. Lett. 102, 225501 (2009). 17 R. Leturcq, C. Stampfer, K. Inderbitzin, L. Durrer, C. Hierold, E. Mariani, M. G. Schultz, F. von Oppen, and K. Ensslin, Nat. Phys. 5, 327 (2009). 18 A. K. Huttel, H. B. Meerwaldt, G. A. Steele, M. Poot, B. Witkamp, L. P. Kouwenhoven, and H. S. J. van der Zant, Phys. Status Solidi B 247, 2974 (2010). 19 G. A. Steele, A. K. Huttel, B. Witkamp, M. Poot, H. B. Merrwaldt, L. P. Kouwenhoven, and H. S. J. van der Zant, Science 325, 1103 (2009). 20 H. Lin, J. Lagoute, V. Repain, C. Chacon, Y. Girard, F. Ducastelle, H. Amara, A. Loiseau, P. Hermet, L. Henrard, and S. Rousset, Phys. Rev. B 81, 235412 (2010). 21 J. Lee, S. Eggert, H. Kim, S.-J. Kahng, H. Shinohara, and Y. Kuk, Phys. Rev. Lett. 93, 166403 (2004). 22 Y.-F. Chen, T. Dirks, G. Al-Zoubi, N. O. Birge, and N. Mason, Phys. Rev. Lett. 102, 036804 (2009). 23 W. Clauss, D. J. Bergeron, M. Freitag, C. L. Kane, E. J. Mele, and A. T. Johnson, Europhys. Lett. 47, 601 (1999). 24 M. Furuhashi and T. Komeda, Phys. Rev. Lett. 101, 185503 (2008). 25 L. C. Venema, J. W. G. Wilder, J. W. Janssen, S. J. Tans, H. L. J. T. Tuinstra, L. P. Kouwenhoven, and C. Dekker, Science 283, 52 (1999). 26 S. G. Lemay, J. W. Janssen, M. van den Hout, M. Mooij, M. J. Bronikowski, P. A. Willis, R. E. Smalley, L. P. Kouwenhoven, and C. Dekker, Nature (London) 412, 617 (2001). 27 M. Ouyang, J.-L. Huang, and C. M. Lieber, Phys. Rev. Lett. 88, 066804 (2002). 28 G. Buchs, D. Bercioux, P. Ruffieux, P. Groning, H. Grabert, and O. Groning, Phys. Rev. Lett. 102, 245505 (2009). 29 S. H. Abedinpour, M. Polini, G. Xianlong, and M. P. Tosi, 71, 075416 (2005). 39 C. Bena, Phys. Rev. B 82, 035312 (2010). 40 H. Suzuura and T. Ando, Phys. Rev. B 65, 235412 (2002). 41 A. De Martino and R. Egger, Phys. Rev. B 67, 235418 (2003). 42 G. D. Mahan, Phys. Rev. B 68, 125409 (2003). 43 G. Pennington and N. Goldsman, Phys. Rev. B 68, 045426 (2003). 44 K. Flensberg, New J. Phys. 8, 5 (2006). 45 E. Mariani and F. von Oppen, Phys. Rev. B 80, 155411 (2009). 46 M. Verissimo-Alves, R. B. Capaz, B. Koiller, E. Artacho, and H. Chacham, Phys. Rev. Lett. 86, 3372 (2001). 47 W. Izumida and M. Grifoni, New J. Phys. 7, 244 (2005). 48 A. Zazunov, D. Feinberg, and T. Martin, Phys. Rev. B 73, 115405 (2006). 49 X. Y. Shen, B. Dong, X. L. Lei, and N. J. M. Horing, Phys. Rev. B 76, 115308 (2007). 50 J. Franck, Trans. Faraday Soc. 21, 536 (1926). 51 E. Condon, Phys. Rev. 28, 1182 (1926). 52 S. Braig and K. Flensberg, Phys. Rev. B 68, 205324 (2003). 53 J. Koch, F. von Oppen, and A. V. Andreev, Phys. Rev. B 74, 205438 (2006). 54 J. Koch and F. von Oppen, Phys. Rev. Lett. 94, 206804 (2005). 55 F. Haupt, F. Cavaliere, R. Fazio, and M. Sassetti, Phys. Rev. B 74, 205328 (2006). 56 M. Merlo, F. Haupt, F. Cavaliere, and M. Sassetti, New J. Phys. 10, 023008 (2008). 57 F. Cavaliere, G. Piovano, M. Sassetti, and E. Paladino, New J. Phys. 10, 115004 (2008). 58 F. Cavaliere, E. Mariani, R. Leturcq, C. Stampfer, and M. Sassetti, Phys. Rev. B 81, 201303(R) (2010). 59 H. Yoshioka and Y. Okamura, J. Phys. Soc. Jpn. 71, 2512 (2002). 60 M. Grifoni and L. Mayrhofer, Eur. Phys. J. B 56, 107 (2007). 61 T. Kleimann, F. Cavaliere, M. Sassetti, and B. Kramer, Phys. Rev. B 66, 165311 (2002). 62 F. Cavaliere, A. Braggio, J. T. Stockburger, M. Sassetti, and B. Kramer, Phys. Rev. Lett. 93, 036803 (2004). 63 F. Cavaliere, A. Braggio, M. Sassetti, and B. Kramer, Phys. Rev. B 70, 125323 (2004). 64 R. A. Jishi, M. S. Dresselhaus, and G. Dresselhaus, Phys. Rev. B 48, 11385 (1993). 65 K. I. Bolotin, K. J. Sikes, J. Hone, H. L. Stormer, and P. Kim, Phys. Rev. Lett. 101, 096802 (2008). 66 G. Piovano, F. Cavaliere, E. Paladino, and M. Sassetti, 69 A. Donarini, A. Yar, and M. Grifoni, arXiv:1109.0723 Phys. Rev. B 83, 245311 (2011). 67 P. Ullersma, Physica (Amsterdam) 32, 27 (1966). 68 A. Metelmann and T. Brandes, arXiv:1107.3762 (2011). (2011). 11 

1304.0950

1

1304

Interplay of Aharonov-Bohm and Berry phases in gate-defined graphene quantum dots

We study the influence of a magnetic flux tube on the possibility to electrostatically confine electrons in a graphene quantum dot. Without magnetic flux tube, the graphene pseudospin is responsible for a quantization of the total angular momentum to half-integer values. On the other hand, with a flux tube containing half a flux quantum, the Aharonov-Bohm phase and Berry phase precisely cancel, and we find a state at zero angular momentum that cannot be confined electrostatically. In this case, true bound states only exist in regular geometries for which states without zero-angular-momentum component exist, while non-integrable geometries lack confinement. We support these arguments with a calculation of the two-terminal conductance of a gate-defined graphene quantum dot, which shows resonances for a disc-shaped geometry and for a stadium-shaped geometry without flux tube, but no resonances for a stadium-shaped quantum dot with a $\pi$-flux tube.

cond-mat.mes-hall

cond-mat

Interplay of Aharonov-Bohm and Berry phases in gate-defined graphene quantum dots Julia Heinl, Martin Schneider, and Piet W. Brouwer Dahlem Center for Complex Quantum Systems and Institut fur Theoretische Physik, Freie Universitat Berlin, Arnimallee 14, 14195 Berlin, Germany (Dated: November 3, 2018) We study the influence of a magnetic flux tube on the possibility to electrostatically confine elec- trons in a graphene quantum dot. Without magnetic flux tube, the graphene pseudospin is responsi- ble for a quantization of the total angular momentum to half-integer values. On the other hand, with a flux tube containing half a flux quantum, the Aharonov-Bohm phase and Berry phase precisely cancel, and we find a state at zero angular momentum that cannot be confined electrostatically. In this case, true bound states only exist in regular geometries for which states without zero-angular- momentum component exist, while non-integrable geometries lack confinement. We support these arguments with a calculation of the two-terminal conductance of a gate-defined graphene quantum dot, which shows resonances for a disc-shaped geometry and for a stadium-shaped geometry without flux tube, but no resonances for a stadium-shaped quantum dot with a π-flux tube. PACS numbers: 73.63.Kv, 73.22.Pr I. INTRODUCTION In recent years, graphene has emerged as a promising material for future nanoelectronical devices.1 -- 4 The pos- sibility to confine electrons is of particular relevance in this context. Experimental activity concentrates on con- finement in quantum dots realized with etched graphene structures5 -- 7 or graphene nanoflakes.8 Electrostatic con- finement with the help of metal gates, which is standard in semiconductor heterostructures, is problematic due to the absence of a band gap in monolayer graphene. In par- ticular, an electron that approaches a region of graphene with zero carrier density -- the closest approximation to an "electrostatic barrier" in graphene -- will penetrate this region with unit probability if at normal incidence. This phenomenon is known as "Klein tunneling".9 -- 11 Theoretical proposals suggest to use magnetic instead of electric fields to shape quantum dots12 or induce a gap in the spectrum.13 The statement that one cannot confine electrons in graphene using gate potentials can be circumvented in certain special cases.14 The reason is that Klein tunnel- ing is effective only at perpendicular incidence, while the reflection probability sharply increases away from nor- mal incidence. Certain integrable geometries, such as a disc,13,15 allow states that exclude perpendicular inci- dence, so that electrons can be effectively confined in a disc-shaped region of graphene with finite carrier den- sity, surrounded by a carrier-free (i.e., undoped, intrin- sic) graphene sheet. On the other hand, for geometries with a chaotic classical dynamics, no such exclusion of perpendicular incidence is possible, and one may expect that no bound states exist in this case. In Ref. 14, as well as in later studies,13,16,17 a circular and a stadium-shaped quantum dot, as prototypes of integrable and chaotic ge- ometries, were embedded in a carrier-free graphene re- gion and coupled to source and drain contacts, as shown schematically in Fig. 1. Bound states are then revealed as sharp resonances in the two-terminal conductance. FIG. 1: (Color online) The geometry under consideration: A quantum dot (here with circular shape), consisting of a region of graphene with a nonzero spatially uniform carrier density surrounded by a carrier-free graphene layer, which is coupled to leads in a two-terminal geometry. The total system has rectangular shape of dimension L × W ; the size of the quantum dot is denoted R. In this article we consider the effect of a magnetic flux tube through the quantum dot (indicated in red). If the flux tube carries half a flux quantum, the Aharonov-Bohm phase precisely cancels the Berry phase that is accumulated in a cyclic orbit inside the quantum dot. the Interestingly, conductance of a carrier-free graphene sheet with a stadium-shaped and disc-shaped quantum dot showed resonant features that were quanti- tatively different, but qualitatively similar.14,17 The quan- titative difference concerns the scaling of the resonance widths with the coupling to the leads, which is deter- mined by the ratio R/L of the quantum dot size R and the distance L between the source and drain contacts, see Fig. 1. Whereas for the stadium-shaped quantum dot the width was proportional to R/L for all resonances, the disc-shaped quantum dot also featured much nar- rower resonances, with a width that scaled proportional to (R/L)n with n ≥ 3. The qualitative similarity was that in the limit R/L → 0 both systems showed con- ductance resonances at all. This contradicts the naive classical expectation that there should be no resonances for a chaotic geometry, because in a chaotic geometry each electron eventually hits the dot boundary at per- pendicular incidence, and an electron that hits the dot boundary at perpendicular incidence exists the quantum dot with unit probability. No resonant structures should exist if the escape probability is unity after a finite time. This deviation from the naive classical expectation can be attributed to the Berry phase in graphene. In graphene, electrons are assigned a pseudospin that cor- responds to the sublattice degree of freedom. The pseu- dospin is locked to momentum. Upon completion of a full rotation the electron collects a Berry phase of π. This Berry phase has the important consequence that  the lowest possible angular momentum is 2 . Perpendic- ular incidence on the boundary of the quantum dot cor- responds to zero angular momentum, so that no states with perpendicular incidence on the surface exist. This then explains why a stadium-shaped quantum dot still shows conductance resonances, in spite of the naive clas- sical expectation that geometries with a chaotic classical dynamics can not be used to confine electrons. In order to support these arguments, in this article we study gate-defined graphene quantum dots in which the Berry phase is compensated by the Aharonov-Bohm phase from a magnetic flux tube through the quantum dot. (The interplay of the two phase shifts is also used experimentally to identify Berry phase effects, see, e.g., Refs. 18,19.) If the magnetic flux tube carries half a flux quantum ("π flux"), the Aharonov-Bohm phase and the Berry phase collected along a closed trajectory around the flux tube precisely cancel. We find that with a π flux the system can reach a state with zero kinematic angular momentum, that cannot be confined by means of gate potentials. The π-flux tube has qualitatively different conse- quences for disc-shaped and stadium-shaped geometries. For the disc-shaped geometry, states have a well-defined kinematic angular momentum. While the states with zero kinematic angular momentum are no longer con- fined, states with nonzero kinematic angular momentum remain confined to the quantum dot. Hence, for the disc- shaped quantum dot the inclusion of the π-flux tube elim- inates some of the resonances, but not all. On the other hand, for the stadium-shaped geometry, all states have a component in the zero-angular-momentum channel, so that inclusion of the π-flux tube leads to the suppression of all resonances.20 The remaining part of the paper is organized as follows: In Section II we calculate the bound states of a disc- shaped quantum dot in the presence of a magnetic flux tube. We find, that the asymptotic behavior of the zero- angular-momentum state is the same as for a free circular wave. Hence, no bound state can exist in this channel. In Section III we present a numerical calculation of the two- terminal conductance setup of Fig. 1, for a circular and a stadium-shaped quantum dot. Upon inclusion of the π-flux tube, we find that sharp resonances persist for the circular dot, while the conductance becomes featureless for the stadium dot in the limit R/L → 0. We conclude in Section IV. 2 II. DISC-SHAPED QUANTUM DOT The electrostatically-defined graphene quantum dot is described by the Hamiltonian H0 = vF(p + eA) · σ + V (r), (1) where vF is the Fermi velocity vF and σ = (σx, σy) are the Pauli matrices. The gate potential V (r) is nonzero and constant inside the quantum dot, and zero elsewhere, (cid:40)−vFV0, V (r) = 0, r < R r > R, (2) where R is the radius of the disc-shaped dot. The con- stant V0 has the dimension of inverse length. We choose V0 > 0, so that the quantum dot is electron doped. The potential V (r) is smooth on the scale of the lattice con- stant, justifying our description in terms of a single Dirac point. The choice of a spatially uniform potential inside dot makes a closed-form solution of the wavefunctions possible and allows for a straightforward comparison to the classical dynamics in the quantum dot, but it is not essential for the existence of bound states.21,22 The struc- ture of quasibound states in the inverted setup (zero po- tential inside, nonzero outside) was considered in Refs. 23,24. The vector potential corresponding to the magnetic flux line is A(r) = h e Φ 2πr eθ, (3) where eθ is the unit vector for the azimuthal angle, and Φ is the magnetic flux measured in units of the flux quan- tum h/e. In polar coordinates, the kinetic part of the Hamiltonian then reads vF(p + eA) · σ = −ivF , (4) (cid:19) (cid:18) 0 D− (cid:19) D+ 0 (cid:18) where we defined the operators D± = e±iθ ∂r ± i r ∂θ ∓ Φ r . (5) With our choice of the vector potential, the Hamil- tonian is invariant under rotation, hence we can look for  eigenstates of the total angular momentum jz = lz + 2 σz. They have the form (cid:18) e−i θ (cid:19) ψm(r) = eimθ 2 ϕm,+(r) 2 ϕm,−(r) ei θ , (6) where m = ±1/2, ±3/2, . . . . Inside the dot, for r < R, the radial wave functions ϕm,± are determined by the coupled equations (cid:0)∂r − (m − 1 (cid:0)∂r + (m + 1 r Φ(cid:1) ϕm,+(r) = iV0ϕm,−(r), r Φ(cid:1) ϕm,−(r) = iV0ϕm,+(r). r − 1 r + 1 2 ) 1 2 ) 1 (7) Outside the dot the equations decouple, and the radial wave functions show a power law behavior ϕm,+(r) = a+rm−1/2+Φ, ϕm,−(r) = a−r−m−1/2−Φ, (8) with coefficients a±. A. Without flux tube We first review the solutions without flux tube, for Φ = 0.14 With the requirement that the wavefunction is regular for r → 0, we find for the solution inside the dot ϕm,+(r) = Jm−1/2(V0r), ϕm,−(r) = isgn(m)Jm+1/2(V0r), (9) where Jn(x) is the Bessel function. Outside the dot, the wave function must not diverge, which gives the con- straints a+ = 0 (m > 0) and a− = 0 (m < 0). From continuity of the wavefunction at r = R, we find the resonance condition Jm−1/2(V0R) = 0. (10) The wavefunction outside the dot is decaying as ∝ r−(m+1/2). In Section III we connect the quantum dots and the surrounding undoped graphene layer to source and drain contacts. The distance between the contacts is denoted L and the quantum dot is placed halfway between the In the limit L (cid:29) R, the bound contacts, see Fig. 1. states are then revealed as resonances in the two-terminal conductance as a function of the gate potential V0. These resonances have a finite width Γ, which can be estimated as14 ΓR ∼ ψ(L)2L/ψ(R)2R. We conclude, that the width of the resonances without flux tube scales as . (11) For m = 1/2 the wavefunction decays proportional to 1/r, which is marginally non-normalizable. Despite the absence of a bound state in the strict sense, the con- ductance nevertheless shows a resonance, with a width ΓR ∝ (R/L).14,16,17 B. With flux tube We now consider a disc-shaped quantum dot with a flux tube carrying half a flux quantum (Φ = 1/2) -- a (cid:18) R (cid:19)2m L ΓR ∝ L (cid:19) 3 "π flux" -- at its center. The results take a form similar to those without flux tube if we consider the kinematical orbital angular momentum, lz,kin = [r × (p + eA)]z, (12) instead of the canonical angular momentum. With the  inclusion of a π-flux, we then find lz,kin = lz + 2 . The wavefunctions from Eq. (6) are then eigenstates of jz,kin with eigenvalue µ, where µ = m + 1/2, i.e. the kine- matical angular momentum takes on integer values. For µ (cid:54)= 0 the calculation for the bound states proceeds in the same way as without flux, and we find that the resonance condition is given by Jµ−1/2(V0R) = 0. (13) Outside the dot, the wavefunction decays proportional to r−(µ+1/2). We conclude that, if the dot and the sur- rounding undoped graphene layer are contacted to source and drain reservoirs, the width Γ of the resonances in the two-terminal conductance scales as (cid:18) R (cid:19)2µ ΓR ∝ . (14) . + b2 ψ(r) = b1 iY1/2(V0r) −iJ1/2(V0r) (cid:18) e−iθJ1/2(V0r) The state with zero kinematical angular momentum (µ = 0) however is special: First of all, inside the dot, the wavefunction is of the form (cid:18) e−iθY1/2(V0r) (cid:19) simple form J1/2(x) = (cid:112)2/πx sin x, and Y1/2(x) = −(cid:112)2/πx cos x, we see that ψ(r) diverges as 1/ (15) Recalling that the half-integer Bessel functions take the V0r at the origin, and that there is no non-trivial choice of co- efficients b1 and b2 which removes this divergence. The root of this singular behavior lies in the vector potential, which is singular upon approaching the origin. The prob- lem can be cured by regularizing the vector potential. One possibility is to let the flux Φ have an r-dependence, such that Φ = 0 for r < ρ and Φ = 1/2 for r > ρ, i.e., the flux is not located at the origin, but on a circle of radius ρ. Obviously, the problem is now well-defined at the origin, and we can take the solution from the case without flux tube, √ (cid:18) e−iθJ1(V0r) −iJ0(V0r) (cid:19) ψ(r) = c , (16) where c is a complex constant. We then match the wave- functions from Eq. (16) and Eq. (15) at r = ρ. Upon taking ρ → 0, we get b2 = 0 as a condition for Eq. (15). The boundary condition at the origin ensures, that there is precisely one solution for zero angular momentum. The µ = 0 state is also special outside the dot, where the wavefunction is proportional to 1√ r in both compo- nents. Thus it has the same decay as a free circular wave in two dimensions and, hence, it does not allow for the formation of a bound state. This conclusion is indepen- dent of the choice of the regularization of the wavefunc- tion near r = 0. Summarizing: Without flux tube, the bound states are labeled by the angular momentum quantum number m, which takes half-integer values. For m = 1/2 one has a "quasi-bound state", because the corresponding wave- function is marginally non-normalizable. With a π flux tube, the bound states are labeled by the kinematic angu- lar momentum quantum number µ, which takes integer values. There is no bound state for µ = 0. III. TWO-TERMINAL CONDUCTANCE Following Refs. 14,16,17 we now attach metallic source and drain contacts to the undoped graphene layer that surrounds the quantum dot. Schematically, this setup is shown in Fig. 1. We then calculate the two-terminal conductance, where bound states of the dot show up as resonant features as a function of the gate voltage V0. The contacts are included by the addition of an addi- tional potential Uleads with25 Uleads = if −L/2 < x < L/2, ∞ if x < −L/2 or x > L/2. (cid:26) 0 (17) (18) We apply periodic boundary conditions in the y direction, with period W . For the vector potential A we take a different gauge than in Sec. II, (cid:26) 0 if 0 < y < W/2, 1 if −W/2 < y < 0, A(r) = δ(x)ex × hΦ e where ex is the unit vector in the x direction. With this choice of the vector potential there are two flux tubes: one, at y = 0, located in the quantum dot, and one, at y = W/2, located outside the quantum dot. The second flux tube is necessary to implement the periodic bound- ary conditions. It does not affect the conductance res- onances in the limit that the sample width W is much larger than the distance L between source and drain con- tacts. The numerical calculation of the two-terminal conduc- tance follows the method of Ref. 26. Details specific to the presence of the flux tube are discussed in the ap- pendix. We now compare results for quantum dots with and without flux tube. We give results for a disc-shaped quantum dot, as a prototype of a quantum dot with in- tegrable dynamics, and a stadium-shaped quantum dot, the prototype of a dot with chaotic dynamics. A. Disc-shaped dot The two-terminal conductance for the case of a disc- shaped quantum dot without and with flux tube is shown in Fig. 2. The figure shows pronounced resonances as a function of the gate voltage V0, with positions that agree 4 FIG. 2: (Color online) Two-terminal conductance of a graphene sheet containing a disc-shaped quantum dot with- out (top) and with (bottom) π-flux tube. Model parameters are R/L = 0.2 and W/L = 6. Without flux tube, resonances have definite angular momentum, with quantum number m indicated at each resonance [data taken from Ref. 14]. With- out flux tube, resonances are labeled by the kinematic angular momentum quantum number µ. No resonance is found for µ = 0. with the ones calculated Sec. II. Without flux tube, the resonances are labeled by the quantum number m = 1/2, 3/2, 5/2, . . . . Their width scales ∝ (R/L)2m as the coupling to the leads is decreased (data not shown), consistent with Eq. (11) and Refs. 14,16,17. With flux tube, the resonances are labeled by the kinematic angular momentum quantum number µ = 1, 2, 3, . . . . There are no resonances for µ = 0. Upon decreasing the coupling to the leads, the resonances become narrower but retain their height, see Fig. 3, and the scaling of the resonance width with the ratio R/L is consistent with Eq. (14) (data not shown) . B. Stadium-shaped dot As a prototypical example of a chaotic quantum dot, we consider a stadium-shaped quantum dot. Here the potential V (r) = −vFV0 for positions r inside the sta- dium and V (r) = 0 otherwise. Without magnetic flux, the two-terminal conductance shows resonances, which, in the limit of small R/L, all behave as the m = 1/2- type resonances of the disc-shaped dot, i.e., their height remains finite, whereas the resonance width scales pro- 5 FIG. 4: (Color online) Two-terminal conductance of a graphene sheet containing a stadium-shaped quantum dot without (top) and with (bottom) a π-flux. Parameters for the calculation are R/L = 0.2, W/L = 12, a/R = 3/2, d = 2a/3. Without flux tube, the calculation for the conduc- tance was done with the method of Ref. 17. √ IV. CONCLUSION In this article we investigated the observation of Refs. 14,16,17, that the two-terminal conductance of a generic gate-defined graphene quantum dot shows resonances in the limit of a weak coupling to the leads, in spite of the naive expectation that electrons can not be confined in such a quantum dot because of Klein tunneling. We at- tribute this observation to the Berry phase in graphene, which quantizes angular momenta to half-integer values. With half-integer angular momenta, strict perpendicular incidence -- the condition for Klein tunneling with unit probability -- does not occur. As a consequence, con- ductance resonances exist in both integrable and chaotic geometries. The only difference between the two cases is a quantitative one: it concerns the scaling of the reso- nance widths with the coupling to the leads.17 can be The Berry phase cancelled against an Aharonov-Bohm phase, when a flux tube containing half a flux quantum is introduced to the system. With a magnetic flux tube, we showed that the relevant angu- lar momentum, the kinematical angular momentum, is quantized to integer values. In this case a state with zero angular momentum is possible. Such a state can not form a bound state or give rise to a conductance res- onance. We showed this by an explicit calculation for the FIG. 3: (Color online) First two resonances for a disc-shaped quantum dot with π-flux tube, for different coupling strengths to the leads. Calculations are performed for W/L = 8 and various R/L, as indicated in the figure. The second resonance is shown enlarged in the inset. portional to R/L.17 The numerical data shown in the top panels of Figs. 4 and 5 clearly reveal these reso- nances, although the asymptotic scaling of the resonance width and resonance height with R/L is somewhat ob- scured by transient contributions for moderate R/L that originate from higher-angular-momentum contributions to the resonances.17 The conductance trace for a stadium-shaped quantum dot with a flux tube carrying half a flux quantum is shown in the bottom panels of Figs. 4 and 5. In order to break inversion symmetry, the stadium is placed asymmetri- cally with respect to the flux tube, see the inset of Fig. 4. The differences with the case of the disc-shaped quantum dot and with the case without a flux tube are significant. We find that the conductance depends on the gate volt- age V0 for finite R/L, but the widths of the "resonances" is independent of the coupling to the leads, which is set by the ratio R/L, whereas the height decreases upon de- creasing R/L. This agrees with the expectation that, since all states in the stadium have a µ = 0 compo- nent, a stadium dot should not support any (quasi)bound states. While for intermediate values of R/L contribu- tions from higher angular momentum channels still give rise to broad "quasi-resonances", in the limit R/L → 0, only the µ = 0 channel is relevant, and the conductance becomes featureless as a function of V0. We remark that, if the flux tube would be placed ex- actly in the middle of the stadium, inversion symme- try would split the resonances into two groups, result- ing from even and odd µ. The "even" resonances have a finite µ = 0 component and disappear upon taking the limit R/L → 0. The "odd" resonances survive in this limit, with a finite resonance height and a resonance width Γ ∝ (R/L)2 (data not shown). Acknowledgments 6 We gratefully acknowledge discussions with Jens Bar- darson, Igor Gornyi, Pavel Ostrovsky, Brian Tarasin- ski, and Silvia Viola Kusminskiy. This work is sup- ported by the Alexander von Humboldt Foundation in the framework of the Alexander von Humboldt Profes- sorship, endowed by the Federal Ministry of Education and Research and by the German Research Foundation (DFG) in the framework of the Priority Program 1459 "Graphene". Appendix A: Numerical approach V (r) is replaced by a potential(cid:80) The numerical approach follows Ref. 26. The potential n Vn(y)δ(x − xn) that is nonzero at N discrete values −L/2 = x0 < x1 < x2 < . . . < xN−1 < xN = L/2 of the x coordinate, with Vn(y) = dxV (x, y), n = 1, 2, . . . , N − 1. (cid:90) (xn+xn+1)/2 (xn−1+xn)/2 Between the discrete points the wavefunction is solved from the free Dirac equation. This solution takes its sim- plest form after Fourier transform with respect to the transverse coordinate y, because the free Dirac equa- tion does not couple different transverse modes. These solutions are then matched by applying the appropri- ate boundary conditions at the discrete points x = xj, j = 1, 2, . . . , N . A numerically stable method to imple- ment this program is to express both the solution of the free Dirac equation and the matching conditions at the discrete points x = xj in terms of scattering matrices. The scattering matrix of the entire sample is then ob- tained from convolution of the scattering matrices of the 2N − 1 individual components. The result of the calcula- tion is the transmission matrix t, which is related to the two-terminal conductance via the Landauer formula, G = tr tt†, 4e2 h (A1) where the trace is taken of the transverse Fourier modes. The number of "slices" N and the number of transverse modes M must be chosen large enough, that the conduc- tance G no longer depends on N and M . The vector potential (18) corresponds to the boundary condition ψ(x, y) = − lim x↓0 lim x↑0 ψ(x, y) for −W/2 < y < 0, (A2) whereas ψ(x, y) is continuous at x = 0 for 0 < y < W/2. In the approach described above, this boundary condition is expressed in terms of a scattering matrix relating incoming and outgoing waves at x ↑ 0 and x ↓ 0. This scattering matrix has no reflective part, whereas the transmission matrix is tmn = −4i/[(km − kn)W ] if m − n even, if m − n odd, (A3) (cid:26) 0 FIG. 5: (Color online) Behavior of the first three quasi reso- nances of the stadium-shaped quantum dot without (top) and with (bottom) π-flux upon changing the coupling to the leads R/L. The other parameters are the same as in Fig. 4. disc-shaped quantum dot in Sec. II, and using numerical calculations for disc-shaped and stadium-shaped quan- tum dots in Sec. III. Once the Aharonov-Bohm phase from the π-flux tube cancels the Berry phase, the re- sults of the full quantum theory are consistent with the simple classical expectations. With a π flux, there is a stark qualitative difference between conductance res- onance for integrable and non-integrable quantum dots: Whereas sharp conductance resonances for the case of an integrable quantum dot continue to exist, in the limit of weak lead-dot coupling the conductance becomes fea- tureless for a generic non-integrable quantum dot. In closing, we make two remarks concerning the possi- ble realization of the scenario we investigated here. First, a flux tube for graphene not necessarily has to be created by a real magnetic field, but it can also be engineered via strain as a pseudo-magnetic field.27,28 The pseudo- magnetic field would have opposite signs for the two val- leys, which is of no consequence for our conclusions, be- cause the two valleys are decoupled for the smooth con- fining potentials we consider here. Second, the appli- cation of a well-defined Aharonov-Bohm phase is more controlled in ring-shaped structures.18,19 In this case, we expect that the our main finding, the strong qualitative difference between integrable and non-integrable geome- tries in the case of a π flux, persists. where the integers m and n label the transverse modes and kn = 2πn/W . This transmission matrix has the special properties that t = t† and t2 = 1. In order to ensure numerical stability, unitarity must be preserved while restricting to a finite number of trans- verse modes M . For the transmission matrix (A3) this can be achieved using the following trick: One first builds the hermitian matrix h = i(eiφ − t)/(t + eiφ) = cot φ− t/ sin φ out of the transmission matrix, where φ is a phase that can be chosen arbitrarily, then truncates h, which can be done straightforwardly without com- promising hermiticity, and then uses the inverse relation t = eiφ(1 + ih)/(1 − ih) to obtain a finite-dimensional transmission matrix. In our numerical calculation we set φ = π/2. We verified that the elements of the resulting finite-dimensional transmission matrix approach the ele- ments of the exact transmission matrix (A3) in the limit that the number of transverse mode M → ∞. 7 1 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 2 C. W. J. Beenakker, Rev. Mod. Phys. 80, 1337 (2008). 3 A. K. Geim, Science 324, 1530 (2009). 4 S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). 5 L. A. Ponomarenko, F. Schedin, M. I. Katsnelson, R. Yang, E. W. Hill, K. S. Novoselov, and A. K. Geim, Science 320, 356 (2008). 6 C. Stampfer, E. Schurtenberger, F. Molitor, J. Guttinger, T. Ihn, and K. Ensslin, Nano Lett. 8, 2378 (2008). 7 A. Jacobsen, P. Simonet, K. Ensslin, and T. Ihn, New J. Phys. 14, 023052 (2012). 8 S. K. Hamalainen, Z. Sun, M. P. Boneschanscher, A. Up- pstu, M. Ijas, A. Harju, D. Vanmaekelbergh, and P. Lil- jeroth, Phys. Rev. Lett. 107, 236803 (2011). 9 O. Klein, Z. Phys. 53, 157 (1929). 10 M. Katsnelson, K. Novoselov, and A. Geim, Nat. Phys. 2, 620 (2006). 17 M. Schneider and P. W. Brouwer, Phys. Rev. B 84, 115440 (2011). 18 J.-B. Yau, E. P. De Poortere, and M. Shayegan, Phys. Rev. Lett. 88, 146801 (2002). 19 F. Qu, F. Yang, J. Chen, J. Shen, Y. Ding, J. Lu, Y. Song, H. Yang, G. Liu, J. Fan, et al., Phys. Rev. Lett. 107, 016802 (2011). 20 A similar situation occurs, if one tries to confine electrons on a rectangular strip. Here, perpendicular incidence is only possible for zero transverse momentum. Whether such a state exists strongly depends on the boundary conditions, see P. G. Silvestrov and K. B. Efetov, Phys. Rev. Lett. 98, 016802 (2007). 21 C. A. Downing, D. A. Stone, and M. E. Portnoi, Phys. Rev. B 84, 155437 (2011). 22 V. V. Mkhitaryan and E. G. Mishchenko, Phys. Rev. B 86, 115442 (2012). 23 A. Matulis and F. M. Peeters, Phys. Rev. B 77, 115423 (2008). 24 P. Hewageegana and V. Apalkov, Phys. Rev. B 77, 245426 11 V. V. Cheianov and V. I. Fal'ko, Phys. Rev. B 74, 041403 (2008). (2006). 12 A. De Martino, L. Dell'Anna, and R. Egger, Phys. Rev. Lett. 98, 066802 (2007). 13 G. Pal, W. Apel, and L. Schweitzer, Phys. Rev. B 84, 075446 (2011). 25 J. Tworzydlo, B. Trauzettel, M. Titov, A. Rycerz, and C. W. J. Beenakker, Phys. Rev. Lett. 96, 246802 (2006). 26 J. H. Bardarson, J. Tworzyd(cid:32)lo, P. W. Brouwer, and C. W. J. Beenakker, Phys. Rev. Lett. 99, 106801 (2007). 27 M. M. Fogler, F. Guinea, and M. I. Katsnelson, Phys. Rev. 14 J. H. Bardarson, M. Titov, and P. W. Brouwer, Phys. Rev. Lett. 101, 226804 (2008). Lett. 102, 226803 (2009). 15 M. Calvo, Phys. Rev. B 84, 235413 (2011). 16 M. Titov, P. M. Ostrovsky, I. V. Gornyi, A. Schuessler, and A. D. Mirlin, Phys. Rev. Lett. 104, 076802 (2010). 28 N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. C. Neto, and M. F. Crommie, Science 329, 544 (2010). 

1604.03931

1

1604

Quasi-two-dimensional Dirac fermions and quantum magnetoresistance in LaAgBi$_2$

We report quasi-two-dimensional Dirac fermions and quantum magnetoresistance in LaAgBi$_2$. The band structure shows several narrow bands with nearly linear energy dispersion and Dirac-cone-like points at the Fermi level. The quantum oscillation experiments revealed one quasi-two-dimensional Fermi pocket and another complex pocket with small cyclotron resonant mass. The in-plane transverse magnetoresistance exhibits a crossover at a critical field $B^*$ from semiclassical weak-field $B^2$ dependence to the high-field unsaturated linear magnetoresistance which is attributed to the quantum limit of the Dirac fermions. Our results suggest the existence of quasi 2D Dirac fermions in rare-earth based layered compounds with two-dimensional double-sized Bi square nets, similar to (Ca,Sr)MnBi$_{2}$, irrespective of magnetic order.

cond-mat.mes-hall

cond-mat

Quasi-two-dimensional Dirac fermions and quantum magnetoresistance in LaAgBi2 Kefeng Wang,1 D. Graf,2 and C. Petrovic1 1Condensed Matter Physics and Materials Science Department, Brookhaven National Laboratory, Upton, New York 11973, USA 2National High Magnetic Field Laboratory, Florida State University, Tallahassee, Florida 32306-4005, USA (Dated: November 8, 2018) We report quasi-two-dimensional Dirac fermions and quantum magnetoresistance in LaAgBi2. The band structure shows several narrow bands with nearly linear energy dispersion and Dirac- cone-like points at the Fermi level. The quantum oscillation experiments revealed one quasi-two- dimensional Fermi pocket and another complex pocket with small cyclotron resonant mass. The in-plane transverse magnetoresistance exhibits a crossover at a critical field B ∗ from semiclassical weak-field B 2 dependence to the high-field unsaturated linear magnetoresistance which is attributed to the quantum limit of the Dirac fermions. Our results suggest the existence of quasi 2D Dirac fermions in rare-earth based layered compounds with two-dimensional double-sized Bi square nets, similar to (Ca,Sr)MnBi2, irrespective of magnetic order. PACS numbers: 72.20.My,72.80.-r,75.47.Np I. INTRODUCTION Dirac fermions with linear energy-momentum disper- sion and corresponding Dirac cone states have been ob- served in two-dimensional graphene1,2 and the surface of topological insulators (TI).3,4 It is believed that the two bands with the opposite (pseudo)spins cross each other without hybridization giving the linear energy dispersion. Unlike the conventional electron gas with parabolic energy dispersion where Landau levels (LLs) are equidistant,5 the distance between the lowest and 1st LLs of Dirac fermions in magnetic field is very large. So the quantum limit where all of the carriers occupy only the lowest LL is easily realized even in moderate fields.6 Consequently some quantum transport phenomena such as quantum Hall effect and large linear magnetoresistance (MR) could be observed in the regular magnetic field in Dirac fermion system.6 -- 10 Thus, Dirac materials are now one of the central topics of condensed matter physics. In addition to nanoengineered or surface materi- als such as graphene and TIs, the Dirac fermions and Dirac nodes were observed or predicted in bulk crystals of iron-based superconductor parent mate- rial BaFe2As2,9 -- 12 (Sr/Ca)MnBi2 bismuth based lay- ered magnetic compounds13 -- 16 and in a molecular or- ganic conductor α−(BEDT-TTF)2I3.17 Among them, SrMnBi2 contains alternatively stacked two MnBi4 tetra- hedron layers and a two-dimensional (2D) Bi square net separated by Ca atoms along the c-axis. The linear energy dispersion originates from the crossing of two Bi 6px,y bands in the double-sized Bi square nets cor- responding to the double-sized N´eel-type antiferromag- netic (AFM) Mn unit cell.13 However, in a SrMnBi2-type structure, the nonmangetic unit cell still contains two Bi atoms in the Bi square net due to the occupation of Sr atoms.18 In previous report,13 all the analysis is based on the electronic structure of SrMnBi2 in antiferromag- netic state. Hence, the effect of AFM order within MnBi4 layers in the formation of Dirac nodes and whether any unit cell symmetry with double-sized Bi unit cell can host Dirac fermions is not completely clear. Besides, the engi- neering of Dirac states is of the great interest. With the change of band parameters in deformed graphene, Dirac points may merge or be completely removed.19 Similarly in bulk crystals, hopping terms and the band parame- ters can be tuned by changing the lattice parameters, hybridization or the space group of the crystallized struc- ture. It was reported that SrMnBi2 and CaMnBi2 host Dirac dispersion of different nature. Even though the conduction and valence bands touch at the Dirac point in both materials, the details are different due to the dif- ferent stacking of nearby alkaline earth atoms and the dif- ferent hybridization. For SrMnBi2, the zero-energy gap is found only at a specific point, while it is found along the continuous line in the momentum space for CaMnBi2.20 Hence exploring new bulk compounds with similar struc- ture to SrMnBi2 may provide more profound comprehen- sion of Dirac band crossing mechanisms in bulk crystal. Here we report quasi-2D Dirac fermions in the LaAgBi2 single crystal which has similar crystal lattice structure with CaMnBi2, but without magnetic ions.21 The band structure shows several narrow bands with nearly linear energy dispersion and Dirac-cone-like points at the Fermi level. The quantum oscillation experiments show one quasi-two-dimensional Fermi pocket and an- other very complex electron pocket with small cyclotron resonant mass. The in-plane transverse magnetoresis- tance exhibits a crossover at a critical field B∗ from semi- classical weak-field B2 dependence to the high-field un- saturated linear magnetoresistance due to the quantum limit of the Dirac fermions. The temperature dependence of B∗ satisfies quadratic behavior, which is attributed to the splitting of linear energy dispersion in high field. Our results demonstrate that Dirac fermions in bulk crys- tals can also be found in the absence of magnetic order and imply possible universal existence of two dimensional Dirac fermions in layered structure compounds with two- dimensional double-sized Bi square nets. II. EXPERIMENTAL Single crystals of LaAgBi2 were grown using a high- temperature self-flux method.21 The resultant crystals are plate-like. X-ray diffraction (XRD) data were taken with Cu Kα (λ = 0.15418 nm) radiation of Rigaku Mini- flex powder diffractometer. Electrical transport measure- ments up to 9 T were conducted in Quantum Design PPMS-9 with conventional four-wire method. In the in- plane measurements, the crystal was mounted on a ro- tating stage such that the tilt angle θ between sample surface (ab-plane) and the magnetic field can be continu- ously changed, with currents flowing in the ab-plane per- pendicular to magnetic field. The de Haas-van Alphen (dHvA) oscillation experiments were performed at Na- tional High Magnetic Field Laboratory, Tallahassee. The crystals were mounted onto miniature Seiko piezoresistive cantilevers which were installed on a rotating platform. The field direction can be changed continuously between parallel and perpendicular to the c-axis of the crystal. First principle electronic structure calculation were per- formed using experimental lattice parameters within the full-potential linearized augmented plane wave (LAPW) method 22 implemented in WIEN2k package.23 The gen- eral gradient approximation (GGA) of Perdew et al., was used for exchange-correlation potential.24 The LAPW sphere radius were set to 2.5 Bohr for all atoms, and the converged basis corresponding to Rminkmax = 7 with additional local orbital were used where Rmin is the min- imum LAPW sphere radius and kmax is the plane wave cutoff. Spin-orbit coupling for all elements were took into account by a second-variational method with the scalar- relativistic orbitals as basis which was implemented in WIEN2k. III. RESULTS AND DISCUSSIONS Fig. 1(a) shows the powder XRD pattern of flux grown LaAgBi2 crystals, which were fitted by RIET- ICA software.25 All reflections can be indexed in the P4/nmm space group. The determined lattice param- eters are a = b = 0.4582(8) nm and c = 1.062(4) nm, in agreement with the published data.21,26 The basal plane of a cleaved crystal is the crystallographic ab-plane where the 2D Bi layers (Bi2, the red balls in Fig. 1(b)) are lo- cated. Contrary to CaMnBi2, the adjacent 2D Bi layers along c-axis are separated by La atoms and AgBi layers without magnetic ions (Fig. 1(b)). This makes LaAgBi2 an ideal system to clarify the role of magnetic ions in the formation of Dirac fermions in CaMnBi2 and other similar compounds. Fig. 1(c) shows the first-principle band structure with- out spin-orbit coupling. Fig 1(d), Fig. 2(a) and Fig. 2(b-d) show the band structure and the density of states (DOS), and Fermi surfaces of LaAgBi2 with spin-orbit coupling, respectively. The band structure in Fig. 1 clearly shows several narrow linear bands. More inter- 2 esting, probably at the Fermi level, there are two Dirac- cone-like points along Γ − M , R − Z directions and at X points in the Brillouin zone (red circles in Fig. 1(d)). Compared to the band structure without spin-orbit in- teraction in Fig. 1(c), the spin-orbit coupling induces the gap at the Dirac-cone-like points and a lowering of the Fermi level around 20 meV in Fig. 1(d). Besides these, the band structure with and without spin-orbit coupling looks similar and the essential features of the FS's remain almost the same In Fig. 2(a), the Fermi level is located at the edge of the gap, and the main peaks of DOS from La, Ag and Bi1 are located far below the Fermi level. The conducting electrons in LaAgBi2 are mainly due to 5p from Bi2, while there is little con- tribution from other atoms as shown in Fig. 2(a). So the linear bands and the Dirac-cone like points at the Fermi level mainly originate from Bi bands. In (Sr/Ca)MnBi2, the antiferromagnetic order of Mn moments doubles the unit cell. Consequently two Bi 6px,y bands in the double- sized Bi square nets cross each other without significant hybridization and form the linear bands and Dirac-cone like points.13,14 There are no magnetic ions in LaAgBi2, but there are still two Bi2 atoms per unit cell because of the occupation of La ions (one above and another below the Bi2 layer), as shown in Fig. 1(b). This will lead to the folding of the dispersive p orbital of Bi2. The two px,y bands from two Bi2 atoms cross each other at a single point and then form the nearly linear band and Dirac- cone-like point around the Fermi level (Fig. 1(c)). Cor- respondingly the Fermi pockets along Γ-M (Fig. 2(d)), and the one along R− Z directions and at X points (Fig. 2(b)) host Dirac fermions. It is important to compare the similarity/difference dispersion between LaAgBi2 and of the Dirac (Sr/Ca)MnBi2. In (Sr/Ca)MnBi2, conduction and valence bands touch at the Dirac point in both mate- rials, but the details are different due to the different stacking of nearby alkaline earth atoms and the different hybridization. For SrMnBi2, the zero-energy gap is found only at a specific point, while it is found along the continuous line in the momentum space for CaMnBi2.20 Since the crystal structure of LaAgBi2 is identical to CaMnBi2, there is also continuous zero-energy gap line in LaAgBi2 which is different from SrMnBi2. But due to the different valence and hybridization between La and Ca ions as well as the nomagnetic Ag ions, some difference is expected in Fermi surfaces and Dirac dispersions. In CaMnBi2, the Dirac pocket (electrons) is only observed along Γ-X line,20 but in LaAgBi2 there are Dirac electron pockets along Γ-X and Γ− M lines (Fig. 1 and Fig. 2). In addition, the magnetic order of Mn ions is important to expel the states from the Fermi level, in contrast several regular parabolic bands cross the Fermi level in LaAgBi2 (Fig. 1(c)). These observations indicate that the magnetic order in (Sr/Ca)MnBi2 is not the key ingredient in the Dirac cone formation mechanism in 2D Bi layers and points out a direction for the search of new Dirac materials. Nevertheless, the 10 8 6 4 2 3 10 8 6 4 2 0 0 2 4 6 8 10 0 0 2 4 6 8 10 FIG. 1. (color online) (a) Powder XRD patterns and struc- tural refinement results. The data were shown by (+) , and the fit is given by the heavy solid line. The difference curve (the light solid line) is offset and the segments indicate the observed peaks. (b) The crystal structure of LaAgBi2. Atoms are distinguished by their size starting from La (largest) to Ag (smallest). Bi atoms in 2D square nets (Bi2) are somewhat smaller than Bi atoms in AgSb4 tetrahedra (Bi1). (c) and (d) The band structure for LaAgBi2 with (c) and without (d) spin-orbit coupling effect. The different bands were indicated by different color. The line at Energy=0 indicates the posi- tion of Fermi level. The red circles denote the position of the Dirac-cone-like points close to the Fermi level. magnetic order or electronic correlation can still remove other parabolic bands from the Fermi level. Linear bands and Dirac fermions have considerable ef- fects on the transport properties of materials. The in- plane resistivity ρab of LaAgBi2 single crystal is metal- lic in the whole temperature range with a significant anomaly at ∼ 30 K (Fig 3(a)). Similar behavior was observed in LaAgSb2 which was attributed to the possi- ble charge density wave (CDW) order and possibly im- plies the same order/transition in LaAgBi2. The exter- nal magnetic field significantly enhances the resistivity below 30 K, but has little influence on the transport be- havior above 30 K (Fig. 3(a)). The magnetoresistance of LaAgBi2 shows significant dependence on the field direc- tion (Fig 3(b,c)). The crystal was mounted on a rotating FIG. 2. (color online) (a) The total density of states (black line) and local DOS from La (dot line) (upper panel), Ag (the second panel),Bi1 (the third panel) and Bi2 (bottom panel) in LaAgBi2. The dotted line indicates the position of the Fermi energy. (b,c,d) The shape of three different Fermi pockets of LaAgBi2. stage such that the tilt angle θ between the crystal surface (ab-plane) and the magnetic field can be continuously changed with currents flowing in the ab-plane perpendic- ular to magnetic field, as shown in the inset of Fig. 3(c). Angular dependent magnetoresistance ρ(B, θ) at T ∼ 2 K is shown in Fig. 3(b) and (c). When B is parallel to the c-axis (θ = 0o, 180o), the MR is maximized and is linear in field for high fields. With increase in the tilt angle θ, the MR gradually decreases and becomes nearly negligible for B in the ab-plane (θ = 90o), as shown in Fig. 3(b). Angular dependent resistivity in B = 9 T and T = 2 K shows wide maximum when the field is par- allel to the c-axis (θ = 0o, 180o), and sharper minimum around θ = 90o, 270o (Fig. 3(c)). Hence, the Fermi sur- face of LaAgBi2 is highly anisotropic along ab-plane and c-axis. LaAgBi2 exhibits very large linear magnetoresistance. At 2 K, the MR is linear in the high field region and reaches ∼ 1200% in 9 T field (Fig. 4 (a)). This linear behavior extends to a very low crossover fields B∗ where the MR naturally reduces to a weak-field semiclassical quadratic response. In order to extrapolate the crossover field B∗, we plot the field derivative of MR, dMR/dB, in 4 1.2 0.9 0.6 R M 0.3 0.0 0.04 B d / R M d 0.03 (a) 12 10 8 R 6 M 4 2 0 1.2 (c) 0.8 B d / R M d 0.4 2 K 10 K 20 K 30 K (b) 40 K 60 K 80 K 100 K (d) B* B* 10 K 30 K 9 0 10 8 6 4 2 0 0 2 4 6 8 10 (color online) (a) In-plane resistivity ρab(T ) of FIG. 3. LaAgBi2 single crystal in 0 T and 9 T magnetic field. (b) In-plane Resistivity ρ vs. magnetic field B of LaAgBi2 crystal with different tilt angle θ between magnetic field and sample surface (ab-plane) at 2 K. (c) In-plane resistivity ρ vs. the tilt angle θ from 0o to 360o at B = 9 T and T = 2 K. Inset shows the configuration of the measurement. Fig. 4(b) and (c). In the low field range (B <1 T at 2 K), dMR/dB is proportional to B (as shown by lines in low- field regions), indicating the semiclassical MR ∼ A2B2. But above a characteristic field B∗, dMR/dB deviates from the semiclassical behavior and saturates to a much reduced slope (as shown by lines in the high-field region). This indicates that the MR for B > B∗ is dominated by a linear field dependence plus a very small quadratic term (MR= A1B + O(B2)). With increasing temperature, the field range where linear MR appears shrinks and MR decreases. Ultimately we cannot observe any linear MR below 9 T and above 100 K. The cross-over field B∗ is defined as the value where the fitting lines cross each other. Below 9 T and 50 K, the evolution of B∗ with temperature is parabolic (Fig. 5(a)). The linear MR which evidently deviates from semi- classical transport behavior in metal has been ob- served in bulk crystals of Ag2−δ(Te/Se),27,28 Bi2Te3,7,8,10 and (Sr/Ca)MnBi2.13 -- 16 Among them, BaFe2As2 Ag2−δ(Te/Se) and Bi2Te3 were found to be topologi- cal insulators hosting topological protected Dirac sur- face states,3,4,29 while BaFe2As2 and (Sr/Ca)MnBi2 host 9 0.0 0 3 6 B (T) 0.02 0.01 60 K 100 K 9 6 3 B (T) FIG. 4. (color online) (a) and (b) The magnetic field (B) dependence of the in-plane magnetoresistance MR at different temperatures. (c) and (d) The field derivative of in-plane MR, dMR/dB, as a function of field (B) at different temperature respectively. The solid lines in high field regions were fitting results using MR = A1B + O(B 2) and the lines in low field regions using MR = A2B 2. linear bands in the bulk.9,11 -- 13 The energy splitting be- tween the lowest and 1st LLs of Dirac fermions can be described by △LL = ±vF√2e¯hB where vF is the Fermi velocity.7,8,30,31 In the quantum limit △LL is larger than both the Fermi energy EF and the thermal fluctuations kBT . All carriers occupy the lowest Landau level and therefore the quantum transport with linear magnetore- sistance dominates the conduction. The critical field B∗ above which the quantum limit is satisfied at spe- cific temperature T is B∗ = 1 (EF + kBT )2.9 The B∗(T ) in LaAgBi2 can be well fitted by the above equa- tion, as shown by the solid line in Fig. 5(a). This con- firms the existence of Dirac fermion states in LaAgBi2. In a multiband system with dominant Dirac states and conventional parabolic-band carriers (including electrons and holes), the coefficient of the low-field semiclassi- cal B2 quadratic term, A2, is related to the effective 2e¯hv2 F  Data Fitting (a) 8 6 ) T ( * B 4 1.5 1.0 0.5 ) s V / 2 m c ( 2 R M 0.0 0 0 20 40 60 80 T (K) -0.5 (b) 0.9 1 0.6 A 0.3 0.0 0 20 40 60 80 T (K) FIG. 5. (color online) (a)Temperature dependence of the crit- ical field B ∗ (black squares) The solid line is the fitting results (EF + kBT )2. (b) The effective MR of B ∗ using B ∗ = 1 2e¯hv2 F mobility µM R (black squares) extracted from the weak-field MR and the fitting coefficient A1 for the linear term in MR. √σeσh σe+σh MR mobility √A2 = (µe + µh) = µM R (where σe, σh, µe, µh are the effective electron and hole conduc- tivity and mobility in zero field respectively). The ef- fective MR is smaller than the average mobility of car- riers µave = µe+µh and gives an estimate of the lower bound.9,10 Fig. 5(b) shows the dependence of µM R on the temperature. At 2 K, the value of µM R is about 1200 cm2/Vs in LaAgBi2 which is larger than the values in conventional metals. With increasing temperature, the linear MR, the linear term coefficient A1 and µM R are suppressed due to the temperature smearing of the Lan- dau level splitting. 2 In order to further clarify the electronic structure and Dirac fermions in LaAgBi2, we performed first principle Fermi surface calculations and dHvA oscillation measure- ment on LaAgBi2 single crystal. Fig. 2(b-d) shows the topology of the theoretical Fermi surfaces of three dif- ferent pockets for LaAgBi2, respectively. Centered at X point, there are very small ellipsoid electronic pock- ets with the long axis along Γ − Z directions, corre- sponding to the Dirac-cone-like point at X point in band structure.This Fermi pocket predicts the presence of a dHvA oscillation frequency of about 82 T. At the cen- ter of the Brillouin zone, there is a big hollow cylindrical hole pocket.This hole pockets is nearly ten times larger than previous ellipsoid pocket and predicts a frequency of ∼ 1000 T. These two pockets are nearly isotropic in ab-plane but different along c-axis. In addition, there is another complex electron pocket along diagonal direc- tion. This pocket is highly anisotropic along three axis, corresponding to the anisotropic point in Γ − M direc- tion in band structure. This complex pocket will give two 5 20 B (T) 30 (a) 0 (b) 0o 12o 22o 32o 42o 53o 63o 74o 84o 90o 10 2 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 ) V m ( e s n o p s e r r e v e l i t n a C ) s t i n u . b r a ( e d u t i l p m A T F F 50 100 150 200 250 300 350 400 450 Frequency (T) FIG. 6. (color online) (a) Quantum oscillation of LaAgBi2 observed with cantilever as a function of the magnetic field (B) with different tilting angle. (b) The Fourier transform spectrum of the SdH oscillation. For the peak labels see the main text. similar frequencies of 370 T and 320 T. It is of interest to note that the theoretical hole and electron Fermi surfaces are compensated, similar to semimetals. The quantum oscillation provide a direct probe of the Fermi surface. In metals, quantum oscillations corre- spond to successive emptying of LLs by the magnetic field and the oscillation frequency is related to the cross section area of the Fermi surface SF by the Onsager re- lation F = (¯h/2πe)SF .7,8,32 From the temperature evo- lution of the oscillation amplitude, one can deduce the effective cyclotron resonant mass by the fitting using Lifshitz-Kosevitch formula.32 From the evolution of these frequencies as a function of the magnetic field orienta- tion (θ) and temperature, one can construct a detailed picture of the size and shape of the Fermi surface. Fig. 6(a) and (b) show the typical dHvA oscillations and the Fourier transform (FFT) spectrum of the oscillations for LaAgBi2 single crystal with different magnetic field direc- tion. When the magnetic field is close to perpendicular to the ab plane (θ close to 0o), the signal exhibits signif- icant oscillation (Fig. 6(a)). The FFT spectrum of the (a) -10 0 10 20 40 (Degree) 30 50 60 70 (b) F ~ 67 T F ~ 292 T F ~ 323 T ) T ( ) ( s o c * F 350 300 250 200 150 100 50 ) s t i n u . b r a ( e d u t i l p m A T F F 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 0 10 20 30 40 50 60 70 80 90 T (K) FIG. 7. (color online) (a) The evolution of the de Hass-van Alphen oscillation frequencies F plotted as F ∗cos(θ) with the magnetic field angle θ. (b) Temperature dependence of the oscillation amplitudes (Osc. Amp.) in quantum oscillation using cantilevert. The discrete symbols are the experimen- tal results and the solid lines are the fitting results giving cyclotron mass. oscillations (Fig. 6(b)) exhibit two peaks at ∼ 67 and 300 T which are labeled as α and β. There is another peak at 135 T which corresponds to the double frequency of peak α. With increasing θ, oscillations are weaker and the peak α shifts to higher frequency. More interestingly, the peak β splits to two peaks (labeled as β and η). One of them (β) shifts to lower frequency but another (η) shifts toward higher frequency with increasing angle. The detailed angular dependence of the three FFT fre- quencies in the oscillation are shown in Fig. 7(a). For a 2D FS (a cylinder), the cross section has SF (θ) = S0/ cos(θ) angular dependence and the oscillation fre- quencies should be inversely proportional to cos(θ). For a sphere Fermi pocket, the cross section for all magnetic field direction is a constant and the oscillation frequen- cies should not be angular dependent.7,32 In Fig. 7(a), the angular dependence of the oscillation frequency α (∼ 67 T at θ = 0) multiplied by cos θ (open squares) does not show significant dependence on the field angle θ, indicating a quasi-2D Fermi pocket. The tempera- ture evolution of the oscillation amplitude gives a cy- clotron mass m ∼ 0.056me where me is the bare electron 6 mass (Fig. 7(b)) and similar to previous observation.21 The oscillation frequency and mass is close to the el- lipsoid electron pocket at X point. Other two pock- ets β and η should not come from the hollow cylin- der hole pocket at the center of the Brillouin zone, be- cause the hole pocket is nearly one hundred times big- ger than the ellipsoid electron pocket but the oscilla- tion frequencies for β and η is only ten times bigger the α. Moreover, these two pockets show similar effective mass (mβ ∼ 0.14(2)me, mη ∼ 0.16(5)me) (Fig. 7(b)). These two oscillation frequencies should correspond to the highly anisotropic pockets locating along the edge and diagonal directions in Brillouin zone (Fig. 2(d)) since the oscillation frequencies were close to the the- oretical values. but in our in-plane measurement, the magnetic field is along [100] direction. Since we cannot change the in-plane field direction, we can not distinguish these two pockets from the angular dependent oscillation frequency in present measurement. For pocket η, the quantum oscillation frequency, F × cos θ decreases a lit- tle bit with increasing θ, but its change is much smaller when compared to pocket β. This is consistent with the calculated Fermi surface (Fig. 2(d)), which shows that the dispersion along kz direction is much smaller than the value along kx and ky directions. Besides that, the Dirac fermions with linear bands have much larger mo- bility and Fermi velocity than the regular carriers, and will dominate the transport properties. Hence, the an- gular dependent magnetoresistance (Fig. 3(b) and (c)) should come from the quasi-two-dimensional Dirac Fermi pockets. Our results demonstrate the possible universal exis- tence of two dimensional Dirac fermions in layered struc- ture compounds with two-dimensional Bi square nets, irrespective of magnetic order. So it is important to study the relationship between the CDW transition and the Dirac fermions. For the systems with charge den- sity wave or spin density wave, the phase transition of- ten induces band-folding of some Fermi surface sections. CDW was found in LaAgSb2 and LaSb2, where calcu- lated Fermi surfaces without CDW transition agree very well with low-temperature quantum oscillation results in CDW state.33 -- 35 In LaAgBi2, the observed three frequen- cies are consistent with the calculated Fermi surfaces along Γ-M and Γ-X directions (Fig. 2(b) and (d)) af- ter a moderate energy shift (∼ 20 meV), but the large hole pockets centered at Γ point is absent in the dHvA os- cillation, which is similar to the results in LaAgSb2.33,35 Hence, the CDW transition most likely smears out the large hole pocket and induces the shift of the energy of other three pockets which host Dirac fermions. The x-ray scattering experiments in LaAgSb2 revealed that the nesting of the Fermi surfaces responsible for the two CDWs happens in band 1 centered at Γ point and band 3 extending throughout the zone with the vertices of the squares at the X point. The Fermi surfaces of LaAgBi2 should be similar to these of LaAgSb2 because of the sim- ilar structure. So it could be expected that the nesting happens in the band centered at Γ point (Fig. 2(c)) in LaAgBi2. The band α of LaAgBi2 centered at X point (Fig. 2(a)) which hosts Dirac fermions remains unaf- fected by the nesting or CDW transition. The band 3 in LaAgSb2 separates to two bands (β, η) (Fig. 2(d)) in LaAgBi2 which lose the nesting condition. So the bands which host Dirac fermions in LaAgBi2 are most likely intact at the CDW transition. IV. CONCLUSION In summary, first-principle calculation, de Hass-van Alphen oscillation study of the electronic structure and magnetoresistance behavior of LaAgBi2 show striking similarity to properties of (Sr,Ca)MnBi2. LaAgBi2 has no magnetic ions and is a paramagnetic metal without long range magnetic order. Yet, the band structure clearly shows several narrow bands with nearly linear en- ergy dispersion and Dirac-cone-like points at the Fermi level. This is in agreement with the quantum oscillation experiments that revealed three Fermi pockets with small 7 cyclotron resonant mass. The in-plane transverse magne- toresistance exhibits a crossover at a critical field B∗ from semiclassical weak-field B2 dependence to the high-field unsaturated linear magnetoresistance which is a hallmark of the quantum limit of the Dirac fermions. The tem- perature dependence of B∗ satisfies quadratic behavior, which is attributed to the splitting of linear energy dis- persion in high field. Our results demonstrate the possi- ble universal existence of two dimensional Dirac fermions in layered structure compounds with two-dimensional Bi square nets, irrespective of magnetic order. ACKNOWLEDGMENTS We than John Warren for help with SEM measure- ments. Work at Brookhaven is supported by the U.S. DOE under contract No. DE-AC02-98CH10886. The high magnetic field studies in NHMFL were supported by NSF DMR-0654118, the State of Florida and DOE NNSA DE-FG52-10NA29659. 1 A. H. Castro Neto, F. Guinea, N. M. R. Reres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 2 D. S. L. Abergel, V. Apalkov, J. Berashevich, K. Ziegler, and T. Chakraborty, Adv. Phys. 59, 261 (2010). 3 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 15 K. Wang, D. Graf, H. Lei, S. W. Tozer, and C. Petrovic, Phys. Rev. B 84, 220401(R) (2011). 16 K. Wang, D. Graf, L. Wang, H. Lei, S. W. Tozer, and C. Petrovic, Phys. Rev. B 85, 041101(R) (2012). 17 A. Kobayashi, et al. J. Phys. Soc. Jpn. 76, 034711 (2007). 18 E. Brechtel, G. Cordier and H. Schaffer, Z. Naturforsch. B (2010). 35, 1 (1980). 4 X. L. Qi and S. C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). 5 A. A. Abrikosov, Fundamentals of the Theory of Metals (North-Holland, Amsterdam, 1988). 19 P. Dietl, F. Pi´echon, and G. Montambaux, Phys. Rev. Lett. 100, 236405 (2008). 20 Geunsik Lee, Muhammad A. Farhan, Jun Sung Kim, Ji 6 Y. Zhang, Z. Jiang, Y.-W. Tan, H. L. Stormer, and P. Kim, Hoon Shim, arXiv:1301.1087v1 (2013). Nature 438, 201 (2005). 21 C. Petrovic, S. L. Bud'ko, J. D. Strand, and P. C. Canfield, 7 D.-X. Qu, Y. S. Hor, J. Xiong, R. J. Cava, and N. P. Ong, J. Magn. Magn. Mater. 261, 210 (2003). Science 329, 821 (2010). 8 J. G. Analytis, R. D. McDonald, S. C. Riggs, J.-H. Chu, G. S. Boebinger, and I. R. Fisher, Nature Phys. 6, 960 (2010). 9 K. K. Huynh, Y. Tanabe, and K. Tanigaki, Phys. Rev. Lett. 106, 217004 (2011). 10 H.-H. Kuo, J.-H. Chu, S. C. Riggs, L. Yu, P. L. McMa- hon, K. D. Greve, Y. Yamamoto, J. G. Analytis, and I. R. Fisher, Phys. Rev. B 84, 054540 (2011). 11 P. Richard, K. Nakayama, T. Sato, M. Neupane, Y.-M. Xu, J. H. Bowen, G. F. Chen, J. L. Luo, N. L. Wang, X. Dai, Z. Fang, H. Ding, and T. Takahashi, Phys. Rev. Lett. 104, 137001 (2010). 12 T. Morinari, E. Kaneshita, and T. Tohyama, Phys. Rev. Lett. 105, 037203 (2010). 13 J. Park, G. Lee, F. Wolff-Fabris, Y. Y. Koh, M. J. Eom, Y. K. Kim, M. A. Farhan, Y. J. Jo, C. Kim, J. H. Shim, and J. S. Kim, Phys. Rev. Lett. 107, 126402 (2011). 14 J. K. Wang, L. L. Zhao, Q. Yin, G. Kotliar, M. S. Kim, M. C. Aronson, and E. Morosan, Phys. Rev. B 84, 064428 (2011). 22 M. Weinert, E. Wimmer, and A. J. Freeman, Phys. Rev. B 26, 4571 (1982). 23 P. Blaha, K. Schwarz, G. K. H. Madsen, D. Kvasnicka and J. Luitz, WIEN2k, An Augmented Plane Wave + Local Orbitals Program for Calculating Crystal Properties (Karl- heinz Schwarz, Techn. Universitat Wien, Austria), 2001. ISBN 3-9501031-1-2 24 J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). 25 B. Hunter, "RIETICA - A Visual RIETVELD Prog- arm," International Union of Crystallography Commission on Powder Diffractin Newsletter No. 20 (Summer), 1998 (http://www.rietica.org). 26 H. Flandorfer, O. Sologub, C. Godart K. Hiebl, A. Leithe- Jasper, P. Rogl, and H. Noel, Solid State Commun. 97, 561 (1996). 27 R. Xu, A. Husmann, T. F. Rosenbaum, M.-L. Saboungi, J. E. Enderby, and P. B. Littlewood, Nature 390, 57 (1997). 28 M. Lee, T. F. Rosenbaum, M.-L. Saboungi, and H. S. Schnyders, Phys. Rev. Lett. 88, 066602 (2002). 29 W. Zhang, R. Yu, W. Feng, Y. Yao, H. Weng, X. Dai, and Z. Fang, Phys. Rev. Lett. 106, 156808 (2011). 30 Y. Zhang, Z. Jiang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature 438, 201 (2005). 31 D. Miller, K. Kubista, G. Rutter, M. Ruan, W. de Heer, P. First, and J. Stroscio, Science 324, 924 (2009). 32 D. Shoeneberg, Magnetic oscillation in metals (Cambridge University Press, Cambridge, 1984). 33 K. D. Myers, S. L. Budko, V. P. Antropov, B. N. Harmon, P. C. Canfield, and A. H. Lacerda, Phys. Rev. B 60, 13371 (1999). 8 34 R. G. Goodrich, D. Browne, R. Kurtz, D. P. Young, J. F. DiTusa, P. W. Adams, and D. Hall,Phys. Rev. B 69, 125114 (2004) 35 Y. Inada, A. Thamizhavel, H. Yamagami, T. Takeuchi, Y. Sawai, S. Ikeda, H. Shishido, T. Okubo, M. Yamada, K. Sugiyama, N. Nakamura, T. Yamamoto, K. Kindo, T. Ebihara, A. Galatanu, E. Yamamoto, R. Settai, and Y. Onuki, Philosophical Magazine 82, 1867 (2002) 36 S. Song, J. Park, J. Koo, K. -B. Lee, J. Y. Rhee, S. L. Bud'kov, P. C. Canfield, B. N. Harmon, and A. I. Goldman, Phys. Rev. B 68, 035113 (2003). 

1709.05783

1

1709

Exact solution for many-body Hamiltonian of interacting particles with linear spectrum

The exact solution of the Schr\"odinger equation for the one-dimensional system of interacting particles with the linear dispersion law in an arbitrary external field is found. The solution is reduced to two groups of particles moving with constant velocities in the opposite directions with a fixed distance between the particles in each group. The problem is applied to the edge states of the 2D topological insulator.

cond-mat.mes-hall

cond-mat

a Exact solution for many-body Hamiltonian of interacting particles with linear spectrum M.V. Entin1, 2 and L. Braginsky1, 2 1Rzhanov Institute of Semiconductor Physics, Siberian Branch of the Russian Academy of Sciences, Novosibirsk 630090, Russia 2Novosibirsk State University, Novosibirsk 630090, Russia∗ The exact solution of the Schrodinger equation for the one-dimensional system of interacting particles with the linear dispersion law in an arbitrary external field is found. The solution is reduced to two groups of particles moving with constant velocities in the opposite directions with a fixed distance between the particles in each group. The problem is applied to the edge states of the 2D topological insulator. Introduction One-dimensional electron systems with the linear dis- persion are the topical issue now: the edge states1-3 of the 2D topological insulator4, graphene strips, carbon nanotubes5-7, etc. The specificity of these systems is the presence of the double degenerate state at the zero longi- tudinal momentum that has a linear splitting at a finite momentum. The 1D systems with near-linear spectrum are the sub- ject of study in the theory of bosonization and Luttinger liquid8. This approach considers the electron Fermi liq- uid of strongly-interacting electrons assuming their en- ergy spectrum to be approximately linear near the Fermi energy. The bosonization procedure separates the long- (q ≪ pF ) and short-range (q ≈ 2pF ) interactions consid- ering them in different ways. In general, this approach is approximative and applicable near the Fermi points only. Note that the curvature of the energy spectrum caused by the k-p expansion violates the linearity. For example, such non-linear corrections to the energy spectrum ex- ist in graphene where they determine the unconventional character of the e-e scattering9 and e-h coupling to the excitons10. The linear spectrum of the 2D TI edge states is the main reason of disappearance of the electron cor- relation energy13. In11 we have studied the linearity of the energy spec- trum in the edge states of the 2D topological insulator. We concluded that in two models of the edge states12 and1-3 the linearity is absolute, while in other cases the non-linear corrections are extremely weak. This pushes forward the problem of the many-electron states in the system with the linear single-electron energy spectrum. The purpose of the present paper is the general con- sideration of the one-dimensional system with the many- body Hamiltonian H = Xi ( 1 ¯h vσipi + U (xi)) +Xi<j V (xi − xj ). (1) Here U (x) is an external field, and V (xi − xj) is the interaction between the particles, [xi, xj] = 0, [pi, pj] = 0, [pi, xj] = −i¯h. For certainty we consider σi = ± (or equivalently, σi =↑, ↓) as a spin quantum number. Below we shall set ¯h = 1. The coordinates of spin-up and spin- down electrons are denoted as xi and yi, accordingly. The Hamiltonian Eq.(1) is valid for the edge states of electrons in 2D topological insulators11. The linearity of the energy spectrum is the most impor- tant for our consideration. We obtain an exact solution of this quantum problem. Unlike the case of the Lut- tinger liquid, we do not need low temperatures and the vicinity of the energy to the Fermi level. Owning to the exact linearity of the spectrum our results are valid for all energies. Exact solution of many-body Schrodinger equation In the absence of interaction the direction of motion coincides with the spin sign, so that the spin-up and spin-down electrons are rightmovers and leftmovers, cor- respondingly. The exact solution of the Schrodinger equation for a single particle with energy E = pv is ψ(x) = exp(ipx − iR dxU (x)/v). This is the wave func- tion of constant density ψ2. The result essentially differs from that for a particle with a quadratic kinetic energy. Consider now two free (U (x) = 0) particles with posi- tive spins. The Hamiltonian H = vp1,↑ + vp2,↑ + V (x1 − x2) commutes with x1 − x2 conserving, therefore, the dis- tance between the particles. Then the wave function can be chosen as the eigenfunction of x1 − x2 and the total momentum P = p1,↑ + p2,↑: ψ = A exp(iP (x1 + x2)/2)δ(x1 − x2 − a), where a is the eigenvalue of x1 − x2. The corresponding selfenergy is E = P v + V (a). The normalizing coefficient A should be chosen to ex- clude the divergence of the integral R dxδ2(x). This di- vergence inevitably appears when one uses the selffunc- tions of the coordinate operator. It can be formally fixed by the choice A2 = 1/δ(0). For the case of n rightmovers we write ψ(x1, ....xn) = An exp(iP X)(Πkδ(xk+1 − xk − ak)), (2) where X = Pi xi/n, P = Pi pi,↑ is the total momentum. The selfenergy is E = P v +Xi<j V (aij ) , aij = j−1 Xi ak. (3) impurity, potential U (xi) To include an external, e.g. into consideration, we multiply the wave function Eq. (2) by the factor u = exp(−iZ dx1Xj U (x1 + aj)). The antisymmetry of the coordinate part of the wave function can be achieved by means of the Slatter deter- minant composed of the functions fi,j = δ(xi − xj − ai,j). Thus, the wave function of the n right- (left-)movers is ψ(x1, ....xn) = u exp(iP X)det(fi,j). (4) Consider now two particles of the opposite spins with the Hamiltonian: H = vp↑ − vp↓ + V (x − y). The wave function can be chosen as the eigenfunction of x + y with selfvalue 2c: ψ = exp(cid:18)iP ′(x − y)/2 − iZ dxV (2b − 2x)/v(cid:19) × Aδ(x + y − 2c). (5) Here P ′ = p1 − p2, the corresponding selfenergy is E = P ′v. Finally, consider a general case of n spins up and m spins down. The Hamiltonian commutes with the central point between each leftmover and rightmover (x1 +y1)/2. The sufficient condition is a fixation of one of these vari- ables, e.g., x1 + y1 = 2c. Hence, the state with the quan- tum numbers P, c, {ak}, {bk} is P, c, {a1, ...an−1}, {b1, ...bm−1}i = ψ(x1, ..xn; y1, ...ym) = Φ exp(iP ζ)Aδ(y1 + x1 − 2c) × n−1 m−1 Aδ(xk+1 − xk − ak) × Yk Yk Aδ(yk+1 − yk − bk)),(6) where ζ = (x1 − y1)/2, m > 1, n > 1. If m = 1 or n = 1, the corresponding product in Eq.(6)should be replaced by unity. Substituting into the Schrodinger equation, we find the proportionality factor n,m V(x1) = Φ = exp(cid:16) − Xi,j Xi=1 Xk=i U(x1) = ai,j = j−1 n ak, 2 The corresponding energy is E = P v + U , U = Xi>j U (ai,j) +Xi>j U (bi,j) . (9) The case of an arbitrary number of identical electrons with different spins should be considered using the per- mutation symmetry and Young diagrams. The ground state of the system corresponds to the equal numbers of up and down spins, so that the total spin is zero. Note that in this state the average velocity vanishes. Thus, the spin wave function is symmetric with respect to all particles, and the coordinate wave function should be antisymmetric with respect to all coordinates. To con- struct the appropriate coordinate wave function, we need to antisymmetrize the wavefunction. Neglecting the particle exchange, each of subsystems of right- and leftmovers represents a solid superparti- cle, inside which the distances between the electrons are fixed. The right and left moving superparticles (RMS) and (LMF) obey the linear dispersion. The Hamiltonian of the RMS and LMS interaction depends on the distance between the electrons belonging to the different super- particles. Thus, RMS and LMS can be considered as two opposite moving superparticles and the two-particle wave function can be used to describe their relative mo- tion. This is an explanation of Eqs. (6) and (7). Cyclic boundary conditions In the previous consideration we assumed that the co- ordinates change in the infinite domain −∞ < xi < ∞. In this case the energy is expressed via the total and rel- ative momenta of the left- and right-movers (9) and the interaction energies inside the groups. It is important to point out that the interactions between the carriers from different groups as well as the carriers with the impurities do not contribute to the total energy. Actually, this is consequence of the problem formula- tion. We have considered an infinite system with the finite number of interacting electrons. In this case elec- trons with different spins being separated at infinite dis- tance can be characterized by their momenta at the in- finity. In a dense system, however, this is not the case. Consider now a cyclic system of the length L assuming xi + L = xi. Suppose the potentials V (x) and U (x) to be the periodic functions of L. For this reason we replace x by the distance between points on the circle x → (L/2π) sin(2πx/L). In particular, the δ-functions in the previous expressions have to be replaced by their periodic generalizations δ(x) → (2π/L)δ(sin(2πx/L)). The cyclic boundary condition reads ψ(x1, ...xi + L, ...) = ψ(x1, ...xi, ...). Consider first the two-particle problem with the opposite spins. In this case the quan- tization rule Eq. (3) is (7) i v Z x1 dx1(V(x1) + U(x1)(cid:17) V(cid:16)2x1 − 2c + ai,1 − bi,1(cid:17) U(cid:16)x1 + ai,1(cid:17) − Xi=1 m U(cid:16)2c − x1 + bi,1(cid:17) bi,j = j−1 Xk=i bk. (8) E = 2πvN L + 1 L Z L 0 (V (x) + 2U (x)) dx, where N is an integer. The second term here is the av- erage interaction, which has to be added to the total energy. The generalization of the quantization rule to many particles is n,m E = 2πvN (n − m)/L + V (sσ,i,j) + Xσ=±,i,j V (x)dx + (n + m)Z L 0 nmZ L 0 3 magnetic impurity scattering. The transition between the states occurs at the terms crossing points (when the total energies of two states are equal at coinciding elec- tron positions). Consider an impurity whose spin S interacts with the electron spin σi/2. The Hamiltonian of spin-spin inter- action is U (x)dx. (10) Hss = U0Xn δ(xn)(Sσn) (11) In accordance with Eq.(10), the total energy incorporates the intra-group interaction and averages of the external field and inter-group interaction. The physical meaning can be simply understood: electrons with same spins con- serve the distance between each other (and, consequently the sum of the potentials in Eq.(10)) and move through the impurity lattice and the electrons with opposite spins, averaging the interaction with them (the second line in Eq.(10)). It is clear from Eq.(10) that the electron density is not affected by an external potential. This explains the absence of the correlation energy13 for the system with the linear spectrum. Note that P , ak, bk and c compose the full set of the numbers describing the system state. Variable P is a quantum number. It is quantizing in a closed edge the same way as the non-interacting particles momenta. The distances between the same-spin particles ak, bk and the quantity c are the classical variables. This can be seen from the Hamiltonian which corresponds to the limit ¯h → 0. The set of classical variables have arbitrary values; they have infinite masses and are resting. To some extent, this situation reminds the molecular sys- tems where the electron coordinates are quantum quan- tities while the ion coordinates are classical. It is known that the molecular system can be considered via the molecular terms: the electronic levels are determined at fixed arbitrary positions of the ions, while the motion of the latter is considered classically where the terms, play the role of the interaction potential. (This description is limited by crossing of the molecular terms). The relative momentum P is a global variable. Thus, the quantities ak, bk and c obey the Boltzmann statistics. This explains how to make average of the observable quantities. Scattering of interacting electrons at a magnetic impurity The exact solution permits one to include perturbingly other interaction mechanisms that can affect the re- sponses, for example, an interaction with magnetic impu- rities or the spin-orbit interaction with phonons. Here, as an example, we consider backscattering of the elec- trons by a magnetic impurity. The backscattering is for- bidden without such an interaction violated the time- reversibility. The e-e interaction essentially modifies the For the two-electron system we find the matrix elements M = U0/2ei(P +P ′)aδ(a − 2b) and the transition rate T = U 2 0 V (x) dx is a correction to the total energy due to interaction. In the general case, the matrix element between the wave 0 /4δ[E −(P ′−P )v−δV ]. Here δV = V (a)−R L functions (6) is equal to M = U0/2Pij ei(P +P ′)aij δ(aij − Pij Vij − (n − m + 1)R L 0 /4δ[E − (P ′ − P )v − δV ], where δV = 0 V (x) dx includes correction to the interaction energy Vij after transition of the electron from left- to rightmover ensemble or vice-versa. 2b) and T = N U 2 the Now study the many-body problem. states 2i transition between the sider P, c, {a1, ..an−1}, {b1, ...bm−1}i P ′, c′, {a2, ...an−1}, {a1, b1, ...bm−1}i. tering rate at fixed quantum numbers is Con- 1i = = The backscat- and 2πh1Hss2i2 × Xj=1 δ(vP − vP ′ + n−1 m−1 Xj=1 U (aj) − n−1 U (bj)) = 2πU02δ(vP − vP ′ + Xj=1 δ(aj + 2c)δ(bk + 2c′) × Xj,k U (aj) − m−1 Xj=1 U (bj)) (12) Averaging with respect to c, c′ gives δ(aj + 2c)δ(bk + 2c′)i = (n − 1)(m − 1)/L2. hXj,k The physical meaning of Eq.(12) is simple. An electron changing its spin simultaneously changes its interaction energy with all electrons with the same spins to the in- teraction energy with the opposite-spin-electrons. This difference of potential energies is transmitted to the dif- ference of the kinetic energies establishing the thermal equilibrium between the kinetic and potential energies. Classical variables obey the Boltzmann statistics in the thermal equilibrium. Let us now average the energy delta-function over this distribution: R = R Q daidbj exp(−β U )δ(...) R Q daidbj exp(−β U ) . (13) In the nearest-neighbor approximation we obtain R = Z ∞ −∞ dt 2π eitv(P −P ′) exp(cid:26) 1 L Z dahn(e(it−β)U(a) − 1) + m(e(−it−β)U(a) − 1)i(cid:27) (cid:30) exp(cid:26) n + m L Z dahe−βU(a) − 1i(cid:27) 4 Eq. (13) gives a symmetric dependence of the transition probability on P − P ′, which is determined by the func- tion U (a). A careful examination goes beyond the scope of the paper. Conclusions In conclusion, we have found the exact solution of the interacting many-particle 1D system with the lin- ear single-particle spectrum. The Hamiltonian includes an external potential (e.g., impurities) as well. The Schredinger equation solution is reduced to the separa- tion of the system to the groups of right- and left-moving carriers with the constant velocity. The interactions be- tween groups and with impurities are reflected in the phase factor in the wave function. The relative coor- dinates in each group turn out to be classical conserv- ing variables, while the relative momentum of all carriers is a global quantum number. The collective selfenergy consists of the linear in momentum kinetic part and the potential energy of interaction at a fixed interparticle dis- tance inside the groups. In the framework of the Hamil- tonian Eq. (1) the backscattering is absent. The formal solution is applicable to the edge states of the 2D topolog- ical insulator. In a separate paper11 we have found that the edge states have either an exact linear single-electron spectrum in the most models of the 2D topological insu- lator or this spectrum is numerically linear. Hence, the results of the present paper directly pertain to these edge states. Unlike the Luttinger liquid, our solution is not linked It is to the Fermi level of the non-interacting system. also valid in a strongly non-equilibrium situation. The conservation of the distances between the same- spin electrons makes relaxation of such system to the equilibrium impossible, unless some additional term are taken into account. In accordance with the obtained equations the imple- mentation of the e-e and electron-impurities interactions has no effect on the velocity matrix elements. That means that the conductivity of the system is also not changed and stay infinite for the system of interacting electrons. The exact solutions permitted one to include per- turbingly the interaction with the magnetic impurities that was not included in the Hamiltonian (1). It was found that the e-e interaction essentially affects the backscattering. Note that other mechanisms, like the spin-orbit interaction with phonons, can be studied in the same way. One other remark concerns possibility of the gener- It obviously can be alization of the Hamiltonian (1). generalized to (vσipi + Ui(xi) + σiU (1) i (xi)) + H = Xi Xi<j (Vij (xi − xj) + σiσj V (1) ij (xi − xj )), (14) with similar consequences. Acknowledgements This research was supported by RFBR grant No 17-02- 00837. The authors thank A.V. Chaplik for stimulating discussions. ∗ Electronic address: [email protected] 1 Xiao-Liang Qi, Shou-Cheng Zhang, Rev.Mod.Phys., 83, 1057 (2011). 2 Bin Zhou, Hai-Zhou Lu, Rui-Lin Chu, Shun-Qing Shen, and Qian Niu, Phys.Rev.Lett., 101, 246807 (2008). 3 M.Konig, H.Buhmann, Laurens W. Molenkamp, T. Hughes, Chao-Xing Liu, Xiao-Liang Qi, Shou-Cheng Zhang, J.Phys.Soc.Jpn., 77, 031007 (2008). 4 B.Andrei Bernevig, Taylor L.Hughes, Shou-Cheng Zhang, 8 Alexei M.Tsvelik. Quantum Field Theory in Condensed Matter Physics 2nd Edition Cambridge University Press, Cambridge UK(2003) 9 L. E. Golub, A. Tarasenko M. V. Entin and L. I. Magarill, Phys. Rev. B, 84 195408 (2011). 10 Europhysics Lett. 102,3712 (2013). 11 M. V. Entin, M. M. Mahmoodian, and L. I. Magarill, Eu- rophys. Lett. 118, 57002 (2017). 12 B.A. Volkov, O.A. Pankratov Pis'ma Zh. Eksp.Teor. Fiz. Science, 314, no. 5806, 1757 (2006). 42,145(1985) [JETP Lett,42, 178(1985)]. 5 D.S. Miserev, M.V. Entin Zh. Eksp. Teor. Fiz. 142, 13 M.V. Entin and L. Braginsky, Phys. Rev. B, 96 115403 784(2012)[JETP 115, 694 (2012)]. 6 D.S. Miserev, M.V. Entin JETP Letters 99, 410 (2014). 7 D.S. Miserev, JETP 149, 1223 (2016) (2017). 

1204.5206

2

1204

Fast Two-Qubit Gates in Semiconductor Quantum Dots using a Photonic Microcavity

Implementations for quantum computing require fast single- and multi-qubit quantum gate operations. In the case of optically controlled quantum dot qubits theoretical designs for long-range two- or multi-qubit operations satisfying all the requirements in quantum computing are not yet available. We have developed a design for a fast, long-range two-qubit gate mediated by a photonic microcavity mode using excited states of the quantum dot-cavity system that addresses these needs. This design does not require identical qubits, it is compatible with available optically induced single qubit operations, and it advances opportunities for scalable architectures. We show that the gate fidelity can exceed 90% in experimentally accessible systems.

cond-mat.mes-hall

cond-mat

Fast Two-Qubit Gates in Semiconductor Quantum Dots using a Photonic Microcavity Dmitry Solenov,1 Sophia E. Economou,1, ∗ and T. L. Reinecke1 1Naval Research Laboratory, Washington, District of Columbia 20375, USA Implementations for quantum computing require fast single- and multi-qubit quantum gate op- erations. In the case of optically controlled quantum dot qubits theoretical designs for long-range two- or multi-qubit operations satisfying all the requirements in quantum computing are not yet available. We have developed a design for a fast, long-range two-qubit gate mediated by a photonic microcavity mode using excited states of the quantum dot-cavity system that addresses these needs. This design does not require identical qubits, it is compatible with available optically induced single qubit operations, and it advances opportunities for scalable architectures. We show that the gate fidelity can exceed 90% in experimentally accessible systems. I. INTRODUCTION Quantum information processing involves the manip- ulation of entanglement carried out by unitary gate op- erations between different quantum bits (qubits). Re- alistic quantum computing architectures require entan- gling gates between distant qubits. Optical photons pro- vide a natural vehicle to implement such interactions in many physical systems.1 As a result, architectures based on optically active qubits that can couple to photonic modes in optical cavities and waveguides, such as quan- tum dots, NV centers, and trapped ions are attractive for large scale quantum computing.2 -- 5 Quantum dots (QDs) in particular hold promise as qubits for such ar- chitectures, in part owing to their large dipole moments, which allow them to couple efficiently to the optical cav- ity modes and to photonic flying qubits for extended ar- chitectures. Qubits encoded by the spin of an electron in a QD have long coherence times which are five to six orders of magnitude longer than the typical picosecond scale of optical control. Successful initialization and read- out, as well as fast optical single spin rotations, have been demonstrated in these systems.6,7 In addition, im- portant advances have recently been achieved in work on coupled cavity-QD systems, including demonstrations of strong coupling and tunability,5,8 -- 15 and very recently full single-qubit control.16 A critical step needed to advance the field is the de- sign of a two-qubit controlled gate operation mediated by an optical cavity mode. A viable two-qubit quantum gate requires that several criteria are met: (i) a long- range switchable physical interaction between qubits is available; (ii) the gate performs a unitary operation on one qubit depending on the state of the other qubit to provide a controlled operation; (iii) the operations are sufficiently fast compared to decoherence rates; (iv) the gate is compatible with single-qubit rotations (to form a universal set of gates); (v) the gate design is consistent with a multi-qubit system for scalability. So far, only local control of entanglement in closely spaced quantum dots (QD 'molecules') has been demon- strated experimentally.17 For an experimental demon- stration of cavity-mediated entangling gates, a theoreti- cal design is needed that satisfies the above criteria, (i)- (v), while being experimentally simple and compatible with current technology. Existing proposals for cavity- mediated gates have not met these requirements; they are either incompatible with single-qubit gates,18 limited to nearest-neighbor qubits,19 and/or require adiabaticity, either through adiabatic evolution19 or through adiabatic elimination of the auxiliary state.20 As a result, they are much slower than what is needed from a quantum infor- mation processing perspective. Moreover, a careful as- sessment of the performance of such gates as a function of system parameters has not been given in the litera- ture, despite the key role it would play in experimental demonstrations. In this paper we give a novel design for an entan- gling control-z (CZ) two-qubit gate21 that satisfies all the above criteria. Our design does not require the QD ener- gies to be equal or dynamically tunable. As a result, our approach is compatible with single qubit operations and has a potential for many-qubit scalable architectures. We obtain fidelities in excess of 90% for realistic parameters. In the following we explain the concept of this all-optical gate, formulate the model, calculate the QD-cavity sys- tem spectrum, and analyze our design of the two-qubit gate protocol. The fidelity of the gate operation as a function of the system parameters is also calculated and provides a guide for experiment. II. TWO-QUBIT GATES The control-z gate is a maximally entangling two-qubit gate, and it is given by UCZ = diag(1, 1, 1,−1). It is equivalent to the more familiar control-NOT (CNOT) op- √ eration up to single-qubit gates. Specifically, UCN OT = (1⊗H)UCZ(1⊗H), where H = (1 1 1 −1)/ 2 is the Hadamard gate. To see the entangling capability of CZ we can look at its action on a product state of two qubits. In partic- ular, when each qubit is in an equal superposition of the basis states, we have (cid:0)1(cid:105) + 0(cid:105)(cid:1) ⊗(cid:0)1(cid:105) + 0(cid:105)(cid:1) = 11(cid:105) + 10(cid:105) + 01(cid:105) − 00(cid:105), UCZ which is a maximally entangled two-qubit state, also known as a two-qubit 'cluster state'. Such a state is equivalent to a Bell state up to single-qubit rotations. 3 1 0 2 n a J 2 2 ] l l a h - s e m . t a m - d n o c [ 2 v 6 0 2 5 . 4 0 2 1 : v i X r a To implement the CZ gate, we need to accumulate a phase factor of −1 selectively to only one of the two-qubit basis states, taken to be 00(cid:105) above. Meanwhile, to be able to perform single-qubit gates, the transition involv- ing state 00(cid:105) and an auxiliary state should be performed in parallel with that involving 01(cid:105) (or 10(cid:105) for rotations of the second qubit) and its corresponding auxiliary state. To avoid dynamically tuning energies -- a process that is costly in time and can compete with coherence times -- we will use different classes of auxiliary states for single- qubit and two-qubit operations. In particular, we will use a near-resonance between the two-photon state of the cavity and the state where both QDs are excited. III. QUANTUM DOTS IN A CAVITY We focus on a system of two (singly) charged self- assembled InAs QDs in a photonic crystal microcav- ity. This structure can support in- and out-of-plane polarizations.22 Due to strain the optical dipole transi- tion matrix elements in the InAs dots are anisotropic, resulting in efficient absorption of light with electric field polarization perpendicular to the QD growth axis. As a result, only the mode with electric field polarized in the plane of the crystal can be coupled to transitions in QDs. We take an external magnetic field to be applied in-plane (Voigt configuration), perpendicular to the QD growth direction. This will enable full single qubit con- trol as explained in Ref. 20. The system can be represented by two separate four- state QDs interacting with a single photon mode, as shown in Fig. 1(a). The two lowest energy states of each four-state system are the spin states of the elec- † tron in each dot, which represent the qubit, ↑(cid:105) = c n,↑(cid:105)0 † and ↓(cid:105) = c n,↓(cid:105)0, where n = 1, 2 refers to the two † ↑(↓) creates an electron of spin ↑ (↓) rela- dots and c tive to the uncharged QD state (cid:105)0. The two excited states in each dot are electron-exciton bound states, called trions (or charged excitons). They are complexes having total angular momentum 3/2. The two ±3/2 states ('heavy holes') are energetically lower than the ±1/2 ('light hole') states and thus form a pseudo spin † † † † ⇑(cid:105) = t n,⇑(cid:105)0 and similarly for ⇓(cid:105), n,↑(cid:105)0 = c n,↓h n,↑c where h† is the creation operator for a heavy hole. The trion states carry the pseudospin of the hole because the two electrons are in a spin singlet. We choose the spin quantization axis along the external magnetic field. The cavity couples to the trion transitions and pre- serves the (pseudo) spin orientation, ↑(cid:105) ↔ ⇑(cid:105) and ↓(cid:105) ↔ ⇓(cid:105). In the rotating-wave approximation the cavity-dot interaction is (cid:88) (cid:16) HQD−C = g (cid:17) † † n,↑cn,↑a + t n,↓cn,↓a + h.c. t (1) n=1,2 where a annihilates a photon in the cavity and g is the coupling between the trion transition and the cavity. We 2 (a) Energies and relevant FIG. 1: The cavity-dot system. states of two QDs and cavity. (b-e) Interacting cavity-dot spectrum as a function of the cavity mode frequency, ω0. (b) Structure of crossings and corresponding states. Panels (c) and (d) show the anti-crossing splittings in the two- and single-excitation subspaces respectively. Panel (e) shows the energy structure of the qubit subspace, which is unaffected by the coupling to the cavity mode. The numbers in (c-e) give the states of the diagonalized Hamiltonian, and the ↑, ↓ show the predominant spin character of each state far (to the right) of the avoided crossings. Vertical dashed lines indicate (ε1 + ε2)/2. choose these coupling constants to be the same for the two QDs to simplify the presentation. This assumption is not important to the proposed procedure and can be relaxed when necessary. The spectrum of the cavity-QD system is shown in Fig. 1 as a function of the cavity frequency ω0. This representation does not suggest the need to tune ω0 dy- namically, but it helps to identify the region of optimal ω0 values. The spectrum is obtained by diagonalizing H0 = † n,ξcn,ξ+ † n,ξ[εn+ωhθ(ξ)]tn,ξ, HC = ω0a†a, ξ =↑,↓, θ(↑) = 0, n,ξ t HQD + HC + HQD−C, where HQD =(cid:80) (cid:80) n,ξωeθ(ξ)c and θ(↓) = 1. The Hamiltonian H0 conserves the total number of ex- citations. As a result the Hilbert space of the system separates into subspaces with different numbers of ex- citations. Each subspace contains several states, corre- sponding to different spin projections. The lowest set of four states (three energy levels) defines the two-qubit (a)(b)qubit subspaceone-particle subspacetwo-particles subspacecavity modedot 1dot 2(c)(d)(e)optical controlsubspace, ↑↑(cid:105), ↑↓(cid:105), ↓↑(cid:105), and ↓↓(cid:105), and has zero cav- ity photons; we call this the 'zero excitation' subspace. The other relevant subspaces are the 'one-excitation' sub- space, that has states with either one cavity photon or one trion, and the 'two-excitation' subspace, that has states with two trions (one per dot), states with one trion and one cavity photon, and states with two cavity pho- tons; see Appendix A. States in the 'one-excitation' part of the spectrum are approximately local to each quan- tum dot and interact with each other only very weakly, ∼ (g/∆ε)2. They are the states that can be used for single-qubit control.23 The two-excitation subspace in- volves hybridized states of the two QDs and are ideal for a two-qubit gate. These states however are not directly accessible from the qubit subspace with a single pulse, so we make use of a series of control pulses. The laser pulses have momentum perpendicular to the photonic crystal plane to avoid Bragg shielding due to the photonic crystal. For definiteness we choose pulses with the same linear polarization as the cavity mode,24 (cid:88) V(t) = Ωp(t − tp)2 cos ωpt (Mnmn(cid:105)(cid:104)m+h.c.) . (2) p; n>m where H0 = U†H0U = (cid:80) (cid:80) j=1,2,ξ(cid:104)nU†(t The total Hamiltonian becomes H(t) = H0 + V(t), n Enn(cid:105)(cid:104)n and Mn,m = † † j,ξtj,ξ)Um(cid:105). The subscript p j,ξcj,ξ + c enumerates the pulses used to perform the gate where each has frequency ωp and is centered at time tp. IV. IMPLEMENTATION OF CZ GATE The CZ gate has a simple diagonal form, which allows for a relatively straightforward design based on phases induced by resonant cyclic excitation of an auxiliary ex- cited state. The idea is to use the property of quantum two-level systems, in which a cyclic evolution from the ground state to the excited state and back to the ground state induces a minus sign to the latter. In the presence of additional, uncoupled states the minus sign is relative and thus constitutes a nontrivial quantum evolution. The pulse performing such an evolution is known as a '2π' pulse. Optical 2π pulses were proposed theoretically for single-qubit rotations in quantum dots23 and two-qubit gates in quantum dot molecules25 and later used in their experimental demonstrations.7,17 In our approach, the phase accumulation will be on state ↑↓(cid:105), while keeping the phases of other basis states unchanged. This can be done by the following pulse se- quence: (i) a population inversion π pulse, pulse A, tuned to transition ω1 = ωA = E4−E0 between qubit state ↑↑(cid:105) and the excited state with similar spin configuration, see Fig 1(d)-(e). The pulse is also in resonance with E6 − E2 transition, and thus it creates a trion in the first QD only: both ↑↑(cid:105) and ↑↓(cid:105) are transformed in the same way and accumulate a phase factor of −i each. (ii) A 2π, or phase, pulse (pulse B) with frequency ω2 = ωB = E16 − E4, 3 see Fig 1(c)-(d). This induces a transition between the 'one-excitation' states previously created and one of the 'two-excitation' states. Note that if g = 0 or we are far detuned, ω (cid:29) ∆ε, the transition E10 − E2 would also occur. This would correspond to a single qubit opera- tion on the second qubit, i.e., ⇑↑(cid:105) and ↓↑(cid:105) would both acquire a phase factor of −1. A nonzero g induces forma- tion of two-excitation states that have different energies, c.f. the energy of state 16 and the sum of energies of states 4 and 8. As a result, the state ⇑↑, or state 4, ac- quires the factor of −1 after the pulse, while state ↓↑(cid:105) does not. (iii) Finally, we apply the population inversion pulse A again, ω3 = ωA, to restore the system to the qubit subspace. This gives additional factors of −i to both ↑↑(cid:105) and ↑↓(cid:105), as mentioned above. The two phase factors of (−i) induce a minus sign to states ↑↓(cid:105) and ↑↑(cid:105), while the 2π pulse cancels that sign in state ↑↑(cid:105). The phase between the control pulses A and B does not enter the result and therefore pulses with unequal frequencies do not have to be phase locked, which is a significant experimental convenience. A physical explanation of this approach is the follow- ing: because each QD is off-resonant from the cavity, when only one of the QDs is excited and no other exci- tations are present in the system the excited QD can be roughly thought of as isolated, i.e., decoupled from the cavity and from the other QD. Thus, single excitations can implement single-qubit operations without disturb- ing the rest of the system. On the other hand, when both QDs are excited they are closer to the resonance with the cavity state. As a result, there is a large mixing between cavity states and the states of both QDs. Thus, using the two-excitation regime is a natural venue for performing two-qubit conditional operations while maintaining the ability to manipulate each QD spin separately. V. FIDELITY Now we consider the gate fidelity, which is a measure of how close our operation is to the target gate. There are two types of fidelity losses, those caused by unin- tended coherent dynamics due to coupling of the lasers to off-resonance transitions and those originating from random processes such as trion recombination. First we focus on the former mechanism. The unintended tran- sitions can cause Ug to deviate from the ideal UCZ and effectively cause loss of coherence in the qubit-subspace, even though the entire operation involving excited states is unitary and coherent. To analyze this type of decoherence we compute the average fidelity, F , of the gate operation, as explained in detail in Appendix B, including transitions 0-4, 4-6, 2-5, 3-7, 0-8, 2-10, 1-9, 3-11 for pulse A and transitions 4-16, 6-18, 0-8, 2-10, 1-9, 3-11 for pulse B. Other transitions are negligible either due to vanishing matrix elements or to large detuning. We chose different pulse widths for pulse A and pulse B σA = 2σB = 2σ. In Fig. 2(a) 4 FIG. 2: Fidelity of CZ gate with imperfections resulting from coupling of the pulses to neighboring off-resonance optical transitions (a) as a function of ω0 for σ/ωe = 0.1 for ∆ε/ωe = 8.33, 16.67, 25.00, 33.33, as indicated, and (b) as a function of the spectral separation ∆ε between the QDs for different values of the pulse bandwidth, σ/ωe = 0.01, 0.02, ..., 0.1, 0.15, ..., 0.3 as indicated by the dashed arrow. Each point is computed for the optimal value of ω0 from Fig. 2(a). The vertical lines mark the values of ω2 from panel (a). In both panels (a) and (b) we used g/ωe = 3.33 and ωh = ωe/3. FIG. 3: (a) Fidelity of the two-qubit CZ gate in presence of decoherence due to trion recombination and cavity decay. The fidelity is plotted as the function of the trion decay time and the cavity mode quality factor. (b) The temporal profiles of the pulse sequence for σ/ωe = 0.2, ωh = ωe/3, ∆ε/ωe = 8.33, g/ωe = 3.33, and ωe = 0.12meV. the fidelity is plotted as a function of the difference be- tween the cavity mode frequency ω0 and the transition frequency of QD1 ε1 for varying values of the frequency of QD2 ε2. The qualitative features of the plots can be understood as follows: when the cavity mode frequency is much smaller or much larger than the QD frequencies, QD-cavity hybridization is negligible, and we are in a regime of two independent qubits. This causes attenua- tion of fidelity towards both sides of the plot. The dip in the middle occurs because, as the cavity is tuned, the target transition of pulse B (transition 4-16) becomes de- generate with transition 3-11, and therefore state ↓↓(cid:105) is also affected by pulse B, resulting in strong unintended dynamics. Note that at its high values the fidelity does not vary strongly with ε1 and ε2. As a result, gates be- tween several different pairs of quantum dot spin qubits can be performed with high fidelity using only one cavity mode to mediate the interactions, which is an intriguing opportunity for scalable architectures. Fig. 2(b) shows the fidelity as a function of the spectral separation ∆ε between the trions in QD1 and in QD2 for different pulse bandwidths σ. When ∆ε is small (com- parable to ωe) the fidelity drops appreciably. This drop is the result of coupling in the 'one-excitation' subspace, i.e., the assumption that an excited QD is isolated from the rest of the system is no longer valid. Thus it also identifies the regime where fast optical single-qubit con- trol is not possible. In the region of higher fidelities, where ∆ε/ωe >∼ 10, the fidelity approaches its maximal value for longer pulses and starts decreasing more rapidly for σ/ωe >∼ 0.2 due to involvement of a larger number of unintended transitions. Next, we consider the effects of decoherence due to losses during the gate. The main contributions come from trion recombination and cavity photon leakage. The typical linewidth of the trion state, Γtr, in InAs QDs is ∼ 1µeV.26 The loss rate associated with the cavity is Γc = ω0/Q. State-of-the-art microcavities15 can have Q's up to ∼ 105, which gives Γc ∼ 10µeV. We calculate the fi- delity using the standard master equation formalism21,27 and include states from 0 to 19, see Appendix B. The fidelity as a function of Q and 1/Γtr in shown in Fig. 3. It is maximized when the pulses overlap to reduce the time the excited states are occupied. We choose ω0 from the maximal fidelities, as in Fig. 2(a) for each point of Fig. 3. We see that fidelities in excess of 90% are possible for realistic values of the parameters. 2510200.60.70.80.91.0(a)(b)0102030400.30.40.50.60.70.80.91.08.3316.6725.0033.33050100150200250arb.u.(a)(b)VI. CONCLUSIONS In summary, we have developed a design for a cavity- mediated entangling gate between two spin qubits that satisfies the criteria for a realistic two-qubit operation. Our control-z gate is compatible with available single qubit operations and with natural inhomogeneities in op- tical resonances. It can thus accommodate several qubits that couple pairwise with appropriate control laser fre- quencies, opening a path to scalable architectures. It may also be useful for hybrid quantum computing ap- proaches with various physical systems.28 We have shown that the gate fidelity is at least 90% for current experi- mental parameters. Higher fidelities can be achieved in various ways such as using pulse shaping techniques29,30 and engineering higher finesse cavities. VII. ACKNOWLEDGEMENTS This work was supported in part by NSA/LPS and in part by ONR. Appendix A: The spectrum In the rotating wave approximation the Hamiltonian [Eq. (1) from the main text] conserves the total number of excitations. Therefore it can be diagonalized indepen- dently in each excitation-number subspace. The lowest energy set of four states (three energy levels) corresponds to a subset with zero excitations. It represents the two- qubit subspace with zero cavity photons, 0 → ↑↑(cid:105)0(cid:105), 1 → ↑↓(cid:105)0(cid:105), 2 → ↓↑(cid:105)0(cid:105), 3 → ↓↓(cid:105)0(cid:105), (A1) where 0(cid:105) is the vacuum state of the cavity. The corre- sponding energies are controlled by the magnetic field via Zeeman splitting. For typical values of magnetic field of ∼ 1 T used in the initialization and readout and single- qubit experiments the splitting between E0,E3 and E1,2 is ∼ 0.1 meV. The micro-cavity optical mode is coupled to the excitonic transitions in each quantum dot with the transition energies ∼ eV. As a result, the qubit subspace is not affected by the cavity. quency, ∼ eV from the qubit subspace energies: ⇑↑(cid:105)0(cid:105),⇓↑(cid:105)0(cid:105),⇑↓(cid:105)0(cid:105),⇓↓(cid:105)0(cid:105), ↑⇑(cid:105)0(cid:105),↑⇓(cid:105)0(cid:105),↓⇑(cid:105)0(cid:105),↓⇓(cid:105)0(cid:105), ↑↑(cid:105)1(cid:105),↑↓(cid:105)1(cid:105),↓↑(cid:105)1(cid:105),↓↓(cid:105)1(cid:105), (A2) (A3) (A4) where 1(cid:105) denotes the state with a single photon in the cavity. The energy gap ∆ε between states (A2) and (A3) is due to the fact that the two dots are not identical in size, shape, and strain environment, which affects the excitonic transitions. The typical variation in trion tran- sition energies is ∼ 1− 20 meV. The energy of the cavity The one-excitation subspace occurs at the optical fre- dot 1 : dot 2 : cavity : 5 mode, ω0, is fixed during the gate operation but can be set to an optimal value during sample growth. The in- teraction with a cavity photon shifts the energies and mixes trion and photon states. The energies of the re- sulting states can be found analytically: note that states (A2-A4) are always coupled in triplets. For example, the state ↑↑(cid:105)1(cid:105) interacts only with ⇑↑(cid:105)0(cid:105) and ↑⇑(cid:105)0(cid:105). For each triplet we have (E−ε1,ξ)(E−ε2,ξ)(E−ω0) = g2(E−ε1,ξ)+g2(E−ε2,ξ),(A5) where ξ =↑ or ↓, εn,↑ = εn and εn,↓ = εn + ωh − ωe. Each triplet forms two anti-crossings when ω0 is swapped across the trion energies [see Fig. 1(b) and Fig. 1(d) of the main text]. When g ∼ ∆ε or g (cid:29) ∆ε, the two excited quantum dot states can mix and form spin-entangled states. For experimentally accessible systems of quantum dots in a micro-cavity the coupling strength g is substan- tially smaller than the variation in trion energies ∆ε and the mixing is negligible. In the limit g (cid:28) ∆ε the interac- tion between one-excitation states from different QDs can be estimated by analyzing the difference δω↑ in transition energies between ω↑ : ↑↑(cid:105) → ⇑↑(cid:105) and ω(cid:48) ↑ : ↑↓(cid:105) → ⇑↓(cid:105). From Eq. (A5) we find ω↑ = ε + g2/f (ωA, ∆ε) and A, ∆ε + ωe − ωh), where f (y, x) = ω(cid:48) ↑ = ε + g2/f (ω(cid:48) x − ω0 − g2/(x − ε + y). Since ωe ∼ ωh ∼ g (cid:28) ∆ε it is easy to show that δω↑ <∼ −g2ωe/∆ε2. This should be compared to the typical inverse lifetime of the trion state, ∼ 1µeV (in energy units) or ∼ ωe/100. As a result for ωe/∆ε ∼ 10, ω↑ and ω(cid:48) ↑ are practically indistinguishable. This result is confirmed numerically by computing the spectrum (and the states) for different values of ∆ε. It also holds for other transitions between the qubit and the one-excitation subspace states. Therefore we conclude that the one excitation subspace cannot be used for a two-qubit operations. It can, however, be used to per- form fast single qubit operations as described in Ref. 23 by using the localized trion state. In order to find useful non-local states that can mediate a two-qubit gate we investigate the two-excitation sub- space. In this subspace the states are coupled in groups of four, e.g. ⇑⇑, 0(cid:105), ⇑↑, 1(cid:105), ↑⇑, 1(cid:105), ↑↑, 2(cid:105): (ε2,ξ +ε1,ξ−E)(ω+ε1,ξ−E)(ω0 +ε2,ξ−E)(2ω−E)(A6) = g2(ε2,ξ +ε1,ξ +2ω0−2E)2. The spectrum has a more complex structure, see Fig. 1(c) of the main text. The two-excitations subspace provides non-local quantum-dot-cavity states, such as state 16, which has two trions (one in each dot). The energy of such state is different from the combined energy of two trion states localized in each dot, such as E4 and E8, ∆E16,4 (cid:54)= ∆E4,0 + ∆E8,0 (A7) where ∆En,n(cid:48) = En − En(cid:48). This is the basis for the two-qubit conditional phase gate in this work. Us- ing a perturbative approach like that above, we obtain ∆E16,4 − (∆E4,0 + ∆E8,0) ∼ −g2/∆ε. Appendix B: Gate Fidelity The fidelity of the gate described in the main text is affected by two type of processes: (i) induced unintended transitions between the states of the qubit-cavity system and (ii) real losses due to cavity leakage and trion re- combination. We first estimate losses due to unintended but coherent dynamics. We include transitions 0-4, 4-6, 2-5, 3-7, 0-8, 2-10, 1-9, 3-11 for pulse A, and 4-16, 6-18, 0-8, 2-10, 1-9, 3-11 for pulse B (ωB). Other transitions are negligible either due to vanishing matrix elements or to large detuning. We compute the wave function after the A-B-A pulse sequence for each basis configuration of the qubit subspace as initial state (evolution is linear and therefore the resultant wave function for any initial qubit state can be easily recovered). To simplify calculations here we resort to analytically solvable Rosen-Zener pulse shapes,31 i.e. Ωp(t) = Ωpsech(σpt) with σA = 2σB = 2σ, to calculated transition amplitudes and phases for reso- nant and off-resonance transitions for each pulse. Given the initial, ψ0(cid:105), and final, ψ(cid:105) = Uψ0(cid:105), wave function, the fidelity can be computed as F (ψ0, ψ) = (cid:104)ψ0U † CZψ(cid:105) (B1) † where U CZ is the evolution operator corresponding to the ideal CZ gate. The value of F (ψ0, ψ) depends on the initial state of the two-qubit system and therefore can vary depending on the choice of algorithm and initial data. We therefore compute the average fidelity F by taking average over all possible initial states of the two- qubit system, F 2 = dψ0F (ψ0, ψ{ψ0})2 δinδjm + δijδnm (B2) † CZUi(cid:105)(cid:104)jU†UCZm(cid:105) (cid:104)nU = 20 ijnm The integration(cid:82) dψ0 is performed over all complex am- plitudes that define the initial state in the basis i(cid:105), and i, j, n, m run over all basis states ↑↑,↑↓,↓↑,↓↓.32 The re- sults are presented in Fig. 2 and the discussion is given in the main text. In order to account for both unintended dynamics and actual losses we have to calculate the reduced density matrix, ρ(t), of the two-qubit sub-system for the duration of the pulse sequence. The reduced density matrix can be found within the Bloch-Redfield master-equation (ME) formalism (cid:20) (cid:88) s i ρ = [H + V (t), ρ] PsρP † iΓs + s − P † s Psρ + ρP † s Ps 2 (cid:21) (B3) (cid:90) (cid:88) 6 where Ps = fs(cid:105)(cid:104)is, is(cid:105) and fs(cid:105) are initial and finial states (in the spin basis) corresponding to the s-th de- cay process with rate Γs. Solving the above equation directly is computationally involving due to the presence of two different time scales: fast, associated with the laser driving frequency, and slow, coming from the time- dependence of the pulse shaped envelope. To simplify the computation we transform the ME to the eigenbasis of H and use the rotating wave approximation, (cid:20) (cid:88) ρ = −i[V (t), ρ] Ps ρP† Γs + s s − P† sPs ρ + ρP† sPs 2 (cid:21) , (B4) where ρ = eiH0tU†ρU e−iH0t, Ps = U†PsU and V (t) = eiH0tV(t)e−iH0t. Note that the trion decay pro- cesses can involve photons with any in-plane polariza- tion (along or perpendicular to the applied magnetic for each trion we have Γs = Γtr: field). Therefore, Ps → {↑(cid:105)(cid:104)⇑,↓(cid:105)(cid:104)⇑,↑(cid:105)(cid:104)⇓,↓(cid:105)(cid:104)⇓}. Leakage of pho- tons from the cavity is modeled as Γs = Γc: Ps → 21(cid:105)(cid:104)2, etc.}. Due to additional (pseudo)spin- {0(cid:105)(cid:104)1, flip electron-hole recombination processes, more states are involved than for the coherent case discussed above and we include states from 0 to 19. We chose to use Gaus- p/π2} for numerical convenience and apply the same pulse se- quence as before with σA = 2σB = 2σ. sian pulse shapes Ωp(t) = (Ωp/(cid:112)π/2) exp{−2t2σ2 √ Since a separable quantum wave function is no longer accessible, fidelity has to be defined differently, F (ψ0, ρ{ψ0}) = (cid:104)ψ0U † CZ ρ UCZψ0(cid:105) (B5) (cid:113) In this case the average fidelity is computed as δinδjm + δijδnm (cid:104)nU † CZρ{i(cid:105)(cid:104)j}UCZm(cid:105) (B6) F 2= ijnm={1,4} 20 (cid:88) which is the generalization of Eq. (B2) for the case of non-unitary evolution of pure initial state. This is possi- ble due to the fact that the evolution of the density ma- trix is still described by a linear (but non-unitary) super- operator, i.e. ρ(t) = T exp(−i(cid:82) t 0 dtLH (t) − tL)ψ0(cid:105)(cid:104)ψ0, where LH O = [H, O]. As a result, the complex coef- ficients that define initial (ψ0(cid:105)) and target (UCZψ0(cid:105)) states in the basis i(cid:105) can be integrated out in exactly the same way as for Eq. (B2). The results are presented in Figs. 2 and 3 in the main text. ∗ [email protected] 1 T. D. Ladd, F. Jelezko, R. Laflamme, Y. Nakamura, C. Monroe, and J. L. OBrien, Nature (London) 464, 45-53 (2010). 2 J. D. Sterk, L. Luo, T. A. Manning, P. Maunz, C. Monroe, arXiv:1112.4489 (2011). 3 D. Englund, B. Shields, K. Rivoire, F. Hatami, J. Vuckovic, H. Park, and M. D. Lukin, Nano Lett. 10, 39223926 (2010). 4 A. Faraon, P. E. Barclay, C. Santori, K-M. C. Fu, and Raymond G. Beausoleil, Nature Photonics 5, 301 (2011). 5 J. P. Reithmaier, G. Sek, A. Lffler, C. Hofmann, S. Kuhn, S. Reitzenstein, L. V. Keldysh, V. D. Kulakovskii, T. L. Reinecke, and A. Forchel, Nature (London) 432, 197 (2004). 6 D. Press, T. D. Ladd, B. Zhang, and Y. Yamamoto, Nature (London) 456, 218 (2008). 7 A. Greilich, S. E. Economou, S. Spatzek, D. R. Yakovlev, D. Reuter, A. D. Wieck, T. L. Reinecke, and M. Bayer, Nature Physics 5, 262 (2009). 8 T. Yoshie, A. Scherer, J. Hendrickson, G. Khitrova, H. M. Gibbs, G. Rupper, C. Ell, O. B. Shchekin, and D. G. Deppe, Nature (London) 432, 200 (2004). 9 A. Badolato, K. Hennessy, M. Atature, J. Dreiser, E. Hu, P. M. Petroff, and A. Imamoglu, Science 308, 1158 (2005). 10 G. Khitrova, H. M. Gibbs, M. Kira, S. W. Koch, and A. Scherer, Nature Physics 2, 81 (2006). 11 K. Hennessy, A. Badolato, M. Winger, D. Gerace, M. Atature, S. Gulde, S. Falt, E. L. Hu, and A. Imamoglu, Nature (London) 445, 896 (2007). 12 D. Pinotsi, P. Fallahi, J. Miguel-Sanchez, and A. Imamoglu, IEEE J. Quant. El. 47, 1371, (2011). 13 S. M. Thon, H. Kim, C. Bonato, J. Gudat, J. Hagemeier, P. M. Petroff, and D. Bouwmeester, arXiv:1109.5016 (2011). 14 E. Gallardo, L. J. Martinez, A. K. Nowak, D. Sarkar, H. P. van der Meulen, J. M. Calleja, C. Tejedor, I. Prieto, D. Granados, A. G. Taboada, J. M. Garcia, and P. A. Postigo, Phys. Rev. B 81, 193301 (2010). 15 S. Reitzenstein, C. Hofmann, A. Gorbunov, M. Strau, S. H. Kwon, C. Schneider, A. Lffler, S. Hfling, M. Kamp, and A. Forchel, Appl. Phys. Lett. 90, 251109 (2007). 7 16 S. G. Carter, T. M. Sweeney, M. Kim, C. S. Kim, D. Solenov, S. E. Economou, T. L. Reinecke, L. Yang, A. S. Bracker, D. Gammon, arXiv:1211.4540 17 D. Kim, S. G. Carter, A. Greilich, A. S. Bracker and D. Gammon, Nature Physics 7, 223 (2011); A. Greilich, S. G. Carter, D. Kim, A. S. Bracker and D. Gammon, Nature Photonics 5, 702708 (2011). 18 J-Q Zhang, Y-F Yu, and Z-M Zhang, J. Opt. Soc. Am. B 28, 1959 (2011). 19 T. D. Ladd and Y. Yamamoto, Phys. Rev. B 84, 235307 (2011). 20 A. Imamoglu, D. D. Awschalom, G. Burkard, D. P. Di- Vincenzo, D. Loss, M. Shermin, and A. Small, Phys. Rev. Lett. 83, 4204 (1999). 21 M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, Cambridge, UK, 2000). 22 M. Notomi, Rep. Prog. Phys. 73 (2010). 23 S. E. Economou, L. J. Sham, Y. Wu, and D. G. Steel, Phys. Rev. B 74, 205415 (2006); S. E. Economou and T. L. Reinecke, Phys. Rev. Lett. 99, 217401 (2007). 24 This assumption is not crucial and can be easily relaxed to accommodate pulses with other polarizations. 25 S. E. Economou and T. L. Reinecke, Phys. Rev. B 78, 115306 (2008). 26 S. Cortez, O. Krebs, S. Laurent, M. Senes, X. Marie, P. Voisin, R. Ferreira, G. Bastard, J-M. Gerard, and T. Amand, Phys. Rev. Lett. 89, 207401 (2002). 27 D. Solenov, D. Tolkunov, and V. Privman, Phys. Rev. B 75, 035134 (2007). 28 E. Waks and C. Monroe, Phys. Rev. A 80, 062330 (2009). 29 J. P. Palao and R. Kosloff, Phys. Rev. Lett. 89, 188301 (2002). 30 S. E. Economou, Phys. Rev. B 85, 241401(R) (2012). 31 N. Rosen and C. Zener, Phys. Rev. 40, 502 (1932). 32 L. H. Pedersena, N. M. Møller, K. Mølmer, Phys. Lett. A 367, 47 (2007). 

1201.1690

1

1201

A topological look at the quantum spin Hall state

We propose a topological understanding of the quantum spin Hall state without considering any symmetries, and it follows from the gauge invariance that either the energy gap or the spin spectrum gap needs to close on the system edges, the former scenario generally resulting in counterpropagating gapless edge states. Based upon the Kane-Mele model with a uniform exchange field and a sublattice staggered confining potential near the sample boundaries, we demonstrate the existence of such gapless edge states and their robust properties in the presence of impurities. These gapless edge states are protected by the band topology alone, rather than any symmetries.

cond-mat.mes-hall

cond-mat

A topological look at the quantum spin Hall state National Laboratory of Solid State Microstructures and Department of Physics, Nanjing University, Nanjing 210093, China Huichao Li, L. Sheng,∗ and D. Y. Xing We propose a topological understanding of the quantum spin Hall state without considering any symmetries, and it follows from the gauge invariance that either the energy gap or the spin spectrum gap needs to close on the system edges, the former scenario generally resulting in counterpropagating gapless edge states. Based upon the Kane-Mele model with a uniform exchange field and a sublattice staggered confining potential near the sample boundaries, we demonstrate the existence of such gapless edge states and their robust properties in the presence of impurities. These gapless edge states are protected by the band topology alone, rather than any symmetries. PACS numbers: 72.25.-b, 73.20.At, 73.22.-f, 73.43.-f Since the remarkable discovery of the quantum Hall effect (QHE) [1], the study of edge state physics has attracted much attention on both theoretical and ex- perimental sides. Recently, a new class of topological states of matter has emerged, called the quantum spin Hall (QSH) states [2, 3]. A QSH state of matter has a bulk energy gap separating the valence and conduction bands and a pair of spin-filtered gapless edge states on the boundary. The QSH effect was first predicted in two- dimensional (2D) models [2, 3], and was experimentally confirmed soon after in mercury telluride quantum wells [4]. The QSH systems are 2D topological insulators [5, 6] protected by the time-reversal symmetry (TRS), whose edge states are robust against perturbations such as non- magnetic disorder. A simple model of the QSH systems is the Kane-Mele model [2], defined on a honeycomb lattice, first intro- duced for graphene with spin-orbit couplings (SOCs). It was suggested [2] that the intrinsic SOC in graphene would open a band gap in the bulk and also establish spin-filtered edge states that traverse the band gap, giv- ing rise to the QSH effect. Even though the intrinsic SOC strength in pure graphene is too small to produce an ob- servable effect under realistic conditions [7], the Kane- Mele model captures the essential physics of the QSH state with nontrivial band topology [8, 9]. In the pres- ence of the Rashba SOC and an exchange field, the Kane- Mele model enters a TRS-broken QSH phase [10] charac- terized by nonzero spin Chern numbers [11, 12]. Prodan proved [12] that the spin-Chern numbers are topological invariants, as long as the energy gap and the spectrum gap of the projected spin operator P σzP stay open in the bulk, where P is the projection operator onto the sub- space of the occupied bands and σz the Pauli matrix for the electron spin. Unlike the Z2 invariant [13], the robust properties of the spin-Chern numbers remain unchanged when the TRS is broken [10, 12]. The existence of counterpropagating edge states with opposite spin polarizations is an important characteristic of the QSH state. It is believed that the edge states can be gapless only if the TRS [2] or other symmetries, such as the inversion symmetry [14] or charge-conjugation TRS [15], are present. When the TRS is broken, it was found [10] that a small gap appears in the spectrum of the edge states, which was obtained for a ribbon geome- try under ideal boundaries, i.e., boundaries created by an infinite hard-wall confining potential. However, since the edge states are localized around the sample boundaries, they can be sensitive to the variation of on-site potentials near the boundaries [16]. In this Letter, in order to reveal the general character- istics of the edge states and their connection to the bulk topological invariant in a QSH system, we present a topo- logical argument similar to the Laughlin's Gedanken ex- periment without considering any symmetries. We show that, as required by the nontrivial band topology and gauge invariance, either the energy gap or the spin spec- trum gap (the gap in the spectrum of P σzP ) needs to close on the edges of a QSH system. These two scenarios will lead to gapless or gapped edge modes, respectively. In particular, it is demonstrated that gapless edge states can appear in a TRS-broken QSH system by tuning the confining potential at the boundaries. They are associ- ated with the bulk topological invariant, and are robust against relatively smooth impurity scattering potential. Our result offers an interesting example for counterpropa- gating gapless edge states that are not protected by sym- metries, which sheds light on the underlying mechanism of the QSH effect in a broad sense. Let us first look back on a looped ribbon of the QHE system, with a magnetic flux φ (in units of flux quantum hc/e) threading the ring adiabatically [17 -- 19]. The Fermi energy EF is assumed to lie in an energy gap. In the spirit of the Laughlin's argument, increasing φ from 0 to 1 effectively pumps one occupied state from one edge to the other, giving rise to the transfer of one charge between the edges, essentially because there is a nonzero Chern number (Hall conductivity) in the bulk. On the other hand, the system Hamiltonian is gauge invariant under integer flux changes, i.e., if φ is increased from 0 to 1, the system will reproduce the same eigenstates as at φ = 0. To assure this gauge invariance, there must be gapless edge modes on the edges (when the perimeter of the ring is large), so that the spectral flow can form a closed loop, as illustrated in Fig. 1a, along which the electron states can continuously move with changing φ. (a) (b) (c) EF Conduction bands Valence bands Conduction bands Valence bands Left edge Right edge FIG. 1: (a) A schematic of the flow of electron states in a looped ribbon of the QHE system, with adiabatically increas- ing the magnetic flux φ that threads the ring. The bulk elec- tron states below Fermi energy EF drift from left to right, gapless edge modes ascend through EF on the right edge, states above EF drift from right to left, and gapless edge modes descend through EF on the left edge, forming a closed loop. (b) In the first scenario for the QSH system, gapless edge modes appear on the edges of the ring, so that electron states in each spin sector behave like in the QHE system, but those in two spin sectors move in opposite directions. (c) In the second scenario for the QSH system, the edge states are gapped, whereas the spin spectrum gap closes on the edges. In the bulk, electron states in the two spin sectors drift in the same way as in (b), but on the edges they join together within the valence (conduction) band. We now propose a topological understanding of the QSH system in terms of the same looped ribbon geom- etry. The occupied valence band can be decomposed into two spin sectors by using the projected spin oper- ator [10, 12]. (The unoccupied conduction band can be divided similarly.) The two spin sectors are separated by a nonzero spin spectrum gap in the bulk. They carry op- posite spin Chern numbers, so that increasing φ pumps a state of the spin up sector in the occupied band from one edge to the other, and pumps another state of the spin down sector in the opposite direction. In order for the system to recover the initial eigenstates as φ changes from 0 to 1, the spectral flow needs to form closed loops, simi- larly to the QHE system. However, for the QSH system, if not enforced by any symmetry, two different scenarios can occur on the edges. One is that gapless edge modes appear on the edges, so that states can move between the conduction and valence bands with changing φ to form closed loops in the spin-up and spin-down sectors sepa- rately, as shown in Fig. 1b. In this case, the states in the two loops cannot evolve into each other due to the non- vanishing spin spectrum gap both in the bulk and on the edges. The other scenario is that a closed loop of spec- tral flow is formed between the two spin sectors within 2 the valence (or conduction) band, as shown in Fig. 1c. In this case, the spin spectrum gap must vanish on the edges, but the energy gap may remain open in the edge state spectrum. (b) abrupt potential smooth potential (a) s e g d e r i a h c m r a zigzag edges FIG. 2: (a) A schematic of an armchair honeycomb lattice ribbon with atom sites in two sublattices being labeled by solid dots and hollow dots, where a is the distance between nearest-neighbor sites. (b) Profiles of Vi as functions of xi, for an abrupt confining potential (dashed line) and a relatively smooth confining potential (solid line). The above topological discussion on the QSH system is very general, independent of any symmetries. To demon- strate the two scenarios in Figs. 1b and 1c, in what follows we take the Kane-Mele model [2] for a honeycomb lattice ribbon as an example, by taking into account different confining potentials near the edges of the ribbon. It was shown that in a suitable parameter range, the Kane-Mele system is in the QSH phase protected by the TRS, and it can become a TRS-broken QSH phase [10], when a spin- splitting exchange field is applied. Consider an armchair ribbon along the y direction, as shown in Fig. 2a, in- cluding Nx dimer lines across the ribbon. (Results for a zigzag ribbon are similar.) The boundaries are at x1 = 0 √3 and xNx = 2 (Nx − 1), where the distance between nearest-neighbor sites is chosen as the unit of length. The Hamiltonian can be written as H = HKM + HE + HC with 2i √3 HKM = −tXhiji + iVRXhiji c†i cj + VSO Xhhijii c†i ez · (σ × dij)cj, c†i σ · (dkj × dik)cj (1) , c†i↓ as the Hamiltonian of the Kane-Mele model. Here, the first term is the nearest-neighbor hopping term with c†i = (c†i↑ ) as the electron creation operator on site i and the angular bracket in hi, ji standing for nearest-neighbor sites. The second term is the intrinsic SOC with coupling strength VSO, where σ are the Pauli matrices, i and j are two next nearest neighbor sites, k is their unique common nearest neighbor, and vector dik points from k to i. The third term is the Rashba SOC with coupling strength VR. HE = gPi c†i σzci stands for a uniform exchange field of strength g. HC represents a sublattice staggered confining potential, which is given by HC = Pi Vic†i ci with Vi = ±V0(cid:18)e− xi ξ + e− xNx −xi ξ (cid:19) , (2) where ± is taken to be positive for sites on sublattice A (solid dots) and negative on sublattice B (hollow dots), as shown in the Fig. 2a. In Eq. (2), Vi is strongly x depen- dent across the ribbon, equal to ±V0 at the edges (xi = 0 and xi = xNx ). It decays exponentially away from the edges, with a characteristic length ξ, as shown in Fig. 2b. When the ribbon width is much greater than ξ, Vi essentially vanishes in the middle region of the ribbon. Here, we note that in the case of a uniform staggered potential Vi = ±V0 (Vi being independent of xi), it was shown [10] that with increasing Vi, there is a transition from the TRS-broken QSH phase to an ordinary insu- lator state, where the middle band gap closes and then reopens. Therefore, for the confining potential Vi given by Eq. (2) with large V0, the ribbon in Fig. 2 can be regarded as a TRS-broken QSH ribbon sandwiched in between two trivial band insulators. 0.2 0.0 (a) R↓ L↑ R↑ L↓ (b) R↓ L↑ R↑ L↓ -0.4 -0.2 0.0 (π) y 3k 0.2 0.4 -0.2 0.0 (π) y 3k FIG. 3: (color online) The energy spectrum (a, b) and the spectrum of the projected spin P σzP (c, d) of an armchair ribbon for ξ = 0.1 (a, c) and ξ = 4 (b, d), in which L (R) stands for states on the left (right) edge, and ↑ (↓) for the up (down) spin polarization. The horizontal arrows in (a) point to the small energy gaps in the energy spectrum. At ξ = 0.1, while the energy spectrum (a) is gapped, the spin spectrum (c) is gapless. At ξ = 4, the energy spectrum (b) is gapless, but the spin spectrum (d) is gapped. In order to assure the system in the TRS-broken QSH state, we set the parameters Vso = 0.1t, VR = 0.1t, and g = 0.1t. The length of the armchair ribbon Ny is taken to be infinite. The energy spectrum of the ribbon, together with the corresponding eigenfunctions ϕm(ky), can be numerically obtained by diagonalizing the Hamil- tonian for each momentum ky in the y direction. The calculated energy spectrum of the armchair ribbon with width Nx = 240, for the confining potential with V0 = 12t x=0.1 -0.2 -0.4 1.0 0.5 0.0 (c) x=0.1 ) t ( y g r e n E m u r t c e p s n p S i -0.5 -1.0 x=4.0 (d) x=4.0 0.2 0.4 3 fixed and two different decay lengths, is plotted in Fig. 3a and Fig. 3b. One can see easily that the edge states appear as thin lines in the middle bulk band gap of the energy spectrum. In Figs. 3a and 3b, the spin polariza- tion of the edge states is labeled with ↑ and ↓, indicating that two spin-filtered channels on each edge flow along opposite directions. For the nearly hard-wall confining potential of ξ = 0.1, the sublattice potential Vi is nonzero only on the outermost armchair lines, similar to the as- sumption in Ref. [16]. In this case, two small energy gaps are observed in the edge state spectrum shown in Fig. 3a, in agreement with the previous observation [10], as a consequence of the broken TRS. With increasing the decay length ξ of the confining potential, the energy gaps of the edge states become smaller and smaller. As ξ is large enough, e.g., ξ = 4 in Fig. 3b, interestingly, the edge states become gapless. We now calculate the ky-dependent spectrum of pro- jected spin operator P σzP , whose matrix elements are given by hϕm (ky) σz ϕn (ky)i with m and n running over all the occupied states. By diagonalizing this matrix, the spectrum of the projected spin P σzP can be obtained. For the Kane-Mele model Eq. (1), if VR = 0, σz com- mutes with the Hamiltonian H. Therefore, σz is a good quantum number. It follows that the spectrum of P σzP consists of just two values ±1, which are highly degen- erate. When the Rashba term is turned on, σz and H no longer commute, and the degeneracy is lifted. In this case, the spectra of P σzP between +1 and −1 spread towards the origin, but a gap remains for a bulk sample if the amplitude of the Rashba term dose not exceed a threshold [12]. For the ribbon geometry, the situation is more compli- cated due to the existence of the edges, and numerical calculations are performed to obtain the spin spectrum. The calculated spectrum of P σzP for the same param- eters as those in Figs. 3a and 3b is shown in Figs. 3c and 3d, which exhibits very interesting behavior. For the hard-wall confining potential with ξ = 0.1, while the energy spectrum of edge states are slightly gapped, with increasing ky the spectrum of P σzP continuously change between +1 to −1 without showing any gap, correspond- ing to the second scenario shown in Fig. 1c. On the other hand, for a relatively smooth confining potential with ξ = 4.0, the energy spectrum of the edge states are gapless, but the spectrum of P σzP displays a big gap, and the sudden changes happen at the cross points in the energy spectrum of the edge states, corresponding to the first scenario shown in Fig. 1b. From Fig. 3, it follows that as long as the system is in the QSH state, a gapless characteristic always appears either in the energy spec- trum of edge states or in the spectrum of P σzP , leading to the two types of closed loops for the continuous flow of the electron states illustrated in Figs. (1b, 1c). The result shown in Figs. 3b and 3d, for the relatively smooth confining potential, is of particular interest. It indicates that gapless edge states can exist in the TRS- broken QSH system, accompanied with a gapped spec- trum of P σzP . Such an interesting behavior can be fur- ther understood by the following argument. As long as the bulk energy gap does not close, the projected spin op- erator P σzP is exponentially localized in real space with a characteristic length about λ ∼ ¯hvF /∆E, where vF is the Fermi velocity and ∆E is the magnitude of the energy gap. [12] For the parameter set used in Fig. 3, λ is esti- mated to be between 1 to 2 lattice constants. When the confining potential Vi is varying relatively slowly in space, i.e., ξ ≫ λ, one can find that P σzP roughly commutes with the confining potential. In this case, the confining potential is of no influence on the spectrum of P σzP . Since the spin spectrum has a gap in the bulk [10], this gap remains to open on the smooth edges, as seen from Fig. 3d. As a result, the energy gap has to close due to the topological requirement, resulting in gapless edge modes, as observed in Fig. 3b. (a) l = 1.0 0 U = t 0 0.1 ) t ( y g r e n E 0.0 -0.1 0.0 0.5 (b) (c) l = 3.0 0 = t U 0 0.2 0.1 0.0 0.0 Magnetic flux φ 0.5 1.0 0.0 l = 3.0 0 U = 3t 0 1.0 0.5 FIG. 4: (color online) Eigenenergies of the edge states as a function of the magnetic flux φ threading the looped geometry with size 120×60 for (a) l0 = 1.0, U0 = t, (b) l0 = 3.0, U0 = t, and (c) l0 = 3.0, U0 = 3t, where the impurity concentration is fixed at 1%. The other parameters are taken to be VSO = VR = g = 0.1t, V0 = 12t, and ξ = 4. Arrows in (a) indicate some of the relatively large energy gaps. Finally, we wish to discuss the robustness of the gap- less edge states found in the present TRS-broken QSH system. We consider a Nx × Ny sample forming a looped geometry as that shown in Fig. 1. NI nonmagnetic im- purities are assumed to be randomly distributed in the sample at positions Rα with α = 1,··· NI . An extra term HI = Pi wic†i ci is added to the total Hamiltonian H to describe the effect of the impurity scattering, where wi = Pα U (ri − Rα) with ri as the position of the i-th atom site. The impurity scattering potential is taken to be U (ri−Rα) = (U0/l2 0) exp(−ri−Rα/l0) with l0 as the correlation length and U0 the strength of the scattering potential. By inclusion of 1/l2 0 in the prefactor, the area integral of the impurity potential is set to be indepen- 4 dent of l0. Figure 4 shows the evolution of the calculated eigenenergies of a 120 × 60 system in the band gap upon adiabatic insertion of a magnetic flux φ into the ring, for three different impurity scattering potentials. The num- ber concentration of the impurities is fixed at 1%. For a very short correlation length l0 = 1, for which the im- purity potential is nearly uncorrelated from one site to another, we see from Fig. 4a that at U0 = t, the energy levels of the edge states avoid to cross each other as they move close, resulting in small energy gaps in the spec- trum, as indicated by the arrows. This level repulsion behavior is a signature of the onset of backward scatter- ing [11]. When the characteristic length l0 is increased to l0 = 3 with U0 = t fixed, corresponding to a relatively smooth impurity scattering potential, all the energy gaps vanish, as shown in Fig. 4b. The energy levels move in straight lines and continue to cross each other, a clear indication of quenching of the backward scattering [11]. Such a level crossing feature is intact when U0 is increased up to 3t for fixed l0 = 3, as shown in Fig. 4c. We thus conclude that the edge states remain to be robust in the presence of relatively smooth impurity scattering poten- tial of intermediate strength. This result can be under- stood based upon an argument similar to that in the pure case. When l0 is greater than the characteristic length λ of the projected spin operator P σzP , the impurity scat- tering potential nearly commutes with P σzP , and hence does not affect much the spin spectrum gap, so that the energy gap needs to close on the edges, which explains the level crossing behavior of the edge modes. In summary, based upon a general topological argu- ment without relying on the TRS or other symmetries, we show that in a QSH system either the energy gap or the gap in the spectrum of P σzP needs to close on the edges. We find that a TRS-broken QSH system can have either gapless or gapped edge states, depending on the properties of the confining potential near the boundaries. The gapless edge states are protected by the bulk topo- logical invariant rather than any symmetries, which can remain to be robust in the presence of impurities. Acknowledgment This work is supported by the State Key Program for Basic Researches of China un- der Grant Nos. 2009CB929504 (LS), 2011CB922103 and 2010CB923400 (DYX), by the National Natural Science Foundation of China under Grant Nos. 11074110 (LS), 11174125, 11074109, and 91021003 (DYX), and by a project funded by the PAPD of Jiangsu Higher Educa- tion Institutions. ∗ Electronic address: [email protected] [1] K. V. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [2] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 5 [3] B. A. Bernevig, T. L. Hughes and S. C. Zhang, Science, 314 1757 (2006). [4] M. Konig, S. Wiedmann, C. Bruene, A. Roth, H. Buh- mann, L. W. Molenkamp, X. L. Qi and S. C. Zhang, Science 318, 766 (2007). [5] J. E. Moore and L. Balents, Phys. Rev. B, 75, 121306 (2007). [6] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007); L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [7] H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Klein- man, and A. H. MacDonald, Phys. Rev. B 74, 165310 (2006); Y. Yao, F. Ye, X. L. Qi, S. C. Zhang, and Z. Fang, Phys. Rev. B 75, 041401(R) (2007). [8] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [9] X. L. Qi and S. C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [10] Y. Y. Yang, Z. Xu, L. Sheng, B. G. Wang, D. Y. Xing, and D. N. Sheng, Phys. Rev. Lett. 107, 066602 (2011). [11] D. N. Sheng, Z. Y. Weng, L. Sheng, and F. D. M. Hal- dane, Phys. Rev. Lett. 97, 036808 (2006); L. Sheng, D. N. Sheng, C. S. Ting, and F. D. M. Haldane, Phys. Rev. Lett. 95, 136602 (2005). [12] E. Prodan, Phys. Rev. B 80, 125327 (2009). [13] C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [14] B. Zhou, H.-Z. Lu, R.-L. Chu, S.-Q. Shen, and Q. Niu, Phys. Rev. Lett. 101, 246807 (2008). [15] Q. F. Sun, and X. C. Xie, Phys. Rev. Lett. 104, 066805 (2010). [16] W. Yao, S. A. Yang, and Q. Niu, Phys. Rev. Lett. 102, 096801 (2009). [17] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). [18] B. I. Halperin, Phys. Rev. B 25, 2185 (1982). [19] J. E. Avron, D. Osadchy, and R. Seiler, Physics Today, August (2003), page 38. 

1607.08207

1

1607

The Role of Multilevel Landau-Zener Interference in Extreme Harmonic Generation

Motivated by the observation of multiphoton electric dipole spin resonance processes in InAs nanowires, we theoretically study the transport dynamics of a periodically driven five-level system, modeling the level structure of a two-electron double quantum dot. We show that the observed multiphoton resonances, which are dominant near interdot charge transitions, are due to multilevel Landau-Zener-Stuckelberg-Majorana interference. Here a third energy level serves as a shuttle that transfers population between the two resonant spin states. By numerically integrating the master equation we replicate the main features observed in the experiments: multiphoton resonances (as large as 8 photons), a robust odd-even dependence, and oscillations in the electric dipole spin resonance signal as a function of energy level detuning.

cond-mat.mes-hall

cond-mat

The Role of Multilevel Landau-Zener Interference in Extreme Harmonic Generation J. Stehlik,1 M. Z. Maialle,2 M. H. Degani,2 and J. R. Petta1 1Department of Physics, Princeton University, Princeton, New Jersey 08544, USA 2Faculdade de Ciencias Aplicadas, Universidade Estadual de Campinas, 13484-350 Limeira, Sao Paulo, Brazil Motivated by the observation of multiphoton electric dipole spin resonance processes in InAs nanowires, we theoretically study the transport dynamics of a periodically driven five-level system, modeling the level structure of a two-electron double quantum dot. We show that the observed multiphoton resonances, which are dominant near interdot charge transitions, are due to multilevel Landau-Zener-Stuckelberg-Majorana interference. Here a third energy level serves as a shuttle that transfers population between the two resonant spin states. By numerically integrating the master equation we replicate the main features observed in the experiments: multiphoton resonances (as large as 8 photons), a robust odd-even dependence, and oscillations in the electric dipole spin resonance signal as a function of energy level detuning. PACS numbers: 71.70.Ej, 73.63.Kv, 76.30.-v, 85.35.Gv I. INTRODUCTION Harmonic generation occurs in a nonlinear system when driving at frequency f results in a physical response of the system at multiples of the driving frequency, e.g. 2f , 3f , 4f , and underpins nonlinear and quantum op- tics [1, 2]. Two-photon absorption can be observed in optically pumped systems at high powers [3, 4]. Har- monic generation has also been observed in semiconduc- tor systems that are driven with terahertz pulses [5] and in electrically driven quantum dots [6, 7]. Generally, mul- tiphoton resonances are only observed at very high drive fields [8]. As a result, experimental observations are often limited to two-photon processes. Multiphoton resonances were recently observed in elec- tric dipole spin resonance (EDSR) in nanowire and pla- nar quantum dots [6, 7, 9]. Early experiments in these systems demonstrated electric driving of single electron spins [10]. These data were largely consistent with the- oretical predictions, with an EDSR response observed when hf = Ezi, where Ezi = giµBB is the Zeeman energy of the i-th dot, h is Planck's constant, f is the frequency of the electric driving field, gi is the electron g-factor of i-th dot, µB is the Bohr magneton and B is the applied magnetic field [11, 12]. However, more detailed investigations revealed that the multiphoton resonances were strongest when the double quantum dot (DQD) was driven near the interdot charge transition [9]. The EDSR harmonics, indexed by integer n, followed the resonance condition nhf = Ezi and showed a remarkable odd-even dependence, wherein the sign of the EDSR signal differed for odd and even n. In this paper we develop a full model of the DQD, building upon a three-level model of Danon and Rud- ner [13]. We start by calculating the time-evolution of a simple five-level system, which captures the physics of a two-electron DQD. These simulations demonstrate that when the DQD is initialized in a spin-blocked state the system can make a Landau-Zener-Stuckelberg-Majorana (LZSM) transition to an intermediate state, before mak- ing a final LZSM transition to a resonant unblocked state. Thus the harmonics can be understood as being a multi- level LZSM effect [14]. The EDSR resonances were observed in transport mea- surements. Therefore, to realistically model the exper- imental system, we add coupling to source-drain elec- trodes, decoherence, and charge noise [15]. Our work ex- tends the simple three-level model presented in Ref. [13] to a complete 5-level system that accurately describes a two-electron singlet-triplet qubit [16]. The additional levels are found to contribute to the observed resonances and allow us to make a quantitative comparison with the experimental data. The outline of the paper is as follows. In Sec. II we describe the dynamics of a driven two-level system (TLS). The celebrated LZSM equation is introduced be- fore showing that periodic driving gives rise to n-photon resonance conditions. In Sec. III we describe the singlet- triplet energy level diagram of a doubly occupied DQD, and show qualitatively how multilevel LZSM interference gives rise to harmonic generation. Finally in Sec. IV, we add the effects of lead coupling and decoherence to our model. The calculated response is compared with the experimental results and shown to reproduce the key features observed in the data [9]. LANDAU-ZENER-ST UCKELBERG-MAJORANA II. TWO-LEVEL DYNAMICS The LZSM problem describes the evolution of a TLS that is forced through an energy level anticrossing [17 -- 21]. LZSM considered a generic TLS with states 0(cid:105) and 1(cid:105), described by the Hamiltonian: (cid:19) (cid:18) 0 ∆ ∆ −
HTLS = . (1) Here the detuning parameter
sets the energy difference between the states. An off-diagonal matrix element ∆ 6 1 0 2 l u J 7 2 ] l l a h - s e m . t a m - d n o c [ 1 v 7 0 2 8 0 . 7 0 6 1 : v i X r a A. Two-level dynamics under periodic driving 2 The effects of quantum interference are revealed when the system is repeatedly driven through an anticross- ing. Consider the case of sinusoidal driving. Here
(t) =
0 +
1 sin (2πf t), where
0 is a fixed detuning set by dc gate voltages in the experiment and
1 = eVac is the amplitude of the ac drive. The ac drive results in two anticrossing traversals for each cycle of the drive field, with an approximate level velocity: (cid:115) (cid:18)
0 (cid:19)2
1 v ≈ 2π
1f 1 − . (3) Figure 1. (a) Energy level diagram described by Eq. 1. Si- nusoidal driving with
(t) =
0 +
1 sin(2πf t) causes repeated traversals of the anticrossing. (b) Energy levels as a func- tion of time for sinusoidal driving. Each passage through the anticrossing results in some population transfer, analo- gous to a beam splitter. A phase Φi is accumulated between beam splitter events. (c) Population of the 1(cid:105) state, P1(cid:105), as a function of time for two different values of
0 [pictured in (a)]. Dashed lines indicate the times of the LZSM transitions (i.e. t's such that
(t) = 0). For
0 = −39 µeV, the accumu- lated phases result in destructive interference and successive LZSM transitions cancel each other out (blue trace). With
0 = −16.5 µeV (red trace) successive LZSM transitions in- terfere constructively, resulting in nearly complete transfer of population to the 1(cid:105) state at t ≈ 2 ns. (cid:19) (cid:18) −2π ∆2 v hybridizes the levels, resulting in an anticrossing of mag- nitude 2∆ at
= 0 [Fig. 1(a)]. In the LZSM problem the energy difference between the states is varied with a linear level velocity v = dE1(cid:105)− E0(cid:105)/dt, where E1(cid:105) (E0(cid:105)) is the energy of the 1(cid:105) (0(cid:105)) state. This can be accomplished by driving the detuning according to
(t) = vt. Starting in state 0(cid:105) at time ti = −∞, the probability of remaining in state 0(cid:105) at time tf = +∞ is given by the LZSM formula [17 -- 20]: , PLZSM = exp (2) where  is the reduced Planck's constant. When v (cid:28) ∆2, PLZSM ≈ 0 and the evolution is adiabatic. The sys- tem remains in the instantaneous eigenstate. In the op- posite limit PLZSM ≈ 1 and the sudden change approxi- mation can be made. Here a TLS that starts in 0(cid:105) will remain in 0(cid:105) after the sweep through the anticrossing. With intermediate level velocities, a sweep through the anticrossing will generate a superposition of states 0(cid:105) and 1(cid:105) [22 -- 24]. This physics has been harnessed for quan- tum control in a variety of systems, including Rydberg atoms [25, 26], nitrogen vacancy centers [27, 28], and GaAs DQDs [29]. In Fig. 1(b) we plot the energy levels as a function of time for sinusoidal driving with f = 2 GHz,
1 = 100 µeV, ∆ = 4 µeV, and
0 = −16.5 µeV. A system initialized in 0(cid:105) at t = 0 will be forced through the anticrossing every time that
0 = −
1 sin (2πf t). For our driving parame- ters the first crossing happens at t = 0.01 ns. After the first sweep through the anticrossing the probability of re- maining in the 0(cid:105) state is approximately PLZSM = 0.9. The non-unity probability results in the system entering a superposition of states 0(cid:105) and 1(cid:105). After the anticrossing the states accumulate a relative phase Φ1 due to their en- ergy difference. At time t = 0.21 ns the system is forced back through the anticrossing, interfering the two-paths of the interferometer. Such interference occurs twice during each cycle and depends on the phases Φ1 and Φ2. Additionally the total accumulated phase Φ1 + Φ2 will result in interference be- tween subsequent cycles. Depending on the precise value of the phase the system will exhibit behavior ranging from constructive to destructive interference. To illus- trate this we plot the occupation of state 1(cid:105), P1(cid:105), as a function of time in Fig. 1(c). We use two different val- ues of the offset detuning
0 and numerically integrate Schrodinger's equation with the Hamiltonian in Eq. 1. For
0 = −16.5 µeV the phase accumulation results in constructive interference and P1(cid:105) oscillates between 0 and 1. For
0 = −39 µeV, however, the interference is destructive. As such the population transfer resulting from one LZSM transition is immediately canceled out by the next LZSM transition. In the fast driving limit (where 1 − PLZSM (cid:28) 1), the condition for constructive interference can be de- rived by considering the phase accumulation Φ1 occur- ring between the first and second LZSM transition and Φ2 occurring between the second and the third LZSM transition [see Fig. 1(b)]. For constructive interference to occur Φ1 − Φ2 = 2nπ for an integer n [21]. With Φi = dt, the reso- dt ≈ ti+1(cid:82) ti+1(cid:82)
0+eVac sin(2πf t) E1(cid:105)−E0(cid:105)   ti ti nance condition reduces to nhf =
0. Here ti is the time of the i-th anticrossing traversal. This can be interpreted as an n-photon resonance condition and has been ob- served in several studies on both superconducting qubits [30, 31] and GaAs charge qubits [32]. ε(t) = ε0+ε1sin(2πft) 10Φ12ΔΦ2E (µeV)ε (µeV)E (µeV)t(ns)1.00.80.20.0P4310t (ns)(a)(c)(b)1ConstructiveDestructive-1000100-10001000.62.04.00.0-100010020.40.63 five-state basis (S(1, 1)(cid:105), S(2, 0)(cid:105), T0(1, 1)(cid:105), T−(1, 1)(cid:105), T+(1, 1)(cid:105)) can be written as: HDQD =  Ez1−Ez2 0 ∆ 2 0 0 ∆ Ez1−Ez2 −
0 ∆so ∆so 2 0 0 0 0 0 ∆so 0 Ez1+Ez2 2 0 0 ∆so 0 0 − Ez1+Ez2 2  . (4) Figure 2. DQD energy levels plotted as a function of detuning
with B = 45 mT. Interdot tunnel coupling ∆ = 16.5 µeV hybridizes the S(2,0) and S(1,1) states, while the spin-orbit coupling ∆so = 4 µeV hybridizes the S(2,0) state with the T+(1,1) and T−(1,1) states. III. PERIODICALLY DRIVEN TWO-ELECTRON DOUBLE QUANTUM DOT EDSR experiments are typically performed near a Pauli-blocked interdot charge transition, where the to- tal number of electrons in the DQD is even [10, 33]. We use a five-level Hamiltonian to capture the singlet-triplet physics of this system. Starting with the DQD initial- ized in a spin-blocked triplet state, we show that multi- level LZSM interference leads to harmonics in the EDSR response near zero detuning [13]. A. Double quantum dot energy level diagram To reflect the experimental conditions we consider a spin-orbit qubit defined in an InAs nanowire, as schemat- ically shown in the inset of Fig. 2. We note that robust Pauli blockade has been observed in many experiments at higher electron occupancies [9, 34, 35]. For simplicity, we therefore consider a DQD in the two-electron regime, with one electron in each dot (1,1), or with two elec- trons in one dot, e.g. (2,0). Here we use the notation (Nl,Nr), where Nl (Nr) is the number of electrons in the left (right) dot. In the (1,1) charge configuration there are four spin states, the singlet state S(1,1) and the three triplet states T−(1,1), T0(1,1), T+(1,1), with total spin components ms = −1, 0, +1. An external magnetic field results in Zeeman splitting of the electronic spin states with Ez1 = g1µBB and Ez2 = g2µBB. The g-factors are generally different due to strong spin-orbit coupling. We set g1 = 7.8 and g2 = 6.8 to match the values mea- sured in Ref. [9]. Due to the tight electric confinement there is a large singlet-triplet splitting Est = 5.4 meV. As a result, the (2,0) triplet manifold can be neglected in most experimental situations. The Hamiltonian in the Here
is the detuning, ∆ is the interdot tunnel coupling, and ∆so generates spin-orbit anticrossings [36, 37]. The resulting energy level diagram is shown in Fig. 2 with pa- rameters ∆ = 16.5 µeV, ∆so = 4 µeV [37 -- 40]. These pa- rameters are taken from Ref. [9] and the well-established material properties of InAs [41 -- 43]. B. Time evolution of the five-level double quantum dot To illustrate the importance of LZSM dynamics we time-evolve the five-level system described by Eq. 4. This simple model reproduces the strong detuning dependence that is observed in the experimental data. The system is initialized in the T+(1, 1) state and propagated under an oscillatory detuning of the form
(t) =
0 +
1 sin (2πf t), with B = 45 mT, and
1 = 1.3 meV. Figures 3(a -- d) show the T0(1,1), T+(1,1), and S(2,0) state occupations as a function of time. For panels (a,c) we choose driv- ing that corresponds to a one-photon resonance between the T+(1,1) and T0(1,1) states. For panels (b,d) the drive corresponds to a two-photon resonance between the T+(1,1) and T0(1,1) states. In the far detuned region (
0 = 1.9 meV) the ac drive does not have a large enough amplitude to force the system through the anticrossings near
= 0. In this case population transfer into the T0(1,1) state is visible for the n = 1 harmonic, as seen in Fig. 3(a) [11]. However, the dynamics for the n = 2 resonance proceed on a significantly slower time scale, as expected from standard spin resonance theory [8]. For both the n = 1 and n = 2 resonance conditions, there is no significant population transfer into the S(2,0) state [see Fig. 3(a,b)]. When
0 (cid:46)
1 the dynamics are radically different. Here the system is repeatedly forced through the level anticrossings, causing a portion of the population to be transferred to the S(2,0) state, from which the system can make further LZSM transitions to either the T+(1,1) or T0(1,1) state. Evidence of these processes can be seen in Fig. 3(c,d). For both the n = 1 and n = 2 resonances clear population transfer is observed between the T+(1,1) and T0(1,1) states. The population transfer is mediated by the S(2,0) state, as evidenced by the periodic jumps of the S(2,0) state population. Since transitions to both the T−(1,1) and S(1,1) states are not resonant, there is no significant population transfer into them. Finally note that the timescale over which population transfer occurs InAs50 nm-0.10-0.050.000.050.10E (meV)0.20.10.0-0.1ε (meV)S(2,0)T-(1,1)T+(1,1)S(1,1)T0(1,1)2Δso2Δso-0.2Ez2=g2µBBEz1=g1µBB4 Figure 4. (a) Illustration of the charge transport cycle. Start- ing from the (1,0) charge state an electron is loaded into the right dot. If a (1,1) triplet state is loaded the transport be- comes blocked until the spin is rotated into the S(1,1) state, from which the system can tunnel to S(2,0) and then into the left lead. (b) Illustration of the various processes included in simulations of the transport dynamics. Black arrows indicate relaxation and incoherent tunneling processes, while green ar- rows indicate coherent processes that arise due to the periodic driving. state. Strong electron-phonon coupling results in fast re- laxation from the S(1,1) state to the S(2,0) state. The electron then tunnels to the left lead with rate ΓL, com- pleting the transport cycle. A. Time evolution We model the time dependence of the periodically driven system by evolving the density matrix ρ using the master equation: dρ dt = − i ΓR 4  [HDQD, ρ] + ΓLD [(1, 0)(cid:105)(cid:104)S(2, 0)] ρ+ (5) (cid:88) D [S(1, 1)(cid:105)(cid:104)(1, 0)] ρ+ D [Tj(1, 1)(cid:105)(cid:104)(1, 0)] ρ+ (cid:88) ΓspinD [S(1, 1)(cid:105)(cid:104)Tj(1, 1)] ρ+ j j ΓS(2,0)D [S(2, 0)(cid:105)(cid:104)S(2, 0)] ρ+ Θ (
) ΓchargeD [S(2, 0)(cid:105)(cid:104)S(1, 1)] ρ+ Θ (−
) ΓchargeD [S(1, 1)(cid:105)(cid:104)S(2, 0)] ρ. Here D [A] ρ = −1/2{A†A, ρ} + AρA† is the Lindblad superoperator describing relaxation and decoherence, j spans the ms = 0, +1, and −1 triplet states. Θ(x) is the Heaviside step function. As shown in Fig. 4(b), terms Figure 3. S(2,0), T0(1,1), and T+(1,1) occupation probabil- ities plotted as a function of time with
1 = 1.3 meV and B = 45 mT. The system is initialized in the T+(1,1) state at t = 0. In (a,c) f is chosen to drive a n = 1 photon EDSR pro- cess between the T+(1,1) and T0(1,1) states, while for (b,d) f is chosen to drive a n = 2 photon EDSR process between the T+(1,1) and T0(1,1) states. There is a remarkable differ- ence between the dynamics when the levels are far detuned [
0 = 1.9 meV in (a,b)] compared to when the levels are near zero detuning [
0 = -0.05 meV (c,d)]. Near zero detuning, spin transfer processes occur with significant population transfer into the S(2,0) state, and occur on a much faster timescale than in the case of far detuning. Since in all cases the S(1,1) and T−(1,1) states (not shown) are not on resonance, there is no significant population transfer into these states. is much shorter when
0 (cid:46)
1. IV. TRANSPORT CYCLE In the previously reported experiments [9], the EDSR response is detected by measuring the dc current through the DQD. The DQD is configured at finite bias in Pauli blockade. Resonant ac driving rotates the electronic spin states, lifting the Pauli blockade, resulting in a small, but measurable current [10, 33]. To make a quantitative comparison with experiment we therefore model the full transport cycle of the DQD. As seen in Fig. 4(a), starting from the empty (1,0) state, tunneling from the right lead results in the (1,1) charge configuration. If a polarized (1,1) triplet state is loaded, the transport cycle becomes blocked as tunneling into the (2,0) charge configuration is forbidden by the Pauli exclusion principle [44]. The Pauli blockade can be lifted by driving an EDSR tran- sition, which rotates a blocked (1,1) state [T−(1,1) or T+(1,1)] to an unblocked state [S(1,1) or T0(1,1)]. In our case T0(1,1) is unblocked due to the difference in g-factors, which leads to further rotation to the S(1,1) 1.00.50.01.00.50.0(a)(c)(b)ε0 = -0.05 meVε0 = 1.9 meVPPt(ns)t(ns)T0(1,1)T+(1,1)S(2,0)T0(1,1)T+(1,1)S(2,0)T0(1,1)T+(1,1)S(2,0)T0(1,1)T+(1,1)S(2,0)3020100n = 13020100n = 2(d)(1,0)T+(1,1)T0(1,1)T-(1,1)S(1,1)S(2,0)ΓR4ΓchargeΓLΓspin(b)Incoherent RelaxationCoherent Processes(a)ΓRΓLEDSRΓR4ΓR4ΓR45 fied in Sec. III. For each value of B and f we initialize the system in the (1,0) state. We then numerically propagate the system in time until the system reaches a steady state (typically after 20 ns of evolution). We can then write the current as I = eΓRP(1,0), where P(1,0) is the extracted steady state population of the (1,0) state. We note that for typical drive parameters the minimum time between the T+(1,1)↔S(2,0) and the S(2,0)↔T0(1,1) LZSM tran- sitions is on the order of a picosecond. As a result, the S(2,0) state can still act as an intermediary that trans- fers population between the other levels, despite the large relaxation and decoherence rates. In Figs. 5(a,b) we compare the spectroscopic data ob- tained in the experiments with our model. Figure 5(a) shows the data that were obtained with
0 = 0. Here the current I is plotted as a function of magnetic field strength B and the applied excitation frequency f . For n = 1 two distinct resonance lines of increased current are visible. These correspond to Ez1 (g1 = 7.8) and Ez2 (g2 = 6.8). Higher photon transitions display a striking odd-even dependence. Multiphoton resonances up to n = 8 are observed. In Fig. 5(b) we plot the current I as a function of B and f , as calculated by the model described above. Following previous work, we include the effects of quasi-static charge noise by using Gaussian smoothing of the response [40, 45, 46]. In this plot the effects of charge noise are included by sampling 30 different randomly cho- sen offset detunings and weighing the final response with a Gaussian of width σcharge = 60 µeV centered around
0 = 0. The effects of the fluctuating nuclear field are in- cluded by smoothing the response in B with a Gaussian of width 3.3 mT, which is the fluctuating Overhauser field measured in Ref. [9]. This Overhauser field is consistent with other values reported in the literature [47]. The results of our model replicate the overall struc- ture of the experimental data. Both the large number of higher photon transitions and the odd-even dependence of the leakage current are in qualitative agreement with the data. At a finer level, there are some slight deviations between the theoretical predictions and the experimental data. The observed current is in general higher then our model predicts. We attribute this to imperfect fitting of the tunneling rates. The simulations also exhibit faint high frequency oscillations (oriented horizontally in the figure) that are largely independent of B. We attribute this to the imperfect modeling of the charge noise. The experimental data also exhibit a characteristic de- tuning dependence. To make a valid comparison with experiments, we define ∆Ires as the change between res- onant and non-resonant leakage current. In the experi- ment this quantity was obtained by measuring the cur- rent along nhf = g1µBB and subtracting from it the current found approximately 5 mT away [9]. Figure 6 plots ∆Ires for
1 = 1.3 meV and f = 4.7 GHz (succes- sive harmonics are achieved by increasing B). The data points are adapted from Ref. [9], while solid lines show the calculated ∆Ires from the model. We note that all experimentally observed features are reproduced. First, Figure 5. (a) Measured current I through the device as a function of magnetic field B and frequency f . With the DQD configured at zero detuning a large number of multiphoton resonances are observed. Adapted from Ref. [9]. (b) Calcu- lated current through the DQD as a function of B and f . The color-scale axes are slightly different since the off-resonance leakage current measured in experiment is device specific. with ΓL (ΓR) account for coupling to the left (right) lead: the ΓL term relaxes the S(2,0) state into the empty (1,0) state, while the ΓR term moves the population from the empty state to one of the four (1,1) spin states. Note that our model assumes unpolarized lead tunneling. There- fore the tunneling probability into any of the (1,1) spin states is equal. Γcharge is included to account for charge relaxation, which is known to take place on nanosecond timescales in semiconductor DQDs. For
> 0, Γcharge relaxes the S(1,1) state into the S(2,0) state, while for
< 0 this process is reversed. This ensures that charge relaxation only takes place from a state of higher energy to a state of lower energy. Γspin models spin relaxation, which is relatively slow in semiconductor DQDs. Finally ΓS(2,0) [not pictured in Fig. 4(b)] results in charge deco- herence. B. Charge transport and the role of decoherence We now simulate the experimental system using realis- tic parameters to account for tunnel coupling to the leads, charge noise, and dephasing. To match typical experi- mental conditions, we set ΓL = ΓR = 2 GHz, Γcharge = 1 GHz, ΓS(2,0) = 10 GHz, and Γspin = 1 MHz. The param- eters in the Hamiltonian describing the DQD are speci- I (pA)f (GHz)B (mT)8642864215010050I (pA)f (GHz)(b)(a)n = 1 n = 2 n = 3 n = 4Ez1n = 1 n = 2 n = 3 n = 4302010302010Ez2V. CONCLUSION 6 We have shown that the multiphoton resonances re- cently observed in EDSR experiments are due to multi- level LZSM interference. The fact that these high order processes are possible raises several intriguing possibili- ties. For example, since the mechanism for population transfer is quite distinct from traditional Rabi oscilla- tions, one could obtain very fast population transfer, an attractive proposition for quantum manipulation [48, 49]. With smaller charge noise it would also be possible to per- form a direct measurement of the spin-orbit gaps [50] and investigate the interplay of the spin-orbit and hyperfine interactions, both of which open gaps between the S(2,0) state and the T+(1,1) and T−(1,1) states [51]. Finally we note that so far experiments studying LZSM processes have focused on the zero detuning region near the singlet state anticrossing. However, similar behavior should also be observable near anticrossings with the states in the (0,2) triplet manifold. Due to the use of transport as a probe of spin states, these have so far been experimen- tally inaccessible. However, recent developments of fast cavity based readout [40, 52] should make this exciting regime within reach of experimental studies. ACKNOWLEDGMENTS We thank Sorawis Sangtawesin for assistance develop- ing the simulation code. This research is funded by the Gordon and Betty Moore Foundation's EPiQS Initiative through Grant GBMF4535, with partial support from the National Science Foundation (DMR-1409556 and DMR- 1420541). MZM and MHD acknowledge support from Fapesp and INCT-DISSE/CNPq, Brazil. Devices were fabricated in the Princeton University Quantum Device Nanofabrication Laboratory. Figure 6. Measured and simulated ∆Ires (resonant change of current) as a function of
0 for (a) n = 1, (b) n = 2, (c) n = 3, (d) and n = 4. the odd-even dependence is evident. Near zero detuning ∆Ires has a maximum (minimum) for odd (even) photon resonances. Second, the number of oscillations in ∆Ires increases with n, as observed in experiment. Lastly, the magnitude of ∆Ires is in good agreement with the data. [1] P. A. Franken, A. E. Hill, C. W. Peters, and G. Weinre- [11] V. N. Golovach, M. Borhani, and D. Loss, Phys. Rev. B ich, Phys. Rev. Lett. 7, 118 (1961). 74, 165319 (2006). [2] M. O. Scully and M. S. Zubairy, Quantum Optics (Cam- [12] C. Flindt, A. S. Sørensen, and K. Flensberg, Phys. Rev. bridge University Press, Cambridge, England, 1997). Lett. 97, 240501 (2006). [3] I. D. Abella, Phys. Rev. Lett. 9, 453 (1962). [4] W. Kaiser and C. G. B. Garrett, Phys. Rev. Lett. 7, 229 (1961). [5] B. Zaks, R. B. Liu, and M. S. Sherwin, Nature (London) 483, 580 (2012). [6] S. Nadj-Perge, V. S. Pribiag, J. W. G. van den Berg, K. Zuo, S. R. Plissard, E. P. A. M. Bakkers, S. M. Frolov, and L. P. Kouwenhoven, Phys. Rev. Lett. 108, 166801 (2012). [7] E. A. Laird, C. Barthel, E. I. Rashba, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Semicond. Sci. Tech- nol. 24, 064004 (2009). [8] B. D. Cohen-Tannoudji, C. and F. Laloe, Quantum Me- [13] J. Danon and M. S. Rudner, Phys. Rev. Lett. 113, 247002 (2014). [14] A. V. Shytov, Phys. Rev. A 70, 052708 (2004). [15] E. Temchenko, S. Shevchenko, and A. Omelyanchouk, Phys. Rev. B 83, 144507 (2011). [16] J. R. Petta, A. C. Johnson, J. M. Taylor, E. A. Laird, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Science 309, 2180 (2005). [17] L. Landau, Phys. Z. Sowjetunion 2, 46 (1932). [18] C. Zener, Proc. R. Sco. London Ser. A 137, 696 (1932). [19] E. Stuckleberg, Helv. Phys. Acta 5, 36 (1932). [20] E. Majorana, Nuovo Cimento 9, 43 (1932). [21] S. Shevchenko, S. Ashhab, and F. Nori, Phys. Rep. 492, chanics Volume One (Wiley, New York, 1977). 1 (2010). [9] J. Stehlik, M. D. Schroer, M. Z. Maialle, M. H. Degani, [22] Y. Gefen and D. J. Thouless, Phys. Rev. Lett. 59, 1752 and J. R. Petta, Phys. Rev. Lett. 112, 227601 (2014). (1987). [10] S. Nadj-Perge, S. M. Frolov, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Nature (London) 468, 1084 (2010). [23] M. Sillanpaa, T. Lehtinen, A. Paila, Y. Makhlin, and P. Hakonen, Phys. Rev. Lett. 96, 187002 (2006). -2-1012-4-202-2-1012ε0 (meV)-6-4-202∆Ires (pA)∆Ires (pA)(a)(b)(c)(d)ε0 (meV)7 [24] Y. Kayanuma, Phys. Rev. A 55, R2495 (1997). [25] M. C. Baruch and T. F. Gallagher, Phys. Rev. Lett. 68, 3515 (1992). [26] S. Yoakum, L. Sirko, and P. M. Koch, Phys. Rev. Lett. 69, 1919 (1992). Mod. Phys. 75, 1 (2002). [40] K. D. Petersson, L. W. McFaul, M. D. Schroer, M. Jung, J. M. Taylor, A. A. Houck, and J. R. Petta, Nature (London) 490, 380 (2012). [41] J. Singh, Physics of semiconductors and their het- [27] L. Childress and J. McIntyre, Phys. Rev. A 82, 033839 erostructures (McGraw-Hill, 1993). (2010). [28] J. Zhou, P. Huang, Q. Zhang, Z. Wang, T. Tan, X. Xu, F. Shi, X. Rong, S. Ashhab, and J. Du, Phys. Rev. Lett. 112, 010503 (2014). [42] D. L. Perry, Handbook of inorganic compounds, 2nd ed. (CRC Press, 2011). [43] D. Liang and X. P. Gao, Nano Lett. 12, 3263 (2012). [44] K. Ono, D. G. Austing, Y. Tokura, and S. Tarucha, [29] J. R. Petta, H. Lu, and A. C. Gossard, Science 327, 669 Science 297, 1313 (2002). (2010). [45] K. D. Petersson, J. R. Petta, H. Lu, and A. C. Gossard, [30] W. D. Oliver, Y. Yu, J. C. Lee, K. K. Berggren, L. S. Phys. Rev. Lett. 105, 246804 (2010). Levitov, and T. P. Orlando, Science 310, 1653 (2005). [46] Y.-Y. Liu, K. D. Petersson, J. Stehlik, J. M. Taylor, and [31] D. M. Berns, M. S. Rudner, S. O. Valenzuela, K. K. Berggren, W. D. Oliver, L. S. Levitov, and T. P. Or- lando, Nature (London) 455, 51 (2008). [32] J. Stehlik, Y. Dovzhenko, J. R. Petta, J. R. Johansson, F. Nori, H. Lu, and A. C. Gossard, Phys. Rev. B 86, 121303 (2012). [33] K. C. Nowack, F. H. L. Koppens, Y. V. Nazarov, and L. M. K. Vandersypen, Science 318, 1430 (2007). [34] A. C. Johnson, J. R. Petta, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 72, 165308 (2005). [35] E. Nielsen, E. Barnes, J. P. Kestner, and S. Das Sarma, Phys. Rev. B 88, 195131 (2013). [36] V. N. Golovach, A. Khaetskii, and D. Loss, Phys. Rev. B 77, 045328 (2008). [37] S. Nadj-Perge, S. M. Frolov, J. W. W. van Tilburg, J. Danon, Y. V. Nazarov, R. Algra, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Phys. Rev. B 81, 201305 (2010). [38] F. H. L. Koppens, C. Buizert, I. T. Vink, K. C. Nowack, T. Meunier, L. P. Kouwenhoven, and L. M. K. Vander- sypen, J. Appl. Phys. 101, 081706 (2007). [39] W. G. van der Wiel, S. De Franceschi, J. M. Elzerman, T. Fujisawa, S. Tarucha, and L. P. Kouwenhoven, Rev. J. R. Petta, Phys. Rev. Lett. 113, 036801 (2014). [47] M. D. Schroer, K. D. Petersson, M. Jung, and J. R. Petta, Phys. Rev. Lett. 107, 176811 (2011). [48] J. W. G. van den Berg, S. Nadj-Perge, V. S. Pribiag, S. R. Plissard, E. P. A. M. Bakkers, S. M. Frolov, and L. P. Kouwenhoven, Phys. Rev. Lett. 110, 066806 (2013). [49] L. Gaudreau, G. Granger, A. Kam, G. C. Aers, S. A. Studenikin, P. Zawadzki, M. Pioro-Ladriere, Z. R. Wasilewski, and A. S. Sachrajda, Nature Phys. 8, 54 (2012). [50] M. D. Shulman, S. P. Harvey, J. M. Nichol, S. D. Bartlett, and A. Yacoby, Nature A. C. Doherty, V. Umansky, Comm. 5, 5156 (2014). [51] J. M. Nichol, S. P. Harvey, M. D. Shulman, A. Pal, V. Umansky, E. I. Rashba, B. I. Halperin, and A. Ya- coby, Nature Comm. 6, 7682 (2015). [52] J. Stehlik, Y.-Y. Liu, C. M. Quintana, C. Eichler, T. R. Hartke, and J. R. Petta, Phys. Rev. Appl. 4, 014018 (2015). 

1111.1591

2

1111

Noise thermometry in narrow 2D electron gas heat baths connected to a quasi-1D interferometer

Thermal voltage noise measurements are performed in order to determine the electron temperature in nanopatterned channels of a GaAs/AlGaAs heterostructure at bath temperatures of 4.2 and 1.4 K. Two narrow two-dimensional (2D) heating channels, close to the transition to the one-dimensional (1D) regime, are connected by a quasi-1D quantum interferometer. Under dc current heating of the electrons in one heating channel, we perform cross-correlated noise measurements locally in the directly heated channel and nonlocally in the other channel, which is indirectly heated by hot electron diffusion across the quasi-1D connection. We observe the same functional dependence of the thermal noise on the heating current. The temperature dependence of the electron energy-loss rate is reduced compared to wider 2D systems. In the quantum interferometer, we show the decoherence due to the diffusion of hot electrons from the heating channel into the quasi-1D system, which causes a thermal gradient.

cond-mat.mes-hall

cond-mat

Noise thermometry in narrow 2D electron gas heat baths connected to a quasi-1D interferometer Sven S. Buchholz,1, 2, ∗ Elmar Sternemann,3, 2 Olivio Chiatti,1 Dirk Reuter,4 Andreas D. Wieck,4 and Saskia F. Fischer1, 2, † 1Neue Materialien, Institut fur Physik, Humboldt-Universitat zu Berlin, 12489 Berlin, Germany 2Werkstoffe und Nanoelektronik, Ruhr-Universitat Bochum, 44780 Bochum, Germany 3Experimentelle Physik 2, Technische Universitat Dortmund, 44227 Dortmund, Germany 4Angewandte Festkorperphysik, Ruhr-Universitat Bochum, 44780 Bochum, Germany (Dated: June 15, 2021) Thermal voltage noise measurements are performed in order to determine the electron temperature in nanopatterned channels of a GaAs/AlGaAs heterostructure at bath temperatures of 4.2 and 1.4 K. Two narrow two-dimensional (2D) heating channels are connected by a quasi-1D quantum interferometer. Under dc current heating of the electrons in one heating channel, we perform cross-correlated noise measurements locally in the directly heated channel and nonlocally in the other channel, which is indirectly heated by hot electron diffusion across the quasi-1D connection. The temperature dependence of the electron energy-loss rate is reduced compared to wider 2D systems. Under nonlocal current heating, which establishes a thermal gradient across the quantum interferometer, we show the decoherence in this structure by Aharonov-Bohm measurements. PACS numbers: 63.20.kd, 72.20.Pa, 73.23.Ad, 85.35.Ds I. INTRODUCTION In recent years, research activities in the field of nano- structured materials have been increasingly focused on thermoelectric properties and thermal non-equilibrium.1 In particular, the creation and detection of thermal gradients and the determination of lattice and charge carrier temperatures at the nanoscale remain crucial issues. Semiconductor heterostructures may be pre- pared as model systems for thermoelectric investiga- tions representing two-dimensional (2D), 1D, and 0D charge carrier systems. Electrical thermometry methods have been implemented on the basis of resistance and mobility measurements,2 -- 4 Shubnikov-de Haas (SdH) measurements,5 -- 9 quantum point contacts (QPCs),9 -- 12 and quantum dots (QDs)13 -- 15 for such low-dimensional systems. However, most of these methods are applicable only at temperatures below roughly 20 K. Thermometry via mobility measurements is limited due to the contri- bution of impurity scattering,6,12 and SdH measurements require a magnetic field which alters the density of states and possibly the energy relaxation.9 SdH, QPC and QD thermometry are limited to low lattice temperatures due to thermal smearing of the discrete energy states. Applying the above thermometry methods enables to study the charge carrier energy-loss to the lattice, which provides fundamental information about electron-phonon interactions. Whereas much effort has been made in the field of 2D charge carrier systems, mostly in the form of wide Hall bar structures,8,9,12 only a few experiments focus on the transition to 1D systems where electron- phonon interactions may be altered.16 -- 18 Here, we apply electronic noise measurements as a di- rect method for the determination of the charge carrier temperature in a particularly narrow 2D GaAs/AlGaAs structure. Thermal (Johnson-Nyquist) noise measure- ments are applicable to different materials with a wide range of operating temperatures, such as diffu- sive metal films and wires,19,20 and semiconductors host- ing high mobility 3D,21 2D,16 and (quasi)-1D electronic systems.16,22 We fabricated a device consisting of a quasi-1D quan- tum interferometer integrated between two narrow 2D heating channels in order to create a temperature dif- ference between the electron reservoirs (heat baths). In these reservoirs, we measured thermal noise locally and nonlocally. The interferometer allows us to study the in- fluence of nonlocal heating on the coherence of electrons in the quantum structure. The electron system is heated to a temperature above the lattice temperature by means of the current heating technique.10,23,24 We extract the electron temperature in the narrow 2D channels for different heating currents, and we find a reduced temperature dependence of the electron energy-loss rate compared to wide 2D electron gases (2DEGs). Resistor network simulations of the ther- mal noise allow us to determine noise contributions of individual parts of the sample and of the external cir- cuitry. Additionally, by Aharonov-Bohm (AB) measure- ments, we show a decoherence effect in the interferometer on the basis of nonlocal current heating. II. EXPERIMENTAL DETAILS A schematic of the device is depicted in Fig. 1(a). The two narrow 2D heating channels are nominally identical and connected by a quasi-1D quantum ring. The device was prepared from a GaAs/AlGaAs heterostructure with a 2DEG 110 nm below the surface, using electron beam lithography and 85-nm-deep wet-chemical etching. The 2D electron density and mobility are ns = 2.07·1011 cm−2 and µ = 2.43 · 106 cm2/Vs at T = 4.2 K in the dark. The heating channels - labeled 'heater I' and 'heater II' in Fig. 1(b) - are geometrically 2 µm wide and 410 µm long. This ensures a high thermal noise signal over the background of the total parasitic noise. SdH measure- ments along the heating channels yield an electron den- sity of ns = 1.84 · 1011 cm−2, where we attribute the deviation from the above-mentioned sheet density to the lateral confinement of 2 µm. A non-alloyed gold flake (not shown) on heating channel II remained from the lift- off process but does not influence the electron transport properties. The quantum wires defining the ring (Fig. 1(c)) and its leads to the heating channels are geometrically 570 nm wide. From separate measurements on simple quantum wires (widths 350 to 550 nm), as well as four-terminal resistance measurements along the quantum ring, we es- timate that about 10 modes of the quasi-1D subband structure are populated in the quantum ring in equilib- rium. Noise measurements were performed in a 4He cryostat at bath temperatures of Tbath ≥ 1.4 K and recorded with an Agilent 89410A spectrum analyzer. At T = 1.4 K, the heating channels have a four-terminal resistance of Rh ≈ 6 kΩ, which corresponds to a Nyquist noise of less than 10−18 V2/Hz. In order to increase the noise signal of the heating channels above the noise of the spectrum analyzer (≈ 10−16 V2/Hz), we used two low-noise voltage preamplifiers with a voltage gain of 103 (Signal Recovery 5184). Cross-correlated measurements were applied to reduce noise contributions from the preamplifiers.25 We measured the noise spectrum along heater I in two different heating current setups allowing for local (Fig. 1(d)) and nonlocal (Fig. 1(e)) heating. In the local setup, we measured the thermal noise of heater I to which the heating current was applied. In the nonlocal setup, thermal noise was measured in heater I, while the heating current was driven through heater II on the other side of the quantum ring. The heating current was applied via a battery-driven voltage source with a 1 MΩ series resistor on each side of the source. A 1 µF capacitor to ground was attached on each side in order to reduce parasitic coupling effects. We chose the resistance of the heating channel such that its noise contribution exceeds that of the remain- ing measurement circuitry. The length of the heating channels exceeds the electron phase coherence and en- ergy relaxation lengths, which yields a diffusive trans- port regime, where the electron temperature Te can be deduced from the thermal white noise SV,w by means of the Nyquist formula. If heated by a current, the electrons are no longer in equilibrium with the lattice and share the heat energy among themselves through electron-electron interactions. Energy relaxation takes place via phonon emission and the diffusion to cold reservoirs. In order to estimate individual thermal noise contri- butions, we simulated the sample and the circuitry in a SPICE model (Cadence PSpice). Next to the individual resistive and capacitive parts of the wiring and circuitry, 2 FIG. 1: (Color online) (a) Sample layout (to scale) with four- terminal heating channels as meander structures to the left and the right of the quantum ring. (b) Microscopic photo- graph of the sample as indicated in (a). (c) Atomic force microscope image of the quantum ring. (d,e) Schemes of the measurement setups (not to scale) for (d) local and (e) non- local heating via Ih and noise measurement Vx. the sample was segmented into discrete resistors account- ing for the individual Ohmic parts of the sample, which were determined experimentally by lock-in measurements at the different bath temperatures. Noise spectra were recorded for bath temperatures in the range of Tbath = 1.4 - 10 K and different local and nonlocal heating currents for a frequency range of 1 Hz to 20 kHz. A recorded voltage noise spectrum SV (f ) results from an average of typically 700 sets of data. In order to obtain the thermal noise of a heating chan- nel SV,h or the equivalent electron temperature Te from the total noise spectrum, we analyzed the data as follows: (a) In order to take parasitic capacities Cpar into account, the measured noise SV,m(f ) was corrected by a first-order low-pass SV (f ) = SV,m(f )(1 + (2πRCpar)2). We deter- mined Cpar = 800 nF and 565 nF from experiments with and without the heating circuit attached, respectively. (b) The total white noise SV,w was extracted as the aver- age value of SV (f ) in a frequency range, where no 1/f - noise was visible, i.e. from f = 15 kHz to 19.25 kHz. (c) We subtracted noise contributions of Ohmic contacts and the wide 2DEG leads SV,Ω, the heating circuitry SV,hc, and the preamplifiers SV,amp by comparison of SV,w(Ih (cid:54)= 0) with SV,w(Ih = 0). SV,amp results from (a)(b)ABCVxIhVxIh(a)VxIhVxIh(c)(d)20 µmheater Iheater IIquantum ring(b)2 µm1234(d)(e)20 µmheater Iheater IIquantumring(b)2 µm1234(a)(c)VxIhVxIhALT:(b)(a)ABCABC32143214the finite current noise of the preamplifiers. All parasitic contributions S0 add to the thermal noise of the heater SV,h: SV,w = SV,h + S0 = SV,h + SV,Ω + SV,hc + SV,amp 3 = 4kBTeRh,eff + 4kBTbathRΩ,eff + SV,hc + 4kBTamp2R−1 ampR2 in, (1) where the latter term is valid for Rin << Ramp. Rin is the total input resistance, and Rh,eff and RΩ,eff are the effective heater and lead (ohmic contacts and wide 2DEG regions) resistances as seen by the preamplifiers. The SPICE network analysis reveals that we can assume Rh,eff ≈ Rh, RΩ,eff ≈ RΩ and Rin ≈ Rh + RΩ with a maximum error of less than 2 %. The preamplifiers with an input resistance of Ramp = 5 MΩ operate at Tamp = 300 K, and the factor 2 accounts for the two preamplifiers. At Tbath = 4.2 K we determined Rh = 6.2 kΩ and RΩ = 1.4 kΩ. The noise contribution of the heating circuit in the lo- cal setup was determined by the simulation as Ssim V,hc = 0.78 · 10−18 V2/Hz, which is in good agreement with the V,hc = 0.73 · 10−18 V2/Hz experimental observations of Sexp V,hc = 0.81· 10−18 V2/Hz at Tbath = 4.2 and 1.4 K, and Sexp respectively (see Fig. 2). With the above resistance val- ues and Eq. 1, we expect a total white noise at 4.2 K with the heating circuit attached in the local setup but Ih = 0 of SV,w = 2.93 · 10−18 V2/Hz, which agrees well with the result of the full simulation SV,w = 2.91 · 10−18 V2/Hz. III. EXPERIMENTAL RESULTS AND DISCUSSION The determined total white noise SV,w for different heating currents in the local and the nonlocal setup is shown in Fig. 2(a) for Tbath = 4.2 K and in Fig. 2(b) for Tbath = 1.4 K. The size of the symbols represents the measurement accuracy. We ascertained that the heater resistance does not change with the heating current. The inset of Fig. 2(a) depicts typical noise spectra SV (f ) for different local heating currents after lowpass correction. For applied heating currents, a 1/f γ depen- dence (γ ≤ 1.3) is visible at f < 13 kHz. At higher frequencies, white noise dominates. With increasing cur- rent, both 1/f and white noise components increase. Here, the 1/f noise components will not be discussed, but we will consider the white noise component. Fig. 2 shows the experimentally determined white noise values Sexp V,w in the local (filled squares and circles) and nonlocal (empty squares and circles) heating setup, as well as fits to the experimental data (broken lines) and simulated Ssim V,w values (triangles). Measured and simu- lated data without the heating circuit attached are given by symbols labeled A. Values labeled B and C were mea- sured and simulated with the heating circuit attached in the nonlocal and the local setup, respectively. The sig- nificant increase of SV,w with the connection of the heat- ing circuit results from the biasing resistors as discussed FIG. 2: Measurements with parabolic fits and simulations of the voltage noise SV,w for local and nonlocal electron heating. (a) Data at Tbath = 4.2 K and (b) Tbath = 1.4 K. The filled and empty symbols represent measured values for local and nonlocal heating, respectively. The half-filled symbols (indi- cated by A) depict measurements without the heating circuit attached. Triangles display results from SPICE simulations, and the broken lines are parabolic fits to the measured data. above. We will first discuss the data at Ih = 0. In order to evaluate the influence of the heating circuit on the to- tal noise quantitatively, we simulated the sample and the measurement circuitry as a resistor network, as explained above, and analyzed the thermal noise. All simulated data points Ssim V,w at A, B and C are slightly lower than the corresponding measured values Sexp V,w, with a maxi- mum deviation of 8%. This systematic error corresponds to the thermal noise of a resistance < 10 Ω at room tem- perature and may result from parts of the measurement setup which were not accounted for in the simulation, such as the cables or a higher Ramp than specified. The good agreement of Ssim V,w shows that the SPICE simulation is a helpful method to investigate the total noise, as well as noise contributions of single components of a complex resistor network. V,w and Sexp (a)(b)ABCVxIhVxIh(a)VxIhVxIh(c)(d)20 µmheater Iheater IIquantum ring(b)2 µm1234(d)(e)20 µmheater Iheater IIquantumring(b)2 µmr1r2r3r4(a)(c)VxIhVxIhALT:(b)(a)ABCABC4 FIG. 3: (a) Electron temperature Te as a function of the heating current Ih in the local setup for Tbath = 4.2 and 1.4 K. The symbols represent data derived from noise measurements, whereas the lines result from parabolic fits. (b) The difference between the electron and the bath temperature ∆Te = Te − Tbath as a function of the dissipated power per electron Pe, for Tbath = 4.2 and 1.4 K (local heating setup). The line results from the parabolic fit for Tbath = 4.2 K. By applying heating currents, the white noise SV,w(Ih (cid:54)= 0) increases with Ih for both local and non- local heating at Tbath = 4.2 and 1.4 K. The parabolic best fits in Fig. 2 (broken lines) indicate a quadratic de- pendence ∆SV,w ∝ I 2 h. In the local heating setup, the electrons of heater I are heated directly. In contrast, un- der nonlocal heating, the electrons of heater I are heated by hot electron diffusion from heater II across the quan- tum ring. Phonon mediated contributions to heat trans- fer were investigated in an additional device (same het- erostructure), which possesses a gate electrode to locally deplete the electron system between two heating chan- nels. In this device, a noise increase under nonlocal cur- rent heating could only be observed if the electron system was conducting. Once the electron system was depleted, no noise dependence on the heating current was detected. Hence, we expect phonon mediated heat transfer to be negligible in our devices at such low temperatures. Fig. 3(a) depicts Te as a function of Ih for the local heating setup, as determined from Eq. 1. In addition to the measurement data (symbols), parabolic fits are displayed. In a first approximation, we observe ∆Te ∝ I 2 h, as expected for Joule heating and observed in other GaAs/AlGaAs 2DEGs.12,26,27 For the nonlocal heating setup, we do not attempt an analogous determination of Te due to the unknown temperature gradient in the narrow channel. Commonly, the current is converted to the dissipated power per electron Pe = I 2 hRh/(nsAh) with Ah the In a steady state, Pe is area of the heating channel. taken as the net power transfer from the electrons to the lattice Pe = Q(Te) − Q(Tlat),8,9,12,16 where Tlat is the lattice temperature and Q denotes the energy-loss rate. Fig. 3(b) shows the electron temperature increase ∆Te = Te − Tbath as a function of Pe in a full-logarithmic graph for the local setup. The line gives the parabolic fit for Tbath = 4.2 K. While ∆Te scales approximately linearly with Pe for low heating powers, it deviates sig- FIG. 4: Dissipated power per electron Pe as a function of the experimentally determined electron temperature Te in the local heating setup for Tbath = 4.2 K and 1.4 K. The broken lines are best fits to Pe = A(T 2.2 e − T 2.2 bath). e −T n nificantly for Pe > 7 · 103 eV/s, since ∆Te << Tlat is no longer satisfied, as discussed previously.12 We point out that the temperature increase is almost independent of the bath temperature (Tbath = 4.2 and 1.4 K). The ab- solute electron heating in the chosen Pe-range is in good agreement with data determined via SdH measurements in GaAs/AlGaAs low dimensional electron systems.5,16 In order to investigate the scattering mechanisms in the heated electron channel, we plot the experimentally determined data for local heating as Pe(Te) in Fig. 4. Here, the broken lines are fits to Pe = A(T n bath) (two- bath-model2) and yield approximately n = 2.2 for both Tbath, and A = 985 and A = 300 eV/sK2.2 for Tbath = 4.2 and 1.4 K, respectively. For Tbath = 1.4 K, the data for Te > 4 K deviate from the fit and were not included in the fit. We assume Tbath = Tlat since the device is thermally well anchored. From the exponent of T n in the two-bath- model, information about electron-phonon interactions can be deduced, as discussed in detail elsewhere.8,9,12 Whereas investigations of the electron energy-loss rates in GaAs/AlGaAs 2DEGs at Tlat = 1 - 10 K by SdH mea- surements typically yield a T 2 or T 3-dependence5,8,9, ex- periments on the basis of QPC thermometry showed a T 5-behavior in the same temperature regime, as well as for temperatures in the mK-range.9,12 From the latter ex- periments, it was deduced that the dominant scattering mechanisms in the Gruneisen-Bloch regime are acoustic phonon scattering via a screened piezoelectric potential9 or an unscreened deformation potential.12 n = 3 sug- gests scattering via an unscreened piezoelectric potential. n = 2, if diffusion to cold contacts dominates the energy relaxation.9 n = 1 is observed in the equipartition regime at higher temperatures of around 20 - 40 K.8 The T 2.2-dependence observed here may suggest that electrons and phonons interact in an intermediate state between the Gruneisen-Bloch and the equipartition regime. However, we point out that the 2 µm wide heating channels are particularly narrow compared with those investigated previously. Investigations of electron- (b)(a)ABCABC(a)(b)(b)(a)ABCABC(a)(b)phonon interactions in 2DEGs were performed in struc- tures of typically several 10 µm width or wider.8,9,12 In GaAs based devices, structures of a width less than roughly 1 µm show signatures of the 1D regime in trans- port, where samples of width between 50 and 300 nm yield clear quantized conductance at temperatures of a few K.28,29 Structures of several µm width, on the other hand, are distinct 2D systems whose fundamental param- eters are investigated by SdH and quantum Hall measure- ments. A system of 2 µm width, as investigated here, re- veals its 2D character in simple transport measurements since the subband separation of its 1D energy states is in the range of tens to hundreds of µeV and thus below the thermal smearing. However, the density of states may be slightly altered toward 1D characteristics, similar to the low-filed SdH effect. One signature for a 2D regime close to the transition to 1D is the decreased electron density in the narrow heating channels compared to wide 2D re- gions, as described in Sec. II. For 1D systems, a reduction of the power law exponent n in the Gruneisen-Bloch regime was predicted30 -- 33 and experimentally observed in etched InGaAs wires (etching widths between 25 nm and 1 µm).17,18 Under perpendic- ular magnetic fields, a reduction of n in a 2DEG has been observed, and it was suggested that the momentum transfer of electron-phonon interactions was restricted due to Landau Levels.9 In analogy to this magnetic con- finement of the electron-phonon interactions, in 1D sys- tems the electronic confinement with its associated mod- ification of the density of states could restrict electron scattering and thus modify the temperature dependence of electron-phonon interactions.31,32 In the etched, nar- row heating channels investigated here, this effect may play a role. We exclude a significant contribution to the energy relaxation by the wide 2DEG leading to the Ohmic contacts since the heating channel length exceeds the hot electron diffusion length. However, the system- atic investigation of the energy-loss rate dependence on the channel width, ranging from wide 2D- to narrow 1D- systems, remains a prospect for future experiments. For the investigation of the impact of nonlocal cur- rent heating on decoherence, we applied a heating current in one heating channel and detected the decoherence by AB measurements in the quantum interferometer. The sample was cooled down in a dilution refrigerator to the base temperature of Tbath = 20 mK. Noise measure- ments could not be performed in this cryostat. While driving the dc heating current Ih through heater I, we measured the quantum ring's four-terminal resistance R41,32 = V32/I41 by lock-in technique as a function of the perpendicular magnetic field B, exploiting the AB interference effect.34 -- 36 Fig. 5(a) shows the oscillatory part of the magnetore- sistance R41,32 for three different Ih after subtraction of the background resistance. For all traces, the resistance oscillates regularly with B with an h/e-period, in accor- dance to the quantum ring geometry. It is noteworthy that the oscillation phase does not change for all applied 5 FIG. 5: Aharonov-Bohm interference measurements and analysis under current heating through heater I. (a) Oscil- latory part of the four-terminal magnetoresistance R41,32 for different heating currents Ih, offset for clarity. (b) Interfer- ence visibility v as a function of Ih. The error amounts to about ±0.3%. R41,32 was measured in a perpendicular mag- netic field B at Tbath = 20 mK. Ih, i.e. phase rigidity is not lifted by current heating in this setup.36 With increasing Ih, the oscillation am- plitude decreases until the AB oscillations are fully sup- pressed for Ih > 1.2 µA. The AB oscillation visibility v = (Rmax − Rmin)/(Rmax + Rmin) is plotted as a function of Ih in Fig. 5(b). The approximately linear decrease of v with Ih in the semi-logarithmic graph suggests the re- lation v(Ih) = v0 exp(−αhIh) = v0 exp(−L/Lφ), with αh the fitting parameter, L the mean length of the inter- ferometer arms, and Lφ the electron phase breaking length.34,36 Under an applied heating current through heater I, hot electrons are created in this reservoir next to the quantum wire interferometer, and a temperature differ- ence between the two heat baths (channel 1 and 2) is invoked (thermal gradient). This leads to the diffusion of hot electrons across the quasi-1D structure, which raises the local electron temperature in the interfero- meter. Previous works have investigated electron dephas- ing in quasi-1D AB interferometers by an increase of the lattice temperature and have shown the relation v(Tlat) = v0 exp(−αT Tlat), which leads to Lφ ∝ T −1 lat .34 -- 36 Here, we find Lφ ∝ I−1 h , in analogy, and conclude that dephasing with Ih is induced by the elevated electron temperature in the interferometer via increased electron-electron scat- tering and thermal averaging.34 -- 36 IV. SUMMARY In conclusion, we have applied voltage noise thermom- etry to a particularly narrow 2D GaAs/AlGaAs elec- tron system at bath temperatures of 4.2 and 1.4 K. Electrons were heated by the application of a dc cur- rent, and the thermal noise was measured in a cross- correlation setup. The device consists of two heating channels (heat baths) with a quasi-1D quantum inter- ferometer in-between. We performed local noise measure- ments in the directly heated channel and nonlocal mea- surements in the indirectly heated channel, and we found (b)(a)ABCABC(a)(b)(a)(b)6 the same functional dependence of the thermal noise on the heating current in the local and the nonlocal heating setup. The indirect heating is explained by hot electron diffusion through the quasi-1D interferometer. The tem- perature dependence of the electron energy-loss rate of T 2.2 is lower than that observed in previous investiga- tions at 2DEGs. This may result from the confinement of the electron system to a narrow 2DEG with a slightly al- tered density of states evoking restrictions in phase space of electron-phonon interactions. We demonstrate that an indirect current heating can be successfully employed as a means to establish thermal gradients between heat baths connected to (quasi-) 1D quantum circuits. The effect of indirect current heating on the electron deco- herence in the quantum interferometer was investigated by Aharonov-Bohm measurements, and an exponential decay of the visibility with an increasing heating current was observed. Acknowledgments The authors gratefully acknowledge financial sup- port from the Deutsche Forschungsgemeinschaft (DFG) within the priority program SPP1386. S.F.F. is grate- ful for support from the 'Junges Kolleg,' North Rhine- Westphalia Academy for Sciences and Arts. D.R. and A.D.W. acknowledge support from the DFG SPP1285 and the BMBF QuaHL-Rep 01BQ1035. We greatly ap- preciate the scientific and technical support from Prof. U. Kunze. ∗ electronic address: [email protected] † electronic address: [email protected] 1 K. Nielsch, J. Bachmann, J. Kimling, and H. Bottner, Adv. Engery Mater. 1, 713 (2011). 2 A. K. M. Wennberg, S. N. Ytterboe, C. M. Gould, H. M. Bozler, J. Klem, and H. Morkoc, Phys. Rev. B 34, 4409 (1986). 3 H. L. Stormer, L. N. Pfeiffer, K. W. Baldwin, and K. W. West, Phys. Rev. B 41, 1278 (1990). 4 A. Mittal, R. G. Wheeler, M. W. Keller, D. E. Prober, Lett. 81, 727 (2002). 18 C. Prasad, D. K. Ferry, and H. H. Wieder, Semicond. Sci. Technol. 19, S60 (2004). 19 M. L. Roukes, M. R. Freeman, R. S. Germain, R. C. Richardson, and M. B. Ketchen, Phys. Rev. Lett. 55, 422 (1985). 20 M. Henny, H. Birk, R. Huber, C. Strunk, A. Bachtold, M. Kruger, and C. Schonenberger, Appl. Phys. Lett. 71, 773 (1997). 21 T. A. Eckhause, O. Sulzer, C. Kurdak, F. Yun, and and R. N. Sacks, Surf. Sci. 361/362, 537 (1996). H. Morkoc, Appl. Phys. Lett. 82, 3035 (2003). 5 K. Hirakawa and H. Sasaki, Appl. Phys. Lett. 49, 889 22 A. Kumar, L. Saminadayar, D. C. Glattli, Y. Jin, and (1986). B. Etienne, Phys. Rev. Lett. 76, 2778 (1996). 6 S. J. Manion, M. Artaki, M. A. Emanuel, J. J. Coleman, 23 R. T. Syme, M. J. Kelly, and M. Pepper, J. Phys.: Con- and K. Hess, Phys. Rev. B 35, 9203 (1987). 7 Y. Ma, R. Fletcher, E. Zaremba, M. D'Iorio, C. T. Foxon, and J. J. Harris, Phys. Rev. B 43, 9033 (1991). 8 for a review see: B. K. Ridley, Rep. Prog. Phys. 54, 169 (1991). 9 N. J. Appleyard, J. T. Nicholls, M. Y. Simmons, W. R. Tribe, and M. Pepper, Phys. Rev. Lett. 81, 3491 (1998). 10 L. W. Molenkamp, H. van Houten, C. W. J. Beenakker, R. Eppenga, and C. T. Foxon, Phys. Rev. Lett. 65, 1052 (1990). 11 O. Chiatti, J. T. Nicholls, Y. Y. Proskuryakov, N. Lump- kin, I. Farrer, and D. A. Ritchie, Phys. Rev. Lett. 97, 056601 (2006). 12 Y. Y. Proskuryakov, J. T. Nicholls, D. I. Hadji-Ristic, and C. B. Sorensen, Phys. Rev. B 75, A. Kristensen, 045308 (2007). 13 A. A. M. Staring, L. W. Molenkamp, B. W. Alphenaar, H. van Houten, O. J. A. Buyk, M. A. A. Mabesoone, C. W. J. Beenakker, and C. T. Foxon, Europhys. Lett. 22, 57 (1993). 14 R. Scheibner, H. Buhmann, D. Reuter, M. N. Kiselev, and L. W. Molenkamp, Phys. Rev. Lett. 95, 176602 (2005). 15 E. A. Hoffmann, H. A. Nilsson, J. E. Matthews, and N. Nakpathomkun, A. I. Persson, L. Samuelson, H. Linke, Nano Lett. 9, 779 (2009). 16 C. Kurdak, D. C. Tsui, S. Parihar, S. A. Lyon, and M. Shayegan, Appl. Phys. Lett. 67, 386 (1995). 17 T. Sugaya, J. P. Bird, D. K. Ferry, A. Sergeev, V. Mitin, K.-Y. Jang, M. Ogura, and Y. Sugiyama, Appl. Phys. dens. Matter 1, 3375 (1989). 24 B. L. Gallagher, T. Galloway, P. Beton, J. P. Oxley, S. P. Beaumont, S. Thoms, and C. D. W. Wilkinson, Phys. Rev. Lett. 64, 2058 (1990). 25 M. Sampietro, L. Fasoli, and G. Ferrari, Rev. Sci. Instrum. 70, 2520 (1999). 26 L. W. Molenkamp and M. J. M. de Jong, Phys. Rev. B 49, 5038 (1994). 27 S. Maximov, M. Gbordzoe, H. Buhmann, L. W. Molenkamp, and D. Reuter, Phys. Rev. B 70, 121308(R) (2004). 28 A. Kristensen, J. B. Jensen, M. Zaffalon, C. B. Sorensen, S. M. Reimann, and P. E. Lindelof, J. Appl. Phys. 83, 607 (1998). 29 G. Apetrii, S. Fischer, U. Kunze, D. Reuter, and A. Wieck, Semicond. Sci. Technol. 17, 735 (2002). 30 U. Bockelmann and G. Bastard, Phys. Rev. B 42, 8947 (1990). 31 A. Y. Shik and L. J. Challis, Phys. Rev. B 47, 2082 (1993). 32 S. Das Sarma and V. B. Campos, Phys. Rev. B 47, 3728 (1993). 33 S. S. Kubakaddi, Phys. Rev. B 75, 075309 (2007). 34 A. E. Hansen, A. Kristensen, S. Pedersen, C. B. Soorensen, and P. E. Lindelof, Phys. Rev. B 64, 045327 (2001). 35 K.-T. Lin, Y. Lin, C. C. Chi, J. C. Chen, T. Ueda, and S. Komiyama, Phys. Rev. B 81, 035312 (2010). 36 S. S. Buchholz, S. F. Fischer, U. Kunze, M. Bell, D. Reuter, and A. D. Wieck, Phys. Rev. B 82, 045432 (2010). 

1004.4913

3

1004

A six degree of freedom nanomanipulator design based on carbon nanotube bundles

Scanning probe imaging and manipulation of matter is of crucial importance for nanoscale science and technology. However, its resolution and ability to manipulate matter at the atomic scale is limited by rather poor control over the fine structure of the probe. In the present communication, a strategy is proposed to construct a molecular nanomanipulator from ultrathin single-walled carbon nanotubes. Covalent modification of a nanotube cap at predetermined atomic sites makes the nanotube act as a support for a functional "tool-tip" molecule. Then, a small bundle of nanotubes (3 or 4) with aligned ends can act as an extremely high aspect ratio parallel nanomanipulator for a suspended molecule, where protraction or retraction of individual nanotubes results in controlled tilting of the tool-tip in two dimensions. Together with the usual SPM three degrees of freedom and augmented with rotation of the system as a whole, the design offers six degrees of freedom for imaging and manipulation of matter with precision and freedom so much needed for advanced nanotechnology. A similar design might be possible to implement with other high-aspect ratio nanostructures, such as oxide nanowires.

cond-mat.mes-hall

cond-mat

A six degree of freedom nanomanipulator design based on carbon nanotube bundles Vasilii I Artyukhov Institute of Biochemical Physics, Russian Academy of Sciences Kosygin st. 4, Moscow, 119334 Russia E-mail: [email protected] Abstract. Scanning probe imaging and manipulation of matter is of crucial importance for nanoscale science and technology. However, its resolution and ability to manipulate matter at the atomic scale is limited by rather poor control over the fine structure of the probe. In the present communication, a strategy is proposed to construct a molecular nanomanipulator from ultrathin single-walled carbon nanotubes. Covalent modification of a nanotube cap at predetermined atomic sites makes the nanotube act as a support for a functional ―tool-tip‖ molecule. Then, a small bundle of nanotubes (3 or 4) with aligned ends can act as an extremely high aspect ratio parallel nanomanipulator for a suspended molecule, where protraction or retraction of individual nanotubes results in controlled tilting of the tool-tip in two dimensions. Together with the usual SPM three degrees of freedom and augmented with rotation of the system as a whole, the design offers six degrees of freedom for imaging and manipulation of matter with precision and freedom so much needed for advanced nanotechnology. A similar design might be possible to implement with other high -aspect ratio nanostructures, such as oxide nanowires. PACS: 81.16.Ta, 82.37.Gk, 81.16.-c, 68.37.Ef, 89.20.Kk 1. Introduction Scanning probe imaging and manipulation of matter crucially relies on the quality of tips, specifically, on their aspect ratio, which determines the spatial resolution, and their wearing behavior, which determines the reliability and lifetime of a probe. From this point of view, carbon nanotubes (particularly, single-wall nanotubes [1, 2]) are the ideal candidate due to their small and uniform diameter and extremely strong graphitic bonds between constituent atoms. Moreover, carbon nanotubes can be either semiconducting or metallic, and the latter is crucial for scanning tunneling microscopy and related atomic manipulation techniques. Immediately after the first demonstration of carbon nanotube usage as scanning microscope probes in 1996 [3], the field literally exploded with various applications and modifications of the technology. Lithography with carbon nanotubes was demonstrated in 1998 [4]. Manipulation of nanoscale objects by multiple-nanotube devices was achieved as early as 1999 [5], promptly followed by fabrication of nanostructures via controlled deposition of specific length segments of the nanotube tips [6]. Covalently functionalized carbon nanotube tips were used to demonstrate scanning chemical force microscopy [7] and chemically sensitive tunneling microscopy [8]. Magnetic force microscopy can be performed using carbon nanotubes functionalized with magnetic metal nanoclusters [9]. Pristine and chemically modified carbon nanotube tips have earned their place as an important tool for scanning probe microscopy in both physical [10] and biological science [11]. However, their application is still limited by various factors: on one hand, the difficulties of positioning and attachment of nanotubes to tips complicate the process of tip fabrication and attachment of the nanotube to the cantilever is often not too reliable, which makes conventional silicon probe technology more practical for most current applications where the extreme aspect ratio of the probe offered by nanotubes is not absolutely essential; on the other hand, there exist fundamental issues such as the fact that the exact type of grown nanotubes is hard to control, and the spatial unpredictability of covalent functionalization, meaning that it is impossible to tell in advance the exact location of the functional group or molecule. In the following, a strategy is proposed to circumvent the latter problem, which ultimately results in the proposal of a novel family of nanoscale parallel manipulators with two extra degrees of freedom compared to the traditional scanning probe technology, hoping that projected gains from such tools can outweigh the associated short-term practical difficulties and stimulate further research effort to mitigate these problems . 2. Nanomanipulator design 2.1. Functionalization of nanotubes at predetermined sites: the isolated pentagon rule It is well known that in order to form a closed surface, an sp2 carbon nanostructure must contain 12 pentagons that create positive surface curvature; therefore, a hemispherical nanotube cap contains 6 pentagons. The angle of an equilateral pentagon is 108°, which deviates strongly from the optimal value of 120° for sp2 hybridization of carbon, and is in fact closer to the 109.5° value that is typical for tetrahedral sp3 hybridization. Therefore, an atom that belongs to two or three pentagons at the same time is ―forced‖ into an sp3-like state by this geometrical strain, which creates effectively unpaired electron density at the atom. Such an atom would be prone to become engaged in addition reactions so that its chemical environment builds up to an sp3 tetrahedron. This result is well-known as the ―isolated pentagon rule‖ (IPR). It explains why the smallest stable fullerene is C60, and it only exists in the truncated-icosahedron C60 isomeric form that obeys the IPR (i.e., all pentagons are separated by hexagons). Smaller fullerenes , such as the C20 dodecahedron containing only pentagons, demonstrate poor stability; at the same time, the dodecahedrane molecule (C20H20) is stable. Figure 1. (6,0) nanotube caps made from the C28 fullerene: (a-d) construction of the fullerene starting from three pentagons sharing a common atom; (e) cutting the fullerene into two caps and insertion of a (6,0) nanotube segment inbetween. The upper cap contains the three-pentagon cluster. Since C60 is the smallest fullerene that has isomers obeying the IPR, any nanotube with a diameter smaller than 0.68 nm (the diameter of C60) will inevitably contain edge- or vertex-sharing pentagons in its cap. These sites will be the most chemically active, making them prime candidates for chemical functionalization. We can further imagine a situation where a nanotube cap contains a unique most preferred spot for covalent functionalization; for this, we need to recall the structure of the C28 fullerene, which contains four clusters of three pentagons sharing a vertex (Td point symmetry group). The clusters are linked by common pentagon edges. The process of constructing a C28 fullerene is shown in four steps starting from our cluster of interest in figure 1 (a–d). The final structure can be sliced in half so that two (6,0) nanotube caps are formed; this is shown in figure 1 (e), where a segment of the nanotube has been inserted between the halves. The upper fragment has a threefold rotational axis (C3v group), while the symmetry of the lower is sixfold (C6v). Since no experimental data exist on the geometry of caps (or whether at all caps can be stable) in such small-diameter nanotubes, the question of which of the two possible cap structures is the correct one has to be addressed using quantum chemical calculations. The relative stability of the two different cap structures can be roughly judged from the deformation energies of corresponding fullerenes that are made by joining two caps of each type. The upper cap structure corresponds to one of the isomers of C44, the lower corresponds to C36. Calculations (see table 1) show that our cap structure is preferred for the (6,0) nanotube since it minimizes the energy per atom: it has one atom in the most unfavorable position possible—shared by three pentagons—but this is overcompensated by there being fewer edge-sharing pentagons (3 common edges instead of 6). It should also be noted that the ―bottom‖ cap can be converted into a ―top‖ cap by adding 4 carbon atoms. The data in table 1 demonstrate that the energies-per-atom of such caps come quite near to that of the (stable) C60 fullerene; chemical functionalization of the tip atom should be expected to substantially stabilize the C3v cap structure even further. Table 1. Relative energies of possible (6,0) nanotube caps. Binding energy per atom (kJ/mol)a Fullerene C28 (figure 1d) C36 (two C6v caps [12]) C40 (one cap of each type) C44 (two C3v caps) –671.3 –691.3 –691.4 –695.8 C60 –720.0 a Calculations were performed using the PBE density functional [13] with an optimized triple-zeta Gaussian basis set. Thus, the (6,0) carbon nanotube is the perfect candidate for use as a chemically functionalized probe since its cap contains a single site, referred to as the ―tip atom‖ in the following, that is especially susceptible to chemical functionalization. The diameter of this nanotube is 0.47 nm, suggesting a very favorable aspect ratio. Incidentally, this tube is metallic, meaning that the operation of the functional group at the tip could be controlled by applied voltage. Another point to note with regard to the (6,0) nanotube is its good lattice match with the (111) diamond surface [14], suggesting the use of diamond cantilevers to support the nanotubes. As for other nanotube types, the complete inventory of carbon nanotube cap structures (up to a diameter of 3 nm) available in the literature [15] should facilitate the search for other systems that offer at least partial site-specificity of chemical functionalization and hence could be used in other designs resembling the present. To conclude the section dedicated to the isolated pentagon rule, it should be noted that substitution of carbon with nitrogen, in a sense, reverses the IPR. The nitrogen atom has an extra electron compared to carbon, and therefore, instead of an effectively unpaired electron, a nitrogen atom shared by three pentagons presents a lone pair. Such a configuration is, in fact, preferential for N-substituted fullerenes [16]. This means that in a capped (6,0) nanotube, a nitrogen atom would predominantly occupy the position at the very tip. While this would actually prevent covalent functionalization of the nanotube at the tip, the nitrogen atom could form a dative bond with, e.g., a boron atom, opening up broad possibilities for reversible assembly and disassembly of complex architectures with boron - substituted carbon nanostructures or with boron nitride structures. Finally, similar arguments should generally hold for boron substitution of nanotubes, i.e., boron atoms should also predominantly occupy the tip position in a capped (6,0) nanotube. 2.2. Parallel nanomanipulators based on nanotube bundles While site-specific functionalization of carbon nanotubes may be interesting for imaging and manipulation of matter at the scale of individual atoms, the above strategy becomes effectively useless if deterministic manipulation of multi-atom molecules is required: free rotation of the grafted molecule about the sp3 bond would render null all the efforts to lock it into the desired location. However, this problem can be solved by simultaneously attaching the molecule to three (or more) nanotubes forming a bundle. As an illustration, an adamantane molecule attached to a bundle of three capped (6,0) nanotubes is shown in figure 2 (a). This molecule was chosen for its tetrahedral shape giving it three natural linking sites and a fourth that remains free for subsequent modification. The relatively small size of the molecule means that intermediary linkers are required. A number of possible linking configurations using carbon, silicon, germanium, oxygen and sulfur atoms were tested; from these, alkane chains with three carbon atoms (i.e., propane) appear to be most suitable, long enough to make up for the small size of the molecule compared to the gap between nanotube tips, but not too long so that the molecule is kept in place tightly. Figure 2. Carbon nanotube bundles covalently grafted with (a) an adamantane molecule and with (b) a carbon dimer deposition tool based on two face-joined adamantane molecules. Dative attachment is also possible (c, d). Boron and nitrogen atoms are shown in green and blue, respectively; violet represents germanium. It is also immediately seen that in such a configuration, relative protraction or retraction of individual nanotubes can tilt the molecule with two angular degrees of freedom. The nanotubes in the bundle are kept together by attractive dispersion forces, but the relative sliding should be easy. Therefore, three-site grafting of the functional molecule, in fact, converts a simple probe into a full-fledged parallel nanomanipulator (see figure 3). Another reason why adamantane has been chosen as the molecule of interest in the present study is the recent proposal of a minimal toolset for positionally controlled diamond mechanosynthesis by Freitas and Merkle [17]. Among the 9 proposed functional molecules (―tooltips‖), 7 represent adamantane molecules with appropriate substitutions of either hydrogen or carbon at one vertex of the molecule, and yet another one is basically a combination of two such tooltips. The final member of the set is the dimer placement tool (DimerP) [18, 19] based on two face-joined Ge-substituted adamantane molecules (this corresponds to a diamond crystal twin boundary) shown in figure 2b suspended on four (6,0) nanotubes. In this case, the molecule connects to the inner nanotubes via cyclopentane rings; furthering the twin boundary analogy, each ring can be viewed as dual propane chains sharing one end and joined at the other. Examples of the same two molecules linked via dative B–N bonds are also provided in figure 2 (c, d). Notice that while adamantane is supported on three nitrogen-substituted nanotube caps, the mirror symmetry of the DimerP molecule would cause problems if boron substitution were to be used in the cyclopentane rings; however, using two boron-substituted nanotubes solves the problem. In these systems, the functional molecules are bound less strongly than in the covalent case. On the other hand, tooltips can be changed relatively easily. One can immediately see that it is straightforward to use various combinations of boron/nitrogen substitutions to design complementary tools that bind selectively to their intended counterparts, which might turn out useful in future complex architectures. Figure 3. Angular flexibility of the manipulator: (a) starting position; (b) one nanotube protracted by ca. 0.5 nm; (c) another nanotube protracted by ca. 0.2 nm. The corresponding strain energies are listed in table 2. Although a detailed technical assessment of the performance of the manipulators in terms of range of motion, positional and angular uncertainty, etc. is beyond the scope of the present communication, some estimate of the amount of strain present in the systems is nevertheless needed to check if these structures could at all exist. Strain energies were (very roughly) estimated using the MM2 molecular mechanics force field [20], and the results are summarized in table 2. It can be seen that, although a certain amount of strain is present in all structures, it is insufficient to cause bond rupture, especially considering the fact that it is distributed over 3 (for adamantane) or 4 (for DimerP) links. This also means that bond configurations in the structures are not too unusual, and the use of molecular mechanics (avoiding expensive quantum-chemical calculations) is justified in this case. In summary, these results show that the designs in figure 2 may be feasible, at least from the thermodynamical point of view, and should work as expected (figure 3). Table 2. Strain energy estimates. Energy (kJ/mol)a Structure Adamantane (untilted) Adamantane (tilted, figure 3b) Adamantane (tilted, figure 3c) DimerP (untilted) 232 280 260 307 Typical alkane C-C bond strength 350 a Energies are listed with respect to nonfunctionalized nanotube bundles and molecules with pre-attached linkers. 3. Discussion 3.1. Implementation pathways Before the implications of the above designs can be discussed, possible strategies of fabricating the proposed structures have to be reviewed. This includes synthesizing the required components and assembling them into a working system. As of present, carbon nanotube probes are typically grown in situ on SPM tips using some variation of chemical vapor deposition process [21], with the possibility of even wafer-scale fabrication [22]. However, CVD-grown nanotubes typically have diameters > 1 nm, which is too large for our purposes, and their type is hard to control precisely. On the other hand, ultrathin SWCNTs down to 0.4 nm diameter can be selectively grown inside zeolite pores [23–26], or inside larger diameter CNTs [27] with the possibility of controlling the type of as-grown nanotube by the choice of catalyst type and external conditions [28]. The inner tube could subsequently be extracted from the resulting double-wall nanotube by mechanical means (so-called ―sword-in-sheath‖ failure of the outer wall) [29] or, for example, using electrical current heating [30]. Even if the nanotubes are grown uncapped, it should nevertheless be possible to close their ends; on-demand capping of carbon nanotubes has previously been demonstrated, at least, for multiwall carbon nanotubes [31]. Given all the difficulties of fabrication and processing of ultrathin carbon nanotubes, it might be desirable to use nanocones [32] or conically-terminated multiwall nanotubes, since these structures can have very sharp tips [33] with clusters of pentagons. Although chemical modification of nanocones is much less explored compared to nanotubes, quantum chemical calculations [34] suggest that functionalization of nanocones should occur predominantly at the tip, offering at least some spatial control over functionalization. Finally, it should be noted that perfect control over the functionalization site is not an absolute necessity: techniques such as field emission measurements with a second movable probe [35] could in principle be utilized to determine functional group position after the functionalization has been carried out, thus enabling the use of other carbon nanostructures besides the (6,0) nanotube. Individual as-grown carbon nanotubes will then have to be transferred onto separate actuators, and their free ends joined together to form a self-supporting bundle. Although in principle, just one degree of freedom per actuator should be sufficient—provided that the tubes are long enough (or the actuators close enough) so that they can be joined using an additional 3-dof manipulator,—assembly would be most easily done if all actuators had three degrees of freedom. This suggests the use of a three-probe scanning microscope design for the first demonstrations of the devices; it should then be noted that in the present case, steric hindrance constraints that plague conventional multiprobe instrument design are somewhat relaxed, because the probes only need to approach each other to a certain distance (determined by the length of the nanotubes); nor need they be parallel or coplanar, providing additional design flexibility to reduce steric congestion. After the bundle has been formed, for example, DNA hairpins [36] could be used to make a ―knot‖ clipping the bundle together and allowing individual nanotubes to be routed to their independent actuators, although it is quite probable that it would be sufficient to simply rely on the mutual attraction of carbon nanotubes. Any excess length of the nanotubes could be trimmed in situ with, e.g., an electron beam [37]. Covalent functionalization of carbon nanotubes is a well-established procedure, and, as long as there exists a preferred spot of functionalization on the nanotube, no insurmountable obstacles are to be expected from this side. Similarly, organic synthesis methods are more than capable of producing molecules with appropriate linkers attached, as long as a desired functional molecule has been chosen. The possibility of successful synthesis of particular tooltip molecules discussed above has already been addressed in the corresponding references. Given the freedom to choose the functional groups on both sides—nanotube tips and the molecules—it appears that the rest (putting the functionalized molecule on a pre-assembled functionalized nanotube bundle) is also within the reach of scanning probe manipulation technologies. 3.2. Design variations In the above, only three degrees of freedom of the systems have been explicitly considered, namely, those associated with the vertical (along the bundle axis) translations of individual nanotubes. These correspond to the vertical translation of the system as a whole and two Euler angles. Three more degrees of freedom that have to be introduced ‗externally‘ are two horizontal translations and the rotation about the bundle axis. These could be split between the manipulator and the substrate, hopefully providing some room to simplify manipulator design. Moreover, one nanotube could be kept fixed to save some complexity on its actuator: the nanotube actuators appear to be the most troublesome spot of the whole system since they have to be made both very small and very precise, and transferring this degree of freedom to the substrate or to manipulator suspension could be very helpful from the engineering point of view. Alternatively, nanotube ends could be statically anchored on a single platform having two ―tilt‖ degrees of freedom. Overall, given the recent progress in the design of complex and precise nanomanipulators (including multistage systems [38]) and multiprobe instruments, it appears that one or another solution to these problems could be found in the near future. A greater range of angular flexibility could be achieved by placing nanotubes at an angle to each other instead of the parallel bundle alignment discussed above. In fact, such a setup could remove the need for long floppy alkane chains to link small molecules (figure 4a), or, at least, allow one to use shorter and simpler linkers (figure 4b). This flexibility, however, would have to come at a substantial additional expense: first, much more precise control of individual nanotubes would be required; second, the resulting pyramidal shape would increase the effective volume of the system and cause additional steric hindrance, which may be critical in certain cases. Figure 4. An adamantane molecule suspended on three converging nanotubes (a) directly and (b) via simple ester (C-O-C) bridges. It should also be noted that the 4-nanotube design shown in figure 2b has one additional degree of freedom that corresponds to movement of outer and inner nanotube pairs in opposing directions. This corresponds to stretching or compression of the grafted molecule. In the present case, this effect could possibly be utilized to additionally enhance the reliability of dimer transfer either from the tool or onto the tool (when recharging), although the high stiffness of the tip molecule would probably preclude any substantial effect. More sophisticated designs could put this additional degree of freedom to use in mechanically controlled chemical reactions, or in even finer mechanical manipulation of individual molecules. As an alternative to carbon nanotubes, other atomically-precise structural elements could be used. Recent examples include silica (0.3-0.4 nm) [39] and titania (0.4-0.5 nm) [40] nanowires. Such structures may even possess certain advantages over carbon nanotubes, such as piezo- or ferroelectricity, as well as there being fewer competing sites of possible covalent functionalization, making the assembly of complex architectures easier. Finally, besides stiff nanotubes and nanowires, more flexible chainlike structures might be utilized in future designs to build bendable manipulators. Fullerene–carbyne composite chains are one possible example [41].  4. Conclusions The present communication describes a class of nanoscale parallel manipulators based on carbon nanotube bundles. The manipulators offer precise control over the position and orientation of individual molecules, thanks to the well-defined structure of constituent nanotubes and to the two additional degrees of freedom that such systems provide, compared to regular scanning probes. An important step is the choice of carbon nanotube type so as to achieve tip functionalization at predictable atomic sites. Functional molecules can then be attached by either strong covalent C-C bonds or reversible dative bonds between substitutional B and N atoms in the parts of the assembly. The designs have been demonstrated to be thermodynamically feasible, and pathways that might eventually lead to their practical implementation have been suggested. In particular, techniques to extract ultrathin carbon nanotubes from zeolite pores, or some alternative methods of free-standing ultrathin nanotube synthesis, would be desirable. Although manipulators such as those described above can be expected to substantially improve the spatial resolution of scanning probe microscopy, the true diversity of potential applications comes from the various kinds of functional molecules that they can support . Even without the possibility to actuate individual nanotubes, rigid locking of the molecules in place will enable improved control over their position and orientation, making this approach far superior to single-nanotube imaging and manipulation [42] in terms of both versatility and precision. Here, designs that can support all 9 tooltips from the minimal toolset for positionally controlled diamond mechanosynthesis [17] have been provided. If built, they may serve as stepping stones from current scanning probe technology towards more efficient autonomous positioning systems [43] required for high-throughput deterministic manipulation of matter at the atomic scale, ultimately leading to the much anticipated prospects of machine-phase diamond [44] and graphitic [45] nanotechnology. Although the research into application of carbon nanotubes in scanning probe technologies appears to have slowed down due to practical difficulties, hopefully, the benefits from the present proposal can outweigh these and trigger further attempts to advance the needed prerequisite techniques, or stimulate the exploration of other possible ways to produce the proposed tools, possibly including some of the alternatives suggested in this communication. Acknowledgments The structures were designed with the NanoEngineer-1 package (http://www.nanoengineer-1.com/). MM2 calculations were performed using the TINKER package (http://dasher.wustl.edu/tinker/). Density functional calculations were performed with the Priroda program [46]. The visualization was done using QuteMol (http://qutemol.sourceforge.net/) [47]. The author thanks D. A. Medvedev, L. A. Chernozatonskii and I. V. Artyuhov for inspiring discussions and useful suggestions on the proposal and the manuscript. References: [1] Iijima S and Ichihashi T 1993 Single-shell carbon nanotubes of 1-nm diameter Nature 384 603-5 [2] Bethune D S, Klang C H, de Vries M S, Gorman G, Savoy R, Vazquez J and Beyers R 1993 Cobalt- catalysed growth of carbon nanotubes with single -atomic-layer walls Nature 384 605-7 [3] Dai H, Hafner J H, Rinzler A G, Colbert D T and Smalley R E 1996 Nanotubes as nanoprobes in scanning probe microscopy Nature 384 147-50 [4] Dai H, Franklin N and Han J 1998 Exploiting the properties of carbon nanotubes for nanolithography Appl. Phys. Lett. 73 1508-10 [5] Kim P and Lieber C M 1999 Nanotube Nanotweezers Science 286 2148-50 [6] Cheung C L, Hafner J H, Odom T W, Kim K and Lieber C M 2000 Growth and fabrication with single-walled carbon nanotube probe microscopy tips Appl. Phys. Lett. 76, 3136-9 [7] Wong S S, Wooley A T, Joselevich E, Cheung C L and Lieber C M 1998 Covalently-Functionalized Single-Walled Carbon Nanotube Probe Tips for Chemical Force Microscopy J. Am. Chem. Soc. 120 8557-8 [8] Nishino T, Ito T and Umezawa Y 2002 Carbon Nanotube Scanning Tunneling Microscopy Tips for Chemically Selective Imaging Anal. Chem. 74 4275-8 [9] Arie T, Nishijima H, Akita S and Nakayama Y 2000 Carbon-nanotube probe equipped magnetic force microscope J. Vac. Sci. Technol. B 18 104-6 [10] Nguyen C V, Chao K-J, Stevens R M D, Delzeit L, Cassell A, Han J and Meyyappan M 2001 Carbon nanotube tip probes: stability and lateral resolution in scanning probe microscopy and application to surface science in semiconductors Nanotechnology 12 363-7 [11] Woolley A T, Cheung C L, Hafner J H and Lieber C H 2000 Structural biology with carbon nanotube AFM probes Chemistry & Biology 7 R193-204 [12] Gal'pern E G, Stankevich I V, Chistyakov A L and Chernozatonskii L A 1992 Atomic and electronic structure of the barrelenes b—Cm with m = 36 + 12n. JETP Lett. 55 483-6 [13] Perdew J P, Burke K and Ernzerhof M 1996 Generalized Gradient Approximation Made Simple Phys. Rev. Lett. 77 3865-8 Perdew J P, Burke K and Ernzerhof M 1997 Phys. Rev. Lett. 78 1396 [14] Shenderova O A, Areshkin D and Brenner D W 2003 Carbon Based Nanostructures: Diamond Clusters Structured with Nanotubes Mater. Res. 6 11-7 [15] Brinkmann G, Fowler P W, Manolopoulos D E and Palser A H R 1999 A census of nanotube caps Chem. Phys. Lett. 315 335-47 [16] Ewels C 2006 Nitrogen Violation of the Isolated Pentagon Rule Nano Lett. 6 890-5 [17] Freitas Jr R A and Merkle R C 2008 A minimal toolset for positional diamond mechanosynthesis J. Comput. Theor. Nanosci. 5 760-861 [18] Merkle R C and Freitas Jr R A 2003 Theoretical analysis of a carbon-carbon dimer placement tool for diamond mechanosynthesis J. Nanosci. Nanotechnol. 3 319-324 [19] Freitas Jr R A, Allis D G and Merkle R C 2007 Horizontal Ge-Substituted Polymantane-Based C2 Dimer Placement Tooltip Motifs for Diamond Mechanosynthesis J. Comput. Theor. Nanosci. 4 433-442 [20] Allinger N L 1977 Conformational analysis. 130. MM2. A hydrocarbon force field utilizing V1 and V2 torsional terms J. Am. Chem. Soc. 99 8127 [21] Hafner J H, Cheung C L and Lieber C M 1999 Direct Growth of Single -Walled Carbon Nanotube Scanning Probe Microscopy Tips J. Am. Chem. Soc. 121 9750-1 [22] Yenilmez E, Wang Q, Chen R J, Wang D, Dai H 2002 Wafer scale production of carbon nanotube scanning probe tips for atomic force microscopy Appl. Phys. Lett. 80 2225-7 [23] Hulman M, Kuzmany H, Dubay O, Kresse G, Li L and Tang Z K 2003 Raman spectroscopy of template grown single wall carbon nanotubes in zeolite crystals J. Chem. Phys. 119 3384-90 [24] Munoz E, Coutinho D, Reidy R F, Zakhidov A, Zhou W, Balkus K J 2004 Synthesis of DAM-1 molecular sieves containing single walled carbon nanotubes Microporous and Mesoporous Mater. 67 61-65 [25] Guo K, Yang C, Li Z M, Bai M, Liu H J, Li G D, Wang E G, Chan C T, Tang Z K, Ge W K and Xiao X 2004 Efficient Visible Photoluminescence from Carbon Nanotubes in Zeolite Templates Phys. Rev. Lett. 93 017402(4) [26] Lortz R, Zhang Q, Shi W, Ye J T, Qiu C, Wang Z, He H, Sheng P, Qian T, Tang Z, Wang N, Zhang X, Wang J and Chan C T 2008 Superconducting characteristics of 4-Å carbon nanotube–zeolite composite Proc. Nat. Acad. Sci. 106 7299-303 [27] Shiozawa H, Pichler T, Grüneis A, Pfeiffer R, Kuzmany H, Liu Z, Suenaga K and Kataura H 2008 A Catalytic Reaction Inside a Single-Walled Carbon Nanotube Adv. Mater. 20 1443-49 [28] Shiozawa H, Silva S R P, Liu Z, Suenaga K, Kataura H, Kramberger C, Pfei ffer R, Kuzmany H and Pichler T 2010 Catalyst and Diameter Dependent Growth of Carbon Nanotubes Determined Through Nano Test Tube Chemistry XXIVth International Winterschool on Electronic Properties of Novel Materials: abstract book p 157 [29] Yu M-F, Lourie O, Dyer M J, Moloni K, Kelly T F and Ruoff R S 2000 Strength and Breaking Mechanism of Multiwalled Carbon Nanotubes Under Tensile Load Science 287 637-640 [30] Kim K, Sussman A and Zettl A 2010 Graphene Nanoribbons Obtained by Electrically Unwrapping Carbon Nanotubes ACS Nano 4 1362-6 [31] de Jonge N, Doytcheva M, Allioux M, Kaiser M, Mentink S A M, Teo K B K, Lacerda R G and Milne W I 2005 Cap Closing of Thin Carbon Nanotubes Adv. Mater. 17 451-5 [32] Ge M and Sattler K 1994 Observation of fullerene cones Chem. Phys. Lett. 220 192-6 [33] Chen I-C, Chen L-H, Ye X-R, Daraio C, Jin S, Orme C A, Quist A and Lal R 2006 Extremely sharp carbon nanocone probes for atomic force microscopy imaging Appl. Phys. Lett. 88 153102(1-3) [34] Trzaskowski B, Jalbout A F and Adamowicz L 2007 Functionalization of carbon nanocones by free radicals: A theoretical study Chem. Phys. Lett. 444 314-8 [35] Baylor L R, Merkulov V I, Ellis E D, Guillorn M A, Lowndes D H, Melechko A V , Simpson M L and Whealton J H 2002 Field emission from isolated individual vertically aligned carbon nanocones J. Appl. Phys. 91 4602-6 [36] Müller K, Malik S and Richert C 2010 Sequence-Specifically Addressable Hairpin DNA−Single- Walled Carbon Nanotube Complexes for Nanoconstruction ACS Nano 4 649–56 [37] Akita S and Nakayama Y 2002 Length Adjustment of Carbon Nanotube Probe by Electron Bombardment Jpn. J. Appl. Phys. 41 4887-9 [38] Heeres E C, Katan A J, van Es M H, Beker A F, Hesselberth M, van der Zalm D J and Oosterkamp T H 2010 A compact multipurpose nanomanipulator for use inside a scanning electron microscope Rev. Sci. Instrum. 81, 023704(1-4) [39] Liu Z, Joung S-K, Okazaki T, Suenaga K, Hagiwara Y, Ohsuna T, Kuroda L and Iijima S 2009 Self- Assembled Double Ladder Structure Formed Inside Carbon Nanotubes by Encapsulation of H8Si8O12 ACS Nano 3 1160-6 [40] Liu C and Yang S 2009 Synthesis of Angstrom-Scale Anatase Titania Atomic Wires ACS Nano 3 1025-31 [41] Sabirov A R, Stankevich I V and Chernozatonskii L A 2004 Hybrids of carbyne and fullerene JETP Lett. 79 121-5 [42] Dzegilenko F N, Srivastava D and Saini S 1998 Simulations of carbon nanotube tip assisted mechano-chemical reactions on a diamond surface Nanotechnology 9 325–30 [43] Merkle R C 1997 A New Family of Six Degree Of Freedom Positional Devices Nanotechnology 8 47-52 [44] Drexler K E 1992 Nanosystems: Molecular Machinery, Manufacturing and Computation (New York: Wiley) [45] Globus A, Bauschlicher Jr C W, Han J, Jaffe R L, Levit C and Srivastava D 1998 Machine phase fullerene nanotechnology Nanotechnology 9 192–9 [46] Laikov D N 1997 Fast evaluation of density functional exchange-correlation terms using the expansion of the electron density in auxiliary basis sets Chem. Phys. Lett. 281 151-6 [47] Tarini M, Cignoni P and Montani C 2006 Ambient Occlusion and Edge Cueing for Enhancing Real Time Molecular Visualization IEEE Transactions on Visualization and Computer Graphics 12 1237-1244 

1006.4840

1

1006

Direct ESR evidence for magnetic behaviour of graphene

Recently, there have appeared theoretical works on the magnetic properties of graphene and graphene nanoribbons envisaging possible spin-based applications along with fundamental scientific insight. The theoretical efforts, however, appear not paralleled by experimental investigation to test magnetic properties. Yet, room temperature ferromagnetism (RTFM) has recently been experimentally reported in graphene (G-600) [Nano. Letters 9, 220 (2009)], the origin of which remains still unexplored. Inspired by this observation, and in attempt to trace the origin of RTFM, we here report on low temperature K-band electron spin resonance (ESR) observations on G-600. Two distinct C-related paramagnetic signals are revealed, both of a Lorentzian shape: a) a broad at g = 2.00278 which can be attributed to graphitic-like (GL) carbon; b) a narrower signal at g = 2.00288 which is associated with free radical like (FL) carbon. No other signals could be detected. We speculate that the GL ESR signal may come from the conductive {\pi}-carriers propagating in the interior of graphene sheets, while the FL ESR signal may stem from the edges of graphene sheets due to non-bonding localized electronic states. It is suggested that the long range direct/indirect exchange interaction between GL and FL C-related magnetic spin centers may lead to the reported RTFM, this pointing to C origin of the later.

cond-mat.mes-hall

cond-mat

1 Direct ESR evidence for magnetic behaviour of graphene S. S. Rao* and A. Stesmans Department of Physics and INPAC-Institute for Nanoscale Physics and Chemistry, University of Leuven, Celestijnenlaan 200 D, B-3001 Leuven, Belgium Y. Wang and Y. Chen Institute of Polymer Chemistry, Nankai University, Tianjin 300071, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210093, China Abstract Recently, there have appeared theoretical works on the magnetic properties of graphene and graphene nanoribbons envisaging possible spin-based applications along with fundamental scientific insight. The theoretical efforts, however, appear not paralleled by experimental investigation to test magnetic properties. Yet, room temperature ferromagnetism (RTFM) has recently been experimentally reported in graphene (G-600) [Nano. Letters 9, 220 (2009)], the origin of which remains still unexplored. Inspired by this observation, and in attempt to trace the origin of RTFM, we here report on low temperature K-band electron spin resonance (ESR) observations on G- 600. Two distinct C-related paramagnetic signals are revealed, both of a Lorentzian shape: a) a broad at g = 2.00278 which can be attributed to graphitic-like (GL) carbon; b) a narrower signal at g = 2.00288 which is associated with free radical like (FL) carbon. No other signals could be detected. We speculate that the GL ESR signal may come from the conductive π-carriers propagating in the interior of graphene sheets, while the FL ESR signal may stem from the edges of graphene sheets due to non-bonding localized electronic states. It is suggested that the long range direct/indirect exchange interaction between GL and FL C-related magnetic spin centers may lead to the reported RTFM, this pointing to C origin of the later. *Corresponding author. Fax: + 32 16 32 79 87. E-mail address: [email protected] (S S Rao)  2 1. Introduction Carbon based materials, such as graphene, graphene oxide (GO) and graphene nanoribbons (GNRs) have recently attracted large interest both from scientific and technological point of view. In particular, graphene, a zero gap semi-metal, is characterized as “the thinnest material in our universe” exhibiting a two-dimensional (2D) honeycomb structure of carbon. It has a band structure showing two intersecting bands at two inequivalent k points in the reciprocal space. It exhibits novel electronic properties such as ballistic transport, massless Dirac fermions, Berry’s phase, minimum conductivity and localization suppression [1-3]. It has been reported that graphene shows very high electron mobility (≈ 105 cm2/V.S) at low temperature [4]. It promises a diverse range of applications in microelectronics, high speed optical communication devices, in spintronics, and as room temperature gas sensors [5-8]. On the other hand, magnetism in C-based materials is a rapidly evolving field of research with strong implications for spin based information technology. C-based magnetism without magnetic elements, having only s and p electrons is intriguing, particularly, when extracted from graphene, a material touted as the basis unit for future spintronic devices owing to its long spin diffusion length, small spin-orbit (SO) as well as hyperfine (hf) couplings [9]. Understanding the origin and basic mechanism behind the magnetic behaviour of C-based materials and engineering ferromagnetic (FM) carbon structures has been of prime importance since long. In their theoretical work, Yazyev et al. [7] have concluded that flat-band quasi localized (QL) states induced by point defects are responsible for magnetism of graphene. Lin et al. [10] have argued the importance of electron-electron interaction and carrier concentration in explaining the FM properties of  3 GNRs. Theoretical calculations of Sawada et al. [11] indicated that zig-zag GNRs (ZGNRs) exhibit anti-ferromagnetic (AFM) phase and also argued that magnetism can be induced by carrier doping. Lehtinen et al. [12] have calculated that the magnetic moment due to an adatom defect is 0.5µ B, µ B being the Bohr magneton while a vacancy has a magnetic moment of about 1µ B, with both defects prone to induce spontaneous long- range FM ordering. In fact, a defective graphene is predicted [13] to show room temperature ferromagnetism (RTFM). A FM phase of mixed sp2 and sp3 pure carbons has been predicted [14]. FM has been observed experimentally by Esquinzai et al. [15] on the highly oriented pyrolytic graphite (HOPG) material with a Tc above 300 K when proton irradiated, and unstable (temporary) FM was observed [16] in light-weight carbon nano- foam also, − the works indicating the origin of the magnetism to be carbon related, not correlated with any external magnetic impurity. Recent experiments [17] have shown that proton irradiation doubles the magnetic moment in comparison to that observed after He ion bombardment. Recent experiments [18] employing magnetic force microscopy (MFM) and scanning tunneling microscopy (STM) techniques performed on HOPG sample revealed that 2D networks of point defects exhibiting localized electronic states are the main source of magnetism. As described above, though there have been several experimental studies on HOPG exploring its magnetic nature, not much experimental work [19,20] has been reported to investigate the magnetic nature of graphene, a building block of HOPG. Theoretical predictions suggested that the edges and defects present in graphene enhance the DOS, which further leads to a FM phase. As predicted, the later exhibits a phase  4 transition from the FM to paramagnetic (PM) state, that is of first order [21]. Such predictions would need experimental verification. Wang et al. [22] have recently experimentally reported the RTFM in graphene (G- 600) by using DC superconducting quantum interference device (SQUID) magnetization measurements. Its origin still remains unexplored. Motivated by this observation, in the present work, we first apply electron spin resonance (ESR) in an attempt to identify paramagnetic (spin) point defects potentially responsible for RTFM in G-600. We present ESR results obtained on samples (G-600), in the pristine state and after annealing in Ar at 800oC (G-800) or H2 at 600oC (H-G-600). The study reveals the occurrence of two distinct, intrinsic C-related types of defects both exhibiting a Lorentzian shape. From this observation, together with the absence of any other signals, it is tentatively suggested the RTFM to be of C origin. 2. Experimental Samples studied were nm-scale graphene, denoted as G-600, prepared by the modified Hummer’s method [22]. Separate samples were additionally annealed at 800o C in Ar (1 atm) for ~ 3 h (sample G-800) or (1 atm) at 600o C for ~ 2 h (sample H-G-600). Conventional K-band (~20.6 GHz) first harmonic ESR measurements were performed at low temperature (T ≤ 10 K) [23], with the spectrometer routinely operated under conditions of adiabatic slow passage. The areal spin density was quantified by the double numerical integration of the derivative absorption spectra of the computer simulated signal. Absolute spin densities were obtained making use of a co-mounted calibrated Si:P intensity marker of g(1.7 K = 1.99876). The modulation amplitude (Bm) of the applied magnetic field and the incident microwave power (Pµ ) were restricted to levels not causing (noticeable) signal distortion. No K-band measurements could be  5 performed for T ≥10 K because of drastic deterioration of the quality factor ‘Q’ of the microwave cavity likely due to the strong semiconducting nature of the graphene samples. 3. Experimental Results and Analysis Atomic absorption spectroscopy (AAS) indicated that G-600 does not contain significant amounts (~ 48.6 ppm) of magnetic impurities (Fe, Co, Ni). DC magnetization measurements were performed on G-600 and G-800 using the SQUID technique by sweeping the magnetic field (H) at fixed Ts of which the first results were published previously [22]. Salient features include: The magnetization (M) versus field (H) isotherms on the pristine sample (G-600) obtained at 2 and 300 K clearly show a tendency towards saturation, with a small coerceive field (Hc) of 40 G (at 300 K), which is a signature of FM; the hysteresis (M-H loop) associated with G-600 is feeble, as expected for a soft FM material. The M-H loop of G-600 was found to disappear upon TA at 800oC (G-800) in Ar. Wang et al. [22] have concluded that the observed RTFM originates from the defects present in G-600. But, a most critical point concerns the nature of the defects causing the long-range FM coupling. For this end, to get in sight, we have applied ESR in an attempt to uncover the magnetic species (defects) giving rise to the appearance and disappearance of FM in G-600 and G-800, and address the possible mechanisms leading to FM. In the following sections, we present the K-band ESR results obtained on G-600, G-800 and H-G-600. 3.1. G-600 Figure 1 shows a K-band (~ 20.6 GHz) ESR spectrum observed on G-600 at 1.7 K. Two symmetric, isotropic ESR signals are observed of zero–crossing g values gc = 2.00278 and gc = 2.00288 with corresponding spin densities 1× 1016 spins/g and 7 × 1013  6 spins/g, respectively. These g values fall within the carbon ESR signal range (g = 2.0022 – 2.0035), and may be ascribed to C-related dangling bonds of spin S = ½ each. Despite intense signal averaging over broad field ranges under various extreme and optimized spectrometer parameter settings, no other signals could be observed. The signals could be computer simulated using Lorentzian line shapes with: a) a broad signal of peak-to-peak width ∆BPP ≈ 80-100 G at g = 2.00278, which can be attributed to graphitic-like (GL1) carbon; b) a narrower signal with ∆BPP ~ 4.5 G, g = 2.00288, attributed to free radical like (FL1) carbon. The fact that no other signals could be detected suggest that the RTFM is of C origin. We speculate that the GL1 ESR signal may come from the mobile π- carriers propagating in the interiors of graphene sheets, while the FL1 ESR signal may stem from non-bonding localized states at the edges of graphene sheets. Similar assignment has been made earlier [24] for the ESR signal originating from graphitic like pyrocarbon. To probe deeper, Q-band ESR spectroscopy was attempted, yet without success (even at low temperatures) because of resonator problems. 3.2. G-800 Wang et al. [22] have shown that FM has disappeared in G-800, which in view of the results obtained on G-600, would make ESR analysis of much interest. Accordingly, K- band ESR measurements have been carried out on G-800 down to1.8 K. The obtained spectrum is shown in Fig 2. Measurements at higher temperatures ( ≥ 4.2 K) were drastically hampered due to excessive cavity loading. As illustrated by the ESR spectrum observed at 1.8 K, no ESR signal could be detected originating from G-800, either intrinsic or extrinsic in nature. This result suggests that the disappearance of FM after TA at 800o C, may be linked to elimination of ESR active centers [22].  3.3. H-G-600 7 To further assess the nature of the above revealed defects (GL1 and FL1), in a next step, G-600 was treated in H2 at 600o C, Fig. 3 showing a K-band ESR spectrum (upper curve) measured on H-G-600. Though the observed signals weakened, we keep observing two Lorentzian shaped signals. The estimated spin densities of GL2 and FL2 are 2 × 1014 and 3 × 1012 spins/g, respectively through computer simulations (bottom curve). Yet, both signals, now labeled GL2 and FL2, appear broadened by a factor ≈ 2 and ≈ 3, respectively, compared to the respective signals observed in G-600; (∆BPP(GL2) = 160-170 G, ∆BPP(FL2) = 15 G), with inferred g values gc (GL2) = 2.0028 and gc(FL2) = 2.00296, that is, unchanged within the experimental accuracy (± 0.0003). Noteworthy is that in the case of H-G-600, the loading of the cavity by the sample has been drastically reduced, indicating that the electronic properties of G-600 are much modified upon H2 treatment. 4. Discussion We speculate that the GL ESR signal arises from the conductive π-carriers propagating in the interiors of graphene sheets, while the FL ESR signal may stem from non-bonding localized electronic states at the edges of graphene sheets. We further propose that the long range direct/indirect exchange interaction between GL and FL C- related magnetic spin centers may lead to the observed RTFM. Previously, various forms of hydrogen-adsorbed defects have been suggested as possible origin of localized magnetic moments. The current results, with no obvious presence of hyperfine (hf) structure, can be taken as negative evidence for that suggestion.  8 The findings of Wang and co-authors [22] are well corroborated by our current ESR results in terms of the appearance and disappearance of ESR signals in G-600 and G-800 respectively. The ESR data would disqualify the extrinsic metallic impurities as the possible origin of RTFM. From the ESR results, we also infer that the proposed defect is not a di-vacancy, as no forbidden ESR transition, a characteristic feature of spin S=1 centers, could be traced. The question of magnetic ordering in defective graphene at finite temperatures remains largely unaddressed. Regarding the mechanism for the formation of the FM state in G-600, among others, the defect mediated mechanism appears to be the favored one. The single atom and localized defect mediated mechanism has been addressed in a number of publications. The current results, reporting the presence of two paramagnetic centers, may suggest such interpretation, meriting some further discussion as to the possible origin of FM in G-600: (i) as known, graphene is a bipartite lattice with two interpenetrating sub-lattices. According to Lieb’s theorem [25], a single vacancy (defect) results in the formation of a local magnetic moment with S = 1/2. From the theoretical calculations, it is known that magnetic moments pertaining to the same sublattice couple ferro-magnetically while moments pertaining to different sub-lattices couple anti- ferromagnetically. Clearly, the later case does not apply here. Based on the fact that FM is observed in G-600, one may argue that the magnetic moments present in the same sub- lattice would give rise to such a phenomena. (ii) Next, it may also be proposed that the RKKY-type interaction [26] among the magnetic moments via conduction carriers present in the same sub-lattice can also lead to FM features. (iii) It may also be possible that the interaction between sp2 and sp3 type spin units can give rise to the reported FM  9 [14].(iv) Finally, it has also been argued that the negative Gaussian curvatures in graphene-layers [27] and weak magnetic coupling between the individual layers may lead to the formation of localized magnetic moments [13]. According to a study [28], based on density functional theory, of magnetism in proton-irradiated graphite, it has been concluded that the H-vacancy complex plays a predominant role in the magnetic behaviour. Yet, as mentioned, we failed to find any indication for such hydrogen-vacancy complexes from ESR finger print signatures. Also, should there be unobserved unpaired electron involved on an edge oxygen, an anisotropic g value is expected which would reflect in the observation of powder pattern line shapes, unlike observations. 5. Summery and Conclusions Low temperature electron spin resonance measurements have revealed the presence of two distinct paramagnetic C-related spin centers in graphene prepared by the modified Hummer’s method, which were found to be eliminated after thermal annealing at 800 oC. Among the suggested possible scenarios for the onset of ferromagnetism in graphene-related materials, we believe that the observed RTFM may originate either from the direct/indirect long range exchange interaction between these two spin centers or through an RKKY type interaction between spin centers pertaining to the same C- sublattice. Our ESR experimental results lead us to reject FM impurities as the origin of the observed magnetism. ESR experiments in combination with thermal treatment (Ar, H2) indicate that the possibility for tuning the magnetization in graphene by thermal steps. A dramatic change in conductivity is noticed upon treatment in H2. Further work is in  10 progress to probe the nature of spin dynamics of the revealed defect centers using X-band spectroscopy.  References 11 [1] A. K. Geim and K. S. Novoselov, Nat. Mater. 6 (2007) 183. [2] A. H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim, Rev. Mod. Phys. 81(2009) 109. [3] Y. Zhang, J. W Tan, H. L. Stormer and P. Kim, Nature 438 (2005) 201. [4] K. I. Bolotin, K. J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, P. Kim and H. L. Stormer, Solid State Commun. 146 (2008) 351. [5] P. Avouris, Z. Chen and V.Perebeinos. Nature Nanotech. 2 (2007) 605. [6] T. Mueller, F. Xia and P. Avouris, Nature Photonics 4 (2010) 297. [7] O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. 100 (2008) 047209. [8] F. Schedin, A. K. Geim, S. V. morozov, E. W. Hill, P. Blake, M. I. Katsnelson and K. S. Novoselov, Nature Materials, 6 (2007) 652. [9] N. Tombros, S. Tanabe, A. Veligura, C. Jozsa, M. Popinciuc, H. T. Jonkman and B. J. van Wees, Phys. Rev. Lett. 101 (2008) 046601. [10] H. H. Lin, T.Hikihara, H-T. Jeng, B-L Huang, C-Y Mou and X. Hu, Phys. Rev. B. 79 (2009) 035405. [11] K. Sawada, F. Ishii, M. Saito, S. Okada and T. Kawai, Nano Lett. 9 (2009) 269. [12] P.O. Lehtinen, A. S. Foster, A. Ayuela, A. Krasheninnikov, K. Nordlund and R.M. Nieminen, Phys. Rev. Lett. 91 (2003) 017202. [13] L. Pisani, B. Montanari and N. M. Harrison, New J. Phys. 10 (2008) 033002. [14] D. Arčon , Z. Jagličič , A. Zorko , A. V. Rode , A. G. Christy , N. R. Madsen, E. G. Gamaly and B. L. Davies, Phys. Rev. B 74 (2006) 014438, and references there in. [15] P. Esquinazi, D. Spemann, R. Hohne, A. Setzer, K.-H. Han, T. Butz, Phys.Rev. Lett. 91 (2003) 227201.  12 [16] A. V. Rode, E. G. Gamaly, A. G. Christy, J. G. F. Gerald, S. T. Hyde, R. G. Elliman, B. L-Davies, A. I. Veinger, J. Androulakis and J. Giapintzakis, Phys. Rev. B 70 (2004) 054407 [17] D. Spemann, K.-H. Han, P. Esquinazi, R . Hohne and T. Butz, Nuc. Inst. and Meth. in Phys. Res. B 219 (2004) 886. [18] J. Cervenka, M. I. Katsnelson and C. F. J. Flipse, Nature Physics 5 (2009) 840. [19] H. S. S. Ramakrishna Matte, K. S. Subrahmanyam and C. N. R. Rao, J. Phys. Chem. C 113 (2009) 9982. [20] Luka iri *, Andrzej Sienkiewicz *, Bálint Náfrádi, Marijana Mioni , Arnaud Magrez, László Forró, Phys. Status Solidi B 246 (2009) 2558. [21] E. V. Castro, N. M. R. Peres, T. Stauber and N. A. P. Silva, Phys. Rev. Lett. 100 (2008) 186803 [22] Y. Wang, Y Huang, Y. Song, X. Zhang, Y. Ma, J. Liang and Y. Chen, Nano. Lett. 9 (2009) 220. [23] A. Stesmans. Phys. Rev. B 48 (1993) 2418. [24] C. Buschhaus and E. Dormann, Phys. Rev. B 66 (2002) 195401. [25] E.H. Lieb, Phys. Rev. Lett. 62 (1989)1201. [26] M. A. H.Vozmediano, M.P. L-Sancho, T. Stauber, F. Guinea, Phys. Rev. B 72 (2005)155121. [27] N. Park, M. Yoon, S. Berber, J. Ihm, E. Osawa, and D. Tom´anek, Phys. Rev. Lett. 91 (2003) 237204. [28] P. O. Lehtinen, A. S. Foster, Y. Ma, A. V. Krasheninnikov and R. M. Nieminen, Phys. Rev. Lett. 93 (2004) 187202.  Figure Captions 13 FIG. 1. Derivative-absorption K-band ESR spectrum (~ 50 cumulative scans) observed at 1.7 K on G-600 using Bm = 0.5 G and Pµ ≈ 2.5 nW. The dashed curve (bottom) represents a computer simulation (combined) of the observed signals at gc (GL1) = 2.00278 and gc (FL1) = 2.00288 of the corresponding peak-to-peak line widths ∆BPP (GL1) = 80-100 G and ∆BPP (FL1) = 4.5 G. The narrow signal at gc = 1.99876 stems from a co-mounted Si:P marker sample. Poor signal-to-noise ratio is due to strong loading the Q factor of the cavity. FIG. 2. K-band ESR spectrum measured on G-800 at 1.8 K (Bm = 0.5 G and Pµ = 2.5 nW). There is no indication of a C-originated ESR signal. FIG. 3. K-band ESR spectrum (top curve) observed at 1.7 K on H-G-600 using Bm = 0.5 G and Pµ = 2.5 nW. The small low field signal might arise from noise. The dashed curve (bottom) represents a computer simulation of the observed signal at gc (GL2) = 2.0028 and gc (FL2) = 2.00296.  Figures FIG. 1. 14  FIG. 2. 15 16 FIG. 3. 

1612.03995

1

1612

Magnetic Bloch Skyrmion Transport by Electric Fields in a Composite Bilayer

We investigate a mechanical method to manipulate magnetic Bloch Skyrmions by applying an electric field in a composite chiral-magnetic (CM)/ferroelectric (FE) bilayer. The magnetoelectric coupling at the interface allows the electric field to stimulate magnetic ordering. Therefore it offers the possibility to generate Skyrmions [Phys. Rev. B 94, 014311 (2016)]. Here, we design a movable and localized electric field source to drive skyrmion transport along the bilayer. A traveling velocity of the electric field source must be carefully chosen to show the stability and effciency of this process. The effects of high speed operation will be discussed.

cond-mat.mes-hall

cond-mat

Magnetic Bloch Skyrmion Transport by Electric Fields in a Composite Bilayer Zidong Wang∗ and Malcolm J. Grimson† Department of Physics, University of Auckland, Private Bag 92019, Auckland, New Zealand (Dated: October 9, 2018) We investigate a mechanical method to manipulate magnetic Bloch Skyrmions by applying an electric field in a composite chiral-magnetic (CM)/ferroelectric (FE) bilayer. The magnetoelectric coupling at the interface allows the electric field to stimulate magnetic ordering. Therefore it offers the possibility to generate Skyrmions [Phys. Rev. B 94, 014311 (2016)]. Here, we design a movable and localized electric field source to drive Skyrmion transport along the bilayer. A traveling velocity of the electric field source must be carefully chosen to show the stability and efficiency of this process. The effects of high speed operation will be discussed. Introduction. - Magnetic Bloch Skyrmions have been introduced theoretically [1, 2] and observed experimen- tally [3, 4] in many works. They occurs in materials which the magnetic order breaks the centrosymmetric nanos- tructure due to the existence of asymmetric exchange interaction (well-known as the Dzyaloshinskii-Moriya in- teraction). Magnetic Skyrmions have been detected ex- perimentally from the range of MnSi [3], FeGe [5], MnGe [6], GaV4S8 [7], Cu2OSeO3 [8], Fe-Co-Si alloys [9] and Mn-Fe-Ge alloys [6]. They offer a great potential for ap- plications in spintronic memory devices, due to their self- protection behavior. To use magnetic Skyrmions as in- formation holders, the current high interest is to control their motion. Several non-mechanical methods have been investigated, such as by using electric current dynamics [10], spin-polarized currents [11, 12], magnetic fields [13], temperature gradients [14, 15], and magnons [16]. But a mechanical method, however, has not been explored. In this paper, a mechanical technique to move magnetic Bloch Skyrmions collinearly with a mobile electric field source is investigated. A previous study has discussed magnetic Bloch Skyrmions induced by the electric field in a composite bilayer in Ref. [17]. Here, we pursue a microscopic ap- proach to Skyrmion transport by moving the electric field source in a plane parallel to the bilayer. In FIG. 1, we (i) construct a bilayer lattice model, which contains a chiral-magnetic (CM) layer with classical magnetic spins and a ferroelectric (FE) layer with electric pseudospins, both of layers are glued by a strong magnetoelectric (ME) coupling, (ii) attach a localized electric field source, which can travel longitudinally along the bilayer film, (iii) carry out a spin dynamics method to determine the time- response behaviors of the magnetic spins and the electric pseudospins, respectively. Movie 1 in the Supplemen- tal Material shows an animation of the dynamics [18]. The results show the creation and propagation of mag- netic Bloch Skyrmions, and we discuss the stability and efficiency of the traveling velocity to Skyrmions. This mechanical technique provides a guide for designing and developing a Skyrmion transport channel in future spin- tronic devices. Model and Simulation. - The total Hamiltonian is con- FIG. 1. A schematic illustration of a composite bilayer con- sisting of a CM layer and a FE layer stacked at the interface. A localized electric field source can be moved longitudinally along the bilayer. structed by effective Hamiltonians in the CM and FE layers and the interfacial interaction, as: H = HCM + HFE + HME. Firstly, the Hamiltonian in CM layer HCM is described by a classical Heisenberg model, as HCM = −JCM [Si,j · (Si+1,j + Si,j+1)] −DCM [Si,j × Si+1,j · x + Si,j × Si,j+1 · y] (cid:88) (cid:88) (cid:88) i,j i,j −KCM (Sz i,j)2, (1) i,j i,j, Sy i,j, Sz where Si,j = (Sx i,j) represents the local mag- netic spin is used to characterize the magnetic moment, i.e., (cid:107)Si,j(cid:107) = 1, and which is a normalized vector, i, j ∈ [1, 2, 3, ..., N ] characterizes the location of each spin in the CM layer. The first term shows the symmet- ric exchange interaction between neighbor spins, where J∗ CM = JCM/kBT represents the dimensionless magnetic exchange coupling. The second term shows the asym- metric exchange interaction, and D∗ CM = DCM/kBT represents the dimensionless Dzyaloshinskii-Moriya co- efficient, with x and y are the unit vectors of the x and y axes, respectively. Weak magnetocrystalline anisotropy exists in CM materials [19]. In general, the last term shows a perpendicular magnetic anisotropy, with K∗ CM = KCM/kBT representing the dimensionless uniaxial anisotropy coefficient along the z axis. k,l, P z k,l, P y In the FE layer, we employ a pseudospin model to solve the dynamical behaviors of electric polarization [20]. The local electric moment is replaced by the pseudospin, shown as Pk,l = (P x k,l), which is regarded as a continuous vector, and k, l ∈ [1, 2, 3, ..., N ] character- izes each pseudospins location. The distinction between a pseudospin and a classical spin, is the variable size and no precession of the pseudospin. Since the electric polariza- tion is defined as the dipole moment density in dielectric materials. The dipole moment density p is proportional to the external electric field, as p =
0χeEext [21]. In the pseudospin model, the size of each electric pseudospin is proportional to the magnitude of its effective field, as (cid:107)Eeff k,l = δH/δPk,l(cid:107) [22]. Hence, (cid:107)Pk,l(cid:107) =
0Ξe(cid:107)Eeff k,l(cid:107), where Ξe is the dimensionless pseudo-scalar susceptibil- ity. Consequently, the electric pseudospin has a variable size as does the behavior of electric dipole. The Hamilto- nian thus be described by a transverse Ising model [23], as [P z k,l(P z k+1,l + P z k,l+1)] (cid:88) (cid:88) k,l HFE = −JFE −ΩFE (P x k,l) (cid:88) k,l −
0χeEz ext P z k,l, (2) k,l where J∗ FE = JFE/kBT represents the dimensionless electric exchange coupling along the Ising z direction. Ω∗ FE = ΩFE/kBT represents the dimensionless transverse field along the +x axis, which is a in-plane field and per- pendicular to the Ising z direction. E∗ ext/kBT represents the dimensionless electric field, which been ap- plied in the +z direction,
0 is the electric permittivity of free space, and χe is the dielectric susceptibility. A particular site of locations k, l presents the pseudospins which in the presence of electric field. Remember that applied electric field is mobile, and spatially attached to the bulk of bilayer film to reduce the edge effects (i.e., edge-merons) in simulations [24]. ext =
0χeEz Interfacial effects between CM and FE layers are caused by a ME coupling. The mechanism behind it can be understood by a strain-stress effect. Since in ferro- electrics, a mechanical strain internally generated from the applied electric field due to the reverse piezoelectric effect. It physically exerts on the CM layer, resulting in a magnetization due to the inverse magnetostrictive effect. Consequently, a series of electric-mechanical-mechanical- magnetic effects constitute the converse ME effect, which emphasizes the influence of electric polarization on the magnetization at interface. Nan et al. have given a de- tailed study of this behavior [25]. The analytic expres- sion of ME effect can be linear or nonlinear, particularly with respect to the thermal effect [26]. In this paper, we only account for low-energy excitations between the CM 2 and FE layers and so we restrict ourselves to the linear expression of ME interaction, as HME = −gME (Sz i,jP z k,l), (3) (cid:88) (i,j)(k,l) where g∗ ME = gME/kBT represents the dimensionless strength of ME coupling. This was discussed by Spaldin [27]. The ME coupling strength is, however, currently unknown. ∂Si,j ∂t The dynamics of magnetic spins has been studied by the well-known Landau-Lifshitz-Gilbert equation [28], which numerically solves the rotation of a magnetic spin in response to its torques, = −γCM[Si,j × H eff i,j )], (4) where γCM is the gyromagnetic ratio, and λCM is the phe- nomenological damping term of CM materials. H eff i,j = −δH/δSi,j is the magnetic effective field acting on each spin, which is the functional derivative of the total Hamil- tonian with respect to each magnetic spin. i,j ] − λCM[Si,j × (Si,j × H eff In the FE layer, pseudospins are used to describe the location of electric dipoles. The electric dipole moment is a measure of the separation of positive and negative charges along the Ising z direction. It is scalar. There- fore, only the z component of pseudospin represents the real polarization, and the time evolution of pseudospins is expected to perform a precession free trajectory [20], as ∂Pk,l ∂t = −λFE[Pk,l × (Pk,l × Eeff k,l)], (5) CM = 1, D∗ CM = 1, K∗ FE = 0.1, g∗ ME = 0.4, γ∗ where λFE is the phenomenological damping term in the k,l = −δH/δPk,l is the electric effective FE structure. Eeff field acting on each pseudospin. Movie 2 in the Sup- plemental Material shows animations for comparing the trajectories of magnetic spin and electric pseudospin [18]. Results. - Dimensionless parameters are used for the numerical simulations: J∗ CM = 0.1, J∗ FE = 0.8, Ω∗ CM = 1, and FE = 0.1. A number of N = 30 × 90 magnetic λ∗ CM = λ∗ spins/electric pseudospins in each layer and free bound- ary conditions are used. Landau-Lifshitz-Gilbert equa- tions are solved by a fourth-order Range-Kutta method. A marginal electric field is applied to order the FE and CM domain walls before the dynamics, then we apply the localized electric field with a magnitude E∗ ext = 10, per- pendicular to the surface. Electric pseudospins quickly complete realignment, but the response of magnetic spins has a delay. The generation process of a Skyrmion in the bulk of CM layer is summarized in FIG. 2 (Movie 3 in the Supplemental Material [18]). Subsequently, this field source is moved along the bilayer with a traveling veloc- ity. The velocity is measured as v∗ = ∆N/∆t∗, where ∆N corresponds to spatial movement to equivalent lo- cations (i.e., spin-site), and ∆t∗ is a dimensionless time 3 FIG. 2. Sequential snapshots present the generation of a Skyrmion in the bulk of CM layer, as the localized electric field is statically applied at the initial position. The color scale represents the magnitude of the local z componential magnetization. step. Figure 3 shows a series of diagrams about the Skyrmion transport in the CM layer collinearly following the polarized pseudospins in the FE layer, for a travel- ing velocity of v∗ = 0.02. Movie 3 in the Supplemen- tal Material presents the full dynamical process [18]. In this propagation process, we can see the Skyrmion track deflecting to the bottom edge is due to the Skyrmion Hall effect [29]. The behavior of a Skyrmion is topolog- ically like a spinning disk, it generates a Magnus force when traveling longitudinally. So it induces a transverse force during the translational motion of the Skyrmion. Figure 3 furthermore shows the electric polarization re- flecting the passage of field source. But the magneti- zation has a component that is non-collinear with the electric response, and shows a spin spiral alignment, due to the existence of a finite Dzyaloshinskii-Moriya interac- tion. CM crystals have a non-centrosymmetric structure enables the magnetic ordering to be broken. The movement of the field source is externally control- lable. We therefore explore the effects of higher traveling speed on the Skyrmion transport. Two results of the Skyrmion transport with different velocities are present in FIG. 4 (Movie 4 in Supplemental Material [18]). The first case with v∗ = 0.05 is shown in FIG. 4(a). The Skyrmion barely struggles to follow the motion of FIG. 3. Propagating a Skyrmion by moving the electric field source with a velocity of v∗ = 0.02. The color scale represents the magnitude of the z component. field source during the propagation process. Eventually, the system becomes much more complicated, because an- other two Skyrmions are formed from edge-merons to complement the energy contribution. In FIG. 4(b), we double the velocity to v∗ = 0.1, and note the Skyrmion been lost immediately. Furthermore, the Skyrmion Hall effect acts in the high speed operation, and the transverse motion of Skyrmion transport may result in its annihila- tion at boundaries. Conclusion. - To summarize, we have investigated a novel mechanical method to control magnetic Bloch Skyrmions by moving a electric field source parallel to the composite CM/FE bilayer system. Skyrmions are sup- ported by the electric polarization through the converse ME effect. The results demonstrate that the Skyrmion is moved collinearly with the field source at a slow speed. 4 V. Tsurkan, and A. Loidl, Nat. Mater. 14, 1116 (2015). [8] T. Adams, A. Chacon, M. Wagner, A. Bauer, G. Brandl, B. Pedersen, H. Berger, P. Lemmens, and C. Pfleiderer, Phys. Rev. Lett. 108, 237204 (2012). [9] W. Munzer, A. Neubauer, T. Adams, S. Muhlbauer, C. Franz, F. Jonietz, R. Georgii, P. Boni, B. Pedersen, M. Schmidt, A. Rosch, and C. Pfleiderer, Phys. Rev. B 81, 041203 (2010). [10] S. Woo, K. Litzius, B. Kruger, M.-Y. Im, L. Caretta, K. Richter, M. Mann, A. Krone, R. M. Reeve, M. Weigand, P. Agrawal, I. Lemesh, M.-A. Mawass, P. Fischer, M. Klaui, and G. S. D. Beach, Nat. Mater. 15, 501 (2016). [11] A. Fert, V. Cros, and J. a. Sampaio, Nat. Nanotech. 8, 152 (2013). [12] J. Iwasaki, N. Nagaosa, and Y. Tokura, Nat. Nanotech. 8, 742 (2013). [13] W. Wang, M. Beg, B. Zhang, W. Kuch, and H. Fangohr, Phys. Rev. B 92, 020403 (2015). [14] L. Kong and J. Zang, Phys. Rev. Lett. 111, 067203 (2013). [15] S.-Z. Lin, C. D. Batista, C. Reichhardt, and A. Saxena, Phys. Rev. Lett. 112, 187203 (2014). [16] C. Schutte and M. Garst, Phys. Rev. B 90, 094423 (2014). [17] Z. Wang and M. J. Grimson, Phys. Rev. B 94, 014311 (2016). [18] See Supplemental 10.13140/RG.2.2.26175.51360, dynamical results. Material at for animations about [19] J. S. White, K. Prsa, P. Huang, A. A. Omrani, I. Zivkovi´c, M. Bartkowiak, H. Berger, A. Magrez, J. L. Gavilano, G. Nagy, J. Zang, and H. M. Rønnow, Phys. Rev. Lett. 113, 107203 (2014). [20] Z. Wang and M. J. Grimson, J. Appl. Phys. 118, 124109 (2015). [21] C. Kittel, Introduction to Solid State Physics (Wiley, Hoboken, NJ, 2005) Chap. 16. [22] Z. Wang and M. J. Grimson, J. Appl. Phys. 119, 124105 (2016). [23] P. G. de Gennes, Solid State Commun. 1, 132 (1963). [24] X. Xing, P. W. T. Pong, and Y. Zhou, J. Appl. Phys. 120, 203903 (2016). [25] C.-W. Nan, M. I. Bichurin, S. Dong, D. Viehland, and G. Srinivasan, J. Appl. Phys. 103, 031101 (2008). [26] L. Chotorlishvili, S. R. Etesami, J. Berakdar, R. Khome- riki, and J. Ren, Phys. Rev. B 92, 134424 (2015). [27] F. Thole, M. Fechner, and N. A. Spaldin, Phys. Rev. B 93, 195167 (2016). [28] Y. Li, K. Xu, S. Hu, J. Suter, D. K. Schreiber, P. Ra- muhalli, B. R. Johnson, and J. McCloy, J. Phys. D Appl. Phys. 48, 305001 (2015). [29] W. Jiang, X. Zhang, G. Yu, W. Zhang, M. B. Jungfleisch, J. E. Pearson, O. Heinonen, K. L. Wang, Y. Zhou, A. Hoffmann, and S. G. E. te Velthuis, Nat. Phys. (2016), 10.1038/nphys3883. FIG. 4. High velocities effects in Skyrmion transport for (a) v∗ = 0.05 and (b) v∗ = 0.1. The color scale represents the magnitude of the z component magnetization. But high traveling speeds may break the stability of Skyrmion transport, and annihilate the Skyrmions at edges. Z.W. gratefully acknowledges Wang Yuhua, Zhao Bingjin, Zhao Wenxia and Wang Feng for support. ∗ [email protected] † [email protected] [1] U. Rossler, A. Bogdanov, and C. Pfleiderer, Nature 442, 797 (2006). [2] N. Nagaosa and Y. Tokura, Nat. Nanotech. 8, 899 (2013). [3] S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Boni, Sci- ence 323, 915 (2009). [4] S. Heinze, K. Von Bergmann, M. Menzel, J. Brede, and A. Kubetzka, R. Wiesendanger, G. Bihlmayer, S. Blugel, Nat. Phys. 7, 713 (2011). [5] X. Z. Yu, N. Kanazawa, Y. Onose, K. Kimoto, W. Z. and Y. Tokura, Nat. Zhang, S. Ishiwata, Y. Matsui, Mater. 10, 106 (2011). [6] K. Shibata, X. Yu, T. Hara, D. Morikawa, N. Kanazawa, K. Kimoto, S. Ishiwata, Y. Matsui, and Y. Tokura, Nat. Nanotech. 8, 723 (2013). [7] I. K´ezsm´arki, S. Bordacs, P. Milde, E. Neuber, L. M. Eng, J. S. White, H. M. Rønnow, C. D. Dewhurst, M. Mochizuki, K. Yanai, H. Nakamura, D. Ehlers, 

1611.00614

1

1611

Surface effects on ferromagnetic resonance in magnetic nanocubes

We study the effect of surface anisotropy on the spectrum of spin-wave excitations in a magnetic nanocluster and compute the corresponding absorbed power. For this, we develop a general numerical method based on the (undamped) Landau-Lifshitz equation, either linearized around the equilibrium state leading to an eigenvalue problem or solved using a symplectic technique. For box-shaped clusters, the numerical results are favorably compared to those of the finite-size linear spin-wave theory. Our numerical method allows us to disentangle the contributions of the core and surface spins to the spectral weight and absorbed power. In regard to the recent developments in synthesis and characterization of assemblies of well defined nano-elements, we study the effects of free boundaries and surface anisotropy on the spin-wave spectrum in iron nanocubes and give orders of magnitude of the expected spin-wave resonances. For an 8 nm iron nanocube, we show that the absorbed power spectrum should exhibit a low-energy peak around 10 GHz, typical of the uniform mode, followed by other low-energy features that couple to the uniform mode but with a stronger contribution from the surface. There are also high-frequency exchange-mode peaks around 60 GHz.

cond-mat.mes-hall

cond-mat

Surface effects on ferromagnetic resonance in magnetic nanocubes R. Bastardis1, F. Vernay1, D.-A. Garanin2 and H. Kachkachi1 1Laboratoire PROMES CNRS (UPR-8521), Université de Perpignan Via Domitia, Rambla de la thermodynamique, Tecnosud, F-66100 Perpignan, France 2Physics Department, Lehman College, City University of New York 250 Bedford Park Boulevard West, Bronx, New York 10468-1589, USA E-mail: [email protected] Abstract. We study the effect of surface anisotropy on the spectrum of spin-wave excitations in a magnetic nanocluster and compute the corresponding absorbed power. For this, we develop a general numerical method based on the (undamped) Landau-Lifshitz equation, either linearized around the equilibrium state leading to an eigenvalue problem or solved using a symplectic technique. For box-shaped clusters, the numerical results are favorably compared to those of the finite-size linear spin-wave theory. Our numerical method allows us to disentangle the contributions of the core and surface spins to the spectral weight and absorbed power. In regard to the recent developments in synthesis and characterization of assemblies of well defined nano-elements, we study the effects of free boundaries and surface anisotropy on the spin-wave spectrum in iron nanocubes and give orders of magnitude of the expected spin-wave resonances. For an 8 nm iron nanocube, we show that the absorbed power spectrum should exhibit a low-energy peak around 10 GHz, typical of the uniform mode, followed by other low-energy features that couple to the uniform mode but with a stronger contribution from the surface. There are also high-frequency exchange-mode peaks around 60 GHz. 6 1 0 2 v o N 2 ] l l a h - s e m . t a m - d n o c [ 1 v 4 1 6 0 0 . 1 1 6 1 : v i X r a 2 Surface effects on ferromagnetic resonance in magnetic nanocubes 1. Introduction During the last decades the development of potential technological applications of magnetic nanoparticles, such as magnetic imaging and magnetic hyperthermia, has triggered a new endeavor for a better control of the relevant properties of such systems. In particular, synthesis and growth of crystalline nanoparticles have reached such a high level of skill and know-how as to produce well defined 2D and 3D arrays of nanoclusters of tailored size, shape and internal crystal structure [1–6]. On the other hand, experimental measurements on nanoscale systems are a step behind inasmuch as they still do not provide us with sufficient space-time resolutions for an unambiguous interpretation of the observed phenomena that are commonly attributed to finite-size or surface effects. Nonetheless, ferromagnetic resonance (FMR), which is a well known and very precise technique for characterizing bulk and layered magnetic media [7–9], benefits from a renewed interest in the context of nanomagnetism. Indeed, some newly devised variants of the FMR technique [10–15] combine the study of dynamic magnetic properties by FMR with the elemental specificity of the chemical composition of the particles. For instance, these techniques can be employed to detect the ferromagnetic resonance of single Fe nanocubes with a sensitivity of 106µB and element-specific excitations in Co-Permalloy structures. Another variant of ferromagnetic resonance spectroscopy is the so-called Magnetic Resonance Force Microscopy (MRFM)[16]. It has recently been used for the characterization of cobalt nanospheres [17]. These techniques hold the prospect of providing a better resolution of the surface properties at the level of a single (isolated) magnetic nanoparticle. For the benefits of theoretical work, these experiments could provide the missing data for resolving the surface response to a time- dependent magnetic field, and thus contribute to assess the validity of surface-anisotropy models. In particular, measurements of the absorbed power in FMR experiments on "isolated" particles or dilute assemblies of nanoparticles could serve these purposes. Indeed, this is a standard observable that is routinely measured in such experiments. From the theory standpoint, it is a well known (dynamic) response of a magnetic system that can be computed by various well established techniques, analytical as well as numerical. In the present work we consider a box-shaped nanocluster modeled as a many-spin system with free boundary conditions, subjected to a time-dependent (small-amplitude) magnetic field. The systems considered here are chosen to model, to some extent, Fe nanocubes studied by several groups [3, 5, 6, 10, 18]. Our main objective is to distinguish and assess the role of surface and core contributions to the FMR absorption spectra. For this we focus on the simple system of an isolated (ferromagnetic) nanocube and study its intrinsic properties, thus ignoring its interactions with other nanocubes that would be included in an assembly and its interactions with the hosting matrix. As shown by Sukhov et al. [19], this assumption is fully justified in the case of dilute samples. Obviously, real systems of magnetic nanoparticles are far more complex. Indeed, Fe nanoparticles may present a variety of morphologies and internal structures, especially in a core/shell configuration where one observes an antiferromagnetic layer coating a ferromagnetic system [20–22]. However, the system we adopt is simple enough to illustrate our study in a clear manner but rich enough to capture the main physics we are interested in. Furthermore, the methods we develop here are quite versatile and can be extended to a given magnetic nanoparticle with arbitrary physical parameters. Consequently, the energy of the nanocluster considered here includes the Zeeman energy, the (nearest-neighbor) spin-spin exchange coupling and on-site anisotropy (core and surface). We also allow for the possibility that exchange interactions involving one or more sites in the surface outer shell to be different from those in the core or at the interface between the core and the surface. Upon solving the (undamped) Landau-Lifshitz equation (LLE) we compute the absorbed power of such systems. Then, the LLE equation is linearized around the equilibrium Surface effects on ferromagnetic resonance in magnetic nanocubes 3 state of lowest energy and the ensuing eigenvalue problem is solved to infer the full spectrum (eigenfrequencies and eigenfunctions) of all spin-wave excitations. Finally, by comparison with the absorbed power of a given mode, we can determine the separate contributions of core and surface of the nanocluster. The paper is organized as follows : in the next Section we present our model and computing methods. We give the model Hamiltonian and then describe the two numerical methods we used to compute the full spin-wave spectrum (eigenfrequencies and eigenvectors) and the absorbed power. In Section III we present and discuss our results for the effects of size and surface anisotropy on the absorbed power. This section ends with a discussion of Fe nanocubes for which we give orders of magnitude and speculate on the possibility to observe the calculated peak in the absorbed power. An appendix has been added on a toy model of a three-layer system in order to illustrate, in a simpler manner, how the various branches in the spin-wave dispersion can be associated with spins of a given type (core or surface) in the system. 2. Model and Methods 2.1. The Hamiltonian We model the magnetic nanocluster as a system of N classical spins si, with si = 1, with the help of the Hamiltonian (1) (2) H = Hex + Han, where Hex = − 1 2 (cid:88) i,j Jijsi · sj is the ferromagnetic Heisenberg exchange interaction and Han the anisotropy contribution. We assume only nearest-neighbor interactions (nn), Jij = J > 0 for i, j ∈ nn and zero otherwise. However, we use a numerical method that allows us to consider different exchange couplings between the core and surface spins according to their loci. More precisely, we may distinguish between core-core (Jc), core-surface (Jcs) and surface-surface (Js) exchange coupling. The anisotropy term in Eq. (1) is assumed to be uniaxial (along the z axis) with constant D > 0 for core spins and of Néel's type with constant DS for surface spins. More precisely, the anisotropy energy is local (on-site), so that Han =(cid:80) i Han,i, and given by  −D (si · ez)2 , (cid:88) j∈nn 1 2 DS Han,i = i ∈ core (si · uij)2 , i ∈ surface. (3) 2 DSs2 Here uij is the unit vector connecting the surface site i to its nearest neighbor on site j. Néel's anisotropy arises due to missing nearest neighbors for surface spins. In particular, for the simple cubic lattice and xy surfaces (perpendicular to the z axis), the Néel anisotropy becomes Han,i = − 1 i,z. This means that for DS > 0 the spins tend to align perpendicularly to the surface, while for DS < 0 the surface spins tend to align in the tangent plane. In a box-shaped nanocluster the Néel anisotropy on the edges along the z axis becomes Han,i = − 1 i,x + s2 i,y or, equivalently, Han,i = 1 for DS > 0 the edge spins tend to align perpendicularly to the edges. On the other hand, it is easy to check that Néel's anisotropy vanishes at the corners and in the core of a box-shaped nanocluster. i,z. As such, (cid:0)s2 2 DSs2 For the sake of simplicity, and for an easier comparison with experiments on iron nanocubes, for instance, the systems investigated in the present work are boxed-shaped with 2 DS (cid:1) Surface effects on ferromagnetic resonance in magnetic nanocubes 4 N = Nx × Ny × Nz with a simple cubic lattice. In this case, the surface anisotropy (SA) favours an ordering along the shortest edges of the particle if DS > 0 and along the longest ones otherwise. Indeed, for an atom on the edge in the x direction, for instance, we have 4 neighbors with uij = ex, uij = −ex, uij = ey, uij = −ez and thereby (using si = 1) we obtain Han,i → DS 2 + DS 2 s2 i,x. 2.2. Excitation spectrum and absorbed power: computing methods Since surface anisotropy is much stronger than the core anisotropy and the fraction of surface spins for nanoclusters is appreciable, SA strongly influences the spin-wave spectrum of the cluster. Experimentally, the most accessible modes are the spin-wave modes that couple to the uniform ac field, as in magnetic-resonance experiments. In the absence of SA, only the uniform-precession mode is seen in the magnetic resonance. The effect of SA is twofold. First, the uniform (or nearly uniform) precession frequency is modified by SA; it increases or decreases depending on geometry. Hence, combining magnetic-resonance experiments with the corresponding theoretical results provides a means for estimating the surface-anisotropy constant. Second, in larger clusters exchange stiffness becomes less restrictive and different groups of spins (such as the core and surface spins) can precess at different frequencies and this leads to several resonance peaks. In this Section we describe two complementary methods we have used to compute the spin- wave spectrum and the absorbed power. The first method consists in linearizing the (undamped) Landau-Lifshitz equation (LLE) around the equilibrium position and then solving the ensuing eigenvalue problem to obtain the eigenfrequencies and the corresponding eigenvectors (spin- wave modes). This method is quite versatile as it can be applied to any nanocluster with arbitrary size, shape and energy parameters. In the case of box-shaped nanoclusters this method is compared with the results of linear spin-wave theory obtained in Refs. [23, 24]. The second numerical method used here consists in directly solving the LLE using the technique of symplectic integrators [25, 26]. As will be seen later these two methods are in a very good agreement. 2.2.1. Linearization of the Landau-Lifshitz equation: normal modes of a nanocluster Here we deal with the numerical solution of the Landau-Lifshitz equation (LLE) of motion  dsi dt = si × Heff,i − λ si × (si × Heff,i) , (4) where the effective field is defined by Heff,i = −δsiH + gµBH(t), with g being the Landé factor and µB the Bohr magneton, λ the dimensionless damping parameter and H(t) the time- dependent magnetic field. In the following we set λ = 0 (Larmor equation) to avoid artificial effects. Internal spin-wave processes in the particle can provide a natural damping of spin waves, especially for larger particles and non-zero temperatures. For nanosize particles, spin- wave modes are essentially discrete [23, 24], while damping requires quasi-continuous excitation branches to satisfy energy conservation in spin-wave processes. In addition, we do not include thermal excitation via stochastic Langevin fields in the model. Thus we expect that the spin wave modes of our particles are undamped. In other words, in this work, we are not seeking the precise result for the microwave absorption. We use these calculations to find positions of spin- wave peaks and compare them with a second approach. Our numerical experiment is short-time whereas damping comes into play at longer times that we are not considering here. One of the goals of the present work is to assess the role of the surface contribution to the energy spectrum of a single nanocluster or to a given physical observable that is easily accessible experimentally, e.g. the absorbed power. So, before we compute the relevant observable, it is necessary to compute the eigenvectors and eigenenergies of the system. Then, it is our aim to try to attribute the various peaks in the energy spectrum to the core or surface contributions and (cid:110) (cid:111) Surface effects on ferromagnetic resonance in magnetic nanocubes to estimate the corresponding statistical weight. The eigenvalue problem by linearizing the LLE (4) around the equilibrium state . This has been done in the system of spherical coordinates in order to reduce the number of equations from 3N to 2N . The main steps of our formalism are summarized in Appendix A. More precisely, we write δsi = si − s(0) , for i = 1, . . . ,N , and expand the first derivative of the energy E (or the effective field) to 1st-order in δsi i=1,...,N s(0) i i (cid:8)s(0)(cid:9) + (cid:34) N(cid:80) j=1 (cid:35)(cid:8)s(0)(cid:9) . (δsj·∇j) Heff,i Then, inserting this into the LLE (4) leads to where (cid:101)Hik is the pseudo-Hessian defined in Eq. (A.4) and j=1 dt i = 1, . . . ,N is a matrix that results from the vector product of si with the effective field Heff,i. The solution of Eq. (6) can be sought in the form δsk (t) = δsk (0) exp (iωt), leading to the eigenvalue problem Heff,i d (δsi) = δsj, (cid:104)(cid:101)HijI(cid:105) (cid:19) (cid:8)s(0) + δs(cid:9) = Heff,i N(cid:88) (cid:18) 0 −1 (cid:16)(cid:101)HijI − iω1 δsj = 0, 0 1 I ≡ N(cid:88) j=1 (cid:17) (cid:104)(cid:101)HijI(cid:105)αβ 5 (5) (6) (7) (8) whose solution yields the excitation spectrum of the nanocluster. Accordingly, the eigenvalue problem (7) is then solved numerically for an arbitrary N -spin nanocluster by diagonalizing the 2N × 2N matrix with elements . This is done in the absence of the time-dependent magnetic field H(t) so that the effective field involved here is given by Heff,i = −δsiH. In order to evaluate the contributions of the surface and core spins to the eigenvector (or mode) δsk, we introduce the corresponding "spectral weight". For this purpose, we first write the eigenvector δsk of wave vector k as N(cid:88) i=1 N(cid:88) (cid:88)  1 δsk (0) = fkiδsi (0) (cid:104)(cid:101)HI(cid:105) with fki are the eigenfunctions of the matrix rewritten as . For later use the equation above can be δsk (0) = i=1 α=x,y,z Dα kieα i (9) where {ei,x, ei,y, ei,z} is the local Cartesian frame and Dα ki are the corresponding coefficients. Then, we may define the spectral weight (per site) associated with the core and surface spins as follows (cid:88)  fki2 W s,c k = 1 N × Ns,c i∈core,surface with the normalization condition NsW s (surface) spins. k + NcW c k = 1, where Nc (Ns) is the number of core 6 Surface effects on ferromagnetic resonance in magnetic nanocubes In Ref. [24] the eigenfunctions fki were calculated analytically using the finite-size spin-wave theory for a boxed-shaped particle. This yields a benchmark for the numerical results obtained here and helps interpret them. The spin-wave excitations were treated perturbatively as small deviations of the spins si from the direction n of the particle's net magnetic moment, namely si (cid:39) n + πi, with n · πi = 0. The eigenfunctions were then obtained in the form (fix,kx × fiy,ky × fiz,kz) πi πk = with (cid:88) (cid:114) 2 ix,iy,iz (10) (11) fiα,kα = 1 + δkα cos [(iα − 1/2) kα] , α = x, y, z , in the case of free boundary conditions, as adopted here. Comparing Eq. (10) and kα = nαπ Nα with Eq. (8) we see that the variables πk used in Ref. [24] are in fact identical to the variables δsk defined in Eq. (8). The normal modes of a magnetic nanocluster have been studied by many authors [see Refs. [27, 28] and references therein]. On the other hand, the ferromagnetic resonance of ensembles of magnetic nanoparticles in the macrospin approximation has also been studied numerically using the Landau-Lifshitz equation [19, 29]. In the present work we use similar methods (analytical and numerical) with the main objective here to investigate the effects of surface anisotropy on the resonant absorption by the spin-wave modes in box-shaped nanoclusters. 2.2.2. Solution of the Landau-Lifshitz equation by symplectic methods In these numerical calculations we set J = 1,  = 1. For simplicity, we consider only cases in which the spins in the equilibrium state are collinear and directed along the z axis. This assumes that the surface anisotropy does not exceed a certain critical value. Typically we have D = 0.01 and DS = 0.1. The ac field is applied along the x axis, if not stated otherwise. The results of this method will be compared to those of the previous methods. Among many existing solvers of systems of ordinary differential equations, we employ a method making explicit rotations of spins around their effective fields [see Refs. [25, 26] and many references therein]. This method conserves the spin length and, in the absence of anisotropy, it also conserves the energy. Since anisotropy is much weaker than the exchange interaction, the energy non-conservation is weak. The evolution operator of the system corresponding to the time interval ∆t can be written in the exponential form U = e L∆t, L = Li. (12) i=1 There is no explicit formula for e L∆t since the precession of one spin changes the effective fields on the others. However, the action of the operators e Li∆t describing the rotation of an individual spin around its effective field with all other spins frozen, can be worked out analytically. In the absence of anisotropy this is simply the precession around a fixed field that conserves both spin length and the energy. In the presence of anisotropy the effective anisotropy field changes as the spin is precessing, thus an analytical description of this precession is possible but cumbersome. However, since the anisotropy field is much smaller than the dominating exchange field, one can use the anisotropy field at the beginning of the interval ∆t, making only a small error. Representing the precession of all spins in the system as a succession of individual precessions induces errors growing with ∆t. This error can be reduced by using a generalization N(cid:88) Surface effects on ferromagnetic resonance in magnetic nanocubes of the second-order Suzuki-Trotter decomposition e( A+ B)h = e Ah/2e Bhe Ah/2 + O(cid:0)h3(cid:1) that, in 7 our case, has the form U = e L1,he L2h . . . e LN−1he LN he LN he LN−1h . . . e L2he L1h (13) with h ≡ ∆t/2. That is, all spins are rotated around their respective effective fields in succession in some order. Then the procedure is repeated in the reversed order. The effective field on the next spin is updated because of rotation of the previous spin. In the presence of a time- dependent field, the best choice is to take the values of the latter in the middle of the two series of successive rotations, that is, at ∆t/4 and 3∆t/4. Our implementation of this method in Wolfram Mathematica (compiled) is rather efficient and will be confirmed by agreement between the results obtained by Eqs. (15) and (16) for a not too small time step, typically ∆t = 0.1. We would like to emphasize that the approaches (analytical and numerical) presented above are complementary and render the same results for box-shaped clusters. However, the (numerical) method presented in Section 2.2.1 is quite versatile as it allows us to compute the excitation spectrum of a nanocluster of arbitrary shape and model Hamiltonian. 2.2.3. Definition and computing method of the absorbed power The power absorbed by a spin system in the presence of a uniform ac magnetic field is defined as tf 0 (cid:88) i Pabs (t) = − 1 tf dt (gµB) 1 N (cid:104)si(cid:105) (t) · dHac(t) dt (14) where the integration is performed over time from the initial instant t = 0, at which all spins are in their (initial) equilibrium state, to the final time tf . Here, (cid:104)si(cid:105) (t) ≡ Tr [ρ(t)si] where ρ is the density matrix of the ferromagnet. Then, the response of the spin system to a time-dependent field is defined by the difference δ (cid:104)sα i (cid:105)0 = Tr (ρ0si), ρ0 being the density matrix of the unpurturbed ferromagnet. However, in our calculations tf spans several periods, i.e. tf = nT and as such, we can replace (cid:104)si(cid:105) (t) by δ (cid:104)si(cid:105) (t) since the contribution of the constant term vanishes. Therefore, the absorbed power becomes i (cid:105)0, with (cid:104)sα i (cid:105) (t) ≡ (cid:104)sα i (cid:105) (t) − (cid:104)sα Pabs = − 1 tf tf 0 dt (gµB) 1 N (cid:88) i δ (cid:104)si(cid:105) (t) · Hac (t) . (15) On the other hand, since our model is conservative, the absorbed energy should also be given by the change (per time) of the energy of the system, leading to the equivalent definition Pabs = 1 tfN [H(tf ) − H(0)] . (16) We use both formulae for the absorbed power that serve as a check on the numerical calculations. In order to clarify the expected form of the absorbed power that we will compute numerically for magnetic nanoparticles, let us first consider the simple case of a damped harmonic oscillator driven by an oscillating force, i.e. x + 2Γ x + ω2 (17) where a coupling constant ξ is introduced for generality. Solving this equation with the initial conditions x(0) = x(0) = 0 and calculating the absorbed power for times tf = NT T , T = 2π/ω, NT being the number of cycles, with the help of Eq. (15) [note that Eq. (16) cannot be used in 0x = ξh0 sin (ωt) , 8 Surface effects on ferromagnetic resonance in magnetic nanocubes the damped case], one obtains different results in different measurement time ranges. At short times the result is that for the undamped harmonic oscillator, 1 − cos [(ω − ω0) tf ] Pabs h2 0 = ξ2 tf 2 (18) The width of the corresponding peak decreases with the measurement time as ∆ω ∼ 1/tf , while its height grows linearly with tf , so that its integrated intensity is independent of time. At long times a Lorentzian peak is formed around the (effective) angular frequency ω0 with [(ω − ω0) tf ]2 , Γtf (cid:28) 1. Pabs h2 0 = ξ2 2 Γ (ω − ω0)2 + Γ2 , Γtf (cid:29) 1. (19) The latter formula is what is used in magnetic resonance experiments. However, in numerical calculations on magnetic nanoparticles it is inconvenient to perform a very long integration of the equations of motion trying to measure damping that can be very small or zero. Eq. (18) that requires a relatively short computation (we mainly use NT = 10) is fully sufficient in finding the positions of resonance peaks and their intensities (parametrized by the coupling constant ξ in the oscillator model). In contrast to the harmonic oscillator, SW modes in magnetic particles become non-linear at high excitation thus leading to saturation and distortion of the results. For this reason, in numerical calculations we have to use the amplitude of the ac field H0 as small as possible without loss of precision in Eqs. (15) and (16). In the limit of a strong exchange coupling all spins are collinear and can be considered as a single (macro-) spin with an effective anisotropy stemming from the core and the surface. In this approximation, the contribution of surface anisotropy is of first order in DS and depends on the particle's shape. For the case DS > 0 and oblate particles in the xy plane, the effective SA has an easy axis in the z direction. For prolate particles or for DS < 0 the z direction becomes a hard axis of the effective SA. For particles of cubic (or spherical) shape the first-order contribution of the SA cancels out. However, there is a second-order contribution ∼ D2 S/J that has a form of cubic anisotropy and which favours an orientation of the particle's spin along the (1, 1, 1) direction of the simple cubic lattice. Indeed, this orientation leads to the largest deviations from the collinear state that lower the total energy [30]. Considering the precession of the macrospin (the particle's net magnetic moment) in the effective field, to first order in DS, one obtains the resonance frequency ω0 = 2D NcoreN Nx(Ny − Nz) Ncore DS D 1 + Ny(Nx − Nz) Ncore DS D 1 + (20) (cid:115)(cid:20) (cid:21)(cid:20) (cid:21) Indeed, for the cubic shape, Nx = Ny = Nz and the effect of DS vanishes. For DS > 0 and oblate particles (Nx, Ny > Nz) the resonance frequency increases, while for DS > 0 and prolate particles the precession mode softens. We have not calculated the second-order effect of SA on ω0 but the form of the effective cubic anisotropy to second order in DS suggests that the precession mode will soften for any sign of DS, for the orientation of spins along z axis. Surface effects on ferromagnetic resonance in magnetic nanocubes 9 3. Results and discussion 3.1. Surface and core contributions to the energy spectrum Figure 1. Spectral weight of spin-wave excitations in a box-shaped particle of size 13× 11× 7 and uniform uniaxial anisotropy, in a magnetic field along the x direction. The procedure to determine the weight of surface and core spins in the energy spectrum has been described in Subsection 2.2.1. In order to compare the spectral weights inferred from the analytical expressions in Eq. (10) et seq to those obtained by the numerical method, we consider a box-shaped particle with a simple cubic lattice. In order to avoid spurious effects that could be due to highly symmetric systems we chose to investigate a particle with sides of different lengths, e.g. Nx = 13, Ny = 11, Nz = 7. In Fig. 1 we present a plot of the spectral weight as a function of the energy ω (here  = 1) in units of the nearest-neighbor exchange coupling J, with Jc = Jcs = Js = J. We have considered a static magnetic field along the x axis and a (uniform) uniaxial anisotropy for both the core and surface spins with a common easy axis along the z direction and anisotropy constant D/J = 1. The large core anisotropy D = J is merely introduced in order to shift the whole spectrum by 2J and thereby to highlight the uniform mode. We can see that the numerical results fully agree with the spectral weight inferred from the analytical eigenfunctions in Eq. (10). The full spin-wave spectrum of such many-spin systems is rather complex as it exhibits many branches, and thence does not lend itself to a simple interpretation of the various involved excitations. To that end, we have considered a representative, though much simpler, system that consists of three coupled spin layers for which the excitation spectrum can be computed, with the possibility to disentangle the contributions of the surface and core layers. This is done in Appendix B. The major difference is that the three-layer toy model exhibits only three branches and we can see that the surface spins dominate the low-frequency excitations. On the other hand, the various branches of the many-spin system correspond to different modes running in the k−space of a simple cubic lattice. For instance, 10 Surface effects on ferromagnetic resonance in magnetic nanocubes Figure 2. Absorbed power in a 8× 5× 4 cubic particle. Left panel: Magnetic resonance peak at ω/J = 0.0174 for two different pumping times. The vertical dotted line shows the position of the peak for DS = 0. Right panel: Parametric resonance peak at the double frequency ω/J = 0.0347. a quick inspection of Fig. B2 shows that the surface is dominant away from the Brillouin zone center. In addition, the effect of the surface exchange coupling (Js) has been checked for the same particle without external magnetic field or anisotropy. We have seen that at low excitation energies, the spectral weights of the surface spins are always higher than those of the core spins. However, as Js increases the branches of excitations that preferentially involve surface spins merge with other branches and thus decrease the surface contribution. This effect is more clearly seen in the framework of the toy-model as shown in Fig. B2. 3.2. Absorbed power Box-shaped nanoparticles To see how our numerical method of Section 2.2.2 is 3.2.1. implemented, we start with a small particle containing 160 (= 8 × 5 × 4) spins that is flat in the xy plane, with the anisotropy axis in the z direction and the ac field applied along the x axis (if not stated otherwise). The magnetic-resonance (MR) peak in Fig. 2 (left panel) is seen at ω/J = 0.0174 that is far to the right of the peak position ω/J = 0.0045 obtained for DS = 0. This can be understood as the result of xy planes having a larger area, their stabilizing action for DS > 0 is stronger than the destabilizing action of other surfaces, in a qualitative agreement with Eq. (20). One can see that increasing the pumping time from NT = 10 to NT = 30 makes the resonance peak narrower and higher, in accord with Eq. (18). Moreover, one can see the zeros of Pabs and small satellite maxima between them. All the numerical work presented below uses NT = 10, as this is sufficient to find the positions of the resonance maxima. This is a shape effect indicating that the precession of spins is elliptic rather than circular. In such cases parametric resonance can be observed. Thus for the same particle, we also performed a parametric-resonance calculation, directing the ac field in the spin direction z. The results showing the initial stages of the exponential parametric instability at the double frequency of the MR peak ω/J = 0.0347 are shown in Fig. 2 (right panel). The parametric-resonance peak has a different structure and its growth accelerates with the pumping time. However, the parametric resonance requires a much stronger amplitude of the ac field and longer pumping times, as compared with MR peaks. In the sequel we will only concentrate on the latter. In order to identify the contributions from the core and surface spins in the absorbed power we have investigated a cluster with a similar aspect ratio as the cluster with 13 × 11 × 7 = 1001 Surface effects on ferromagnetic resonance in magnetic nanocubes 11 Figure 3. Absorbed power in a 13 × 11 × 7 cubic particle. Left panel: Low-frequency peak. The vertical dotted line shows the position of the peak for DS = 0. Right panel: Both low-frequency peaks (far left) and high-frequency peaks. spins [see Fig. 3], studied in Fig. 1 and for which the diagonalization method presented in Section 2.2.1 allows for a discrimination between the contributions from the core and surface. Taking the (space) Fourier transform of the spin si(t) in Eq. (15) we obtain the power absorbed by the k = 0 mode Pabs = − 1 tf = − 1 tf tf 0 tf 0 dt (gµB) δsk=0(t) · Hac (t) (cid:88) dt (gµB) α=x,y,z k=0(t)eα · Hac (t) . δsα Then, setting k = 0 in Eq. (9) δsk=0 (t) = δsα k=0 (0) eiωk=0t =  N(cid:88) (cid:88) j=1 α=x,y,z we obtain tf dteiωk=0t  N(cid:88) (cid:88) j=1 α=x,y,z (cid:16) Dα k=0,j Dα k=0,jeα j  eiωk=0t (cid:17) j · gµB Hac (t) eα (21) (22) (23) Pabs = − 1 tf 0 Now, since the vectors eα simplifies into the following form tf − 1 tf j are all parallel to each other, i.e. eα j = eα, the equation above  N(cid:88) (cid:88) Dα k=0,j (cid:16) (cid:17) eα · Hac (t) (24) Pabs = dteiωk=0t (gµB) 0 j=1 α=x,y,z P α abs = Dα k=0,j dteiωk=0teα · gµB Hac (t)  . 12 Surface effects on ferromagnetic resonance in magnetic nanocubes which suggests that we can introduce the power absorbed by the degree of freedom (mode) corresponding to the component α = x, y, z. Indeed, we can write  N(cid:88) j=1  × − 1 k=0 × (cid:101)P α k=0 tf tf 0 This in turn can be rewritten as P α abs (k = 0) = C α where k=0 ≡ C α N(cid:88) j=1 Dα k=0,j is the statistical weight of the k = 0 mode and (cid:101)P α k=0 ≡ − 1 tf tf 0 dteiωk=0teα · gµB Hac (t) . (25) (26) (27) (28) (29) k=0. This means that the absorbed power (per mode) is proportional to the sum of the coefficients of the wave-functions. As such, instead of calculating the absorbed power as defined by Eq. (15) we can calculate and plot the coefficients C α For a clearer analysis of the modes appearing in the absorbed power spectrum, we first focus on the case of a box-shaped sample with the same exchange constant everywhere, namely Jc = Jcs = Js = J, and without any anisotropy. All the spins are then identical and the excitation spectrum is given by a single energy band in the k-space as in Eq. (10). Hence, each mode can be unequivocally labeled by its wave-vector k only. According to the definition of the coefficients C α k , the power can only be absorbed when the field couples to the uniform mode, i.e for C α k=0 = 1. On the other hand, for all other values of the wave-vector k it can be easily shown that (cid:89) sin (Nαkα) sin(cid:0) kα 2 (cid:1) = 0. C α k(cid:54)=0 = α=x,y,z In contrast to this simple case, for a system with different types of local environments, as a consequence of an inhomogeneous exchange coupling (Jc (cid:54)= Jcs (cid:54)= Js), or of different types of on-site anisotropies (surface and core), different energy bands appear in the k-space. This can be easily understood in the framework of the toy model presented in Appendix B and shown in Fig. B2). The analysis of such a situation requires an additional band index ((cid:96)) in order to label each mode of energy ωk,(cid:96) and coefficients C α k,(cid:96). Consequently, the absorbed power can be attributed to the non-uniform modes at k = 0. This is shown in Fig. 4 which presents the spectral weight and the wave-function coefficients C α k,(cid:96) in the low-frequency regime, with a surface anisotropy DS/J = 0.1. The upper panel shows the weights of the core and surface spins for all low frequency modes. The middle and lower panels respectively present the weights of the power-absorbing modes and the coefficient C α k,(cid:96), normalized by that of the uniform mode (k = (cid:96) = 0). We can see that the peaks in the absorbed power in Fig. 3 (for DS/J = 0.1) coincide with the peaks in black in Fig. 4, i.e the peaks obtained for an ac field applied along the x axis. The peaks in black obtained for ω/J = 0.24, 0.54, 0.88 in Fig. 4 are not seen in Fig. 3 because the intensities of these peaks are too low compared to the satellites obtained from the absorbed power, described in subsection 2.2.3. The first peak ω/J = 0.017 (in Fig. Surface effects on ferromagnetic resonance in magnetic nanocubes 13 Figure 4. Spectral weight of spin-wave excitations in a box-shaped particle of size 13× 11× 7. spectral weight for the low frequency region. Middle panel: weights for Upper panel : k=0,(cid:96) (cid:54)= 0. The lower panel correspond to the coefficient Cα k=0,(cid:96) for the different components Cα of the spins defined in Eq. (26), normalized that of the uniform mode k = (cid:96) = 0. ω/J Surface (%) Core (%) 0.017 50 50 0.33 60 40 0.79 70 30 1.18 60 40 Table 1. Contributions to the spectral weight from the surface and core spins for the cluster 13 × 11 × 7 with DS /J = 0.1 and a time-dependent field applied along the x axis (i. e. Black peaks of Fig.4). 3) corresponds to the uniform mode. The latter corresponds to an equal contribution (50%) to the spectral weight from the core and surface spins. Indeed, we have checked that this is in agreement with the lowest energy mode shown in Fig. 4 for which the core and surface spectral weights coincide [see middle panel]. Since the contribution of both core and surface spins is at its maximum in this case, the low-energy peak in Fig. 3 and 4 exhibits the highest intensity. The higher-frequency peaks in black correspond to the non-uniform mode (k = 0, (cid:96) > 0) due to the anisotropy and therefore they occur with a lower intensity. These peaks have a dominant contribution from the surface spins (see Tab. 1). The peaks in cyan in Fig.4 are obtained for a time-dependent field along the y axis. These peaks appear with the same frequencies as the peaks in black but with different intensities. In addition, the contributions from the surface and core spins may vary from one type of peaks to the other. For the same nanocluster and in accordance with Eq. (20), in Fig. 3 the position of the low-frequency peak shifts to the right as DS increases from zero (compare with the vertical line at DS = 0). However, a further increase of SA reverses this tendency, as can be seen from the curve DS/J = 0.2. This mode softening can be attributed to the second-order effect of surface anisotropy. On the other hand, in the high-frequency part of the spectrum one can observe three peaks that could be attributed to three different types of the nanocluster facets with different local environment (or effective fields). Note that the positions of the peaks are nearly the same for DS/J = 0.1 and DS/J = 0.2, which hints at the predominant exchange origin of these modes. 14 Surface effects on ferromagnetic resonance in magnetic nanocubes Figure 5. Absorbed power for 8 × 8 × 8 and 12 × 12 × 12 particles with a focus on the low-frequency peaks in the left column and the high-frequency peaks in the right column. The vertical dotted line shows the position of the peak for DS = 0. 3.2.2. Size effect and application to nanocubes The investigation of size effects in general (i.e. without any rotational symmetry) is a rather involved task since upon increasing the size the number of modes increases and their degeneracy makes it difficult to disentangle their contributions to the spectral weight. This is one of the reasons for which we have decided to focus on cubic samples. In fact, today samples of (iron) nanocubes are routinely investigated in experiments since their synthesis has become fairly well controlled. Accordingly, the results for the absorbed power for the 8 × 8 × 8 particle (512 spins) are shown in Fig. 5. One can see a strong peak at ω = 0.0039J that corresponds to nearly coherent precession of all spins in the particle. Because of the second-order effect of SA [30] this peak is shifted to the left from its position for DS = 0, shown by the vertical dotted line at ω0 = 2DNcore/N = 0.0084J. Note that the first-order formula, Eq. (20), does not capture this effect. Here one cannot use DS/J = 0.2 because further shift of the peak to the left renders the collinear spin configuration along the z axis unstable. The lower panel of Fig. 5 shows similar results for a larger particle of 12 × 12 × 12 = 1728 spins. Here the low-frequency peak is shifted to the right in comparison with the 8×8×8 particle, and which can be explained by the smaller fraction of surface spins. The leftmost and strongest of high-frequency peaks here is larger and shifted to the left. Note that for both of these sizes high-frequency peaks are much smaller than the main low-frequency peak (notice the difference in scale between the left and right panels). By way of illustration, we consider an Fe nanocube of side a = 8 nm [5, 6, 12, 18, 31, 32]. This corresponds to a nanocluster of size 27 × 27 × 27 Surface effects on ferromagnetic resonance in magnetic nanocubes 15 Figure 6. Same as in Fig. 5 but for the cluster 27 × 27 × 27. particle whose absorption spectrum is shown in Fig. 6. Although the present paper is focused on theoretical aspects, a few predictions can be made for realistic iron nanocubes studied today in many experiments. Both synthesis and recent experimental developments have provided systems with optimized structures that could be mimicked by the simplified model studied here. In particular, using some oxygen and plasma treatment it seems that the ligands and oxide shell could be effectively removed, leaving us with ferromagnetic nanocubes, see e.g. Ref. [18]. Would FMR measurements on such nanocubes become possible, the observed spectrum should exhibit the features described in the present work, e.g. a low-energy peak at around 10 GHz for the uniform mode, followed by higher-energy excitations that couple to the latter. In addition, the aspect-ratio of box-shaped (non-cubic) samples can be figured out by this technique upon checking whether a parametric resonance feature appears in the spectrum. In regards with the values of the physical parameters taken in our calculations, we note that the ratio of the magneto-crystalline anisotropy to the exchange coupling (D/J) is here taken at least an order of magnitude larger than in typical iron systems. The reason is that lower values of this ratio require much more time-consuming calculations while the physical picture remains the same. More precisely, the calculation for typical iron materials with D/J ∼ 10−3 would lead to a down-shift of the low-frequency peak roughly by a factor of 10 while the high-frequency peaks should practically remain the same. As compared with the sizes dealt with above, here the high-frequency peak is even larger and even more shifted to the left, so that the low- and high-frequency spectra can be plotted on the same graph. In addition, the high-frequency peak resolves into two peaks. The main high-frequency peak, shown in the right column of Fig. 5, can be interpreted as being due to the precession of the spins located near the facets of the cube. Since this precession mode is non-uniform (it has a non-zero k = 0 component) there is exchange energy involved and this is why the precession frequency is high. With an increasing size, the exchange energy per spin in this mode decreases, and so does its frequency. The splitting of the main high-frequency peak seen for the 27× 27× 27 particle can be explained by the fact that SA induces an increase of the 16 Surface effects on ferromagnetic resonance in magnetic nanocubes mode stiffness at the two xy planes (the small peak on the right) and to a decrease of the mode stiffness at the four other surfaces (the big peak on the left). 4. Conclusion Through a systematic numerical investigation, backed by analytical calculations for special cases, we have studied and distinguished the role of surface and core spins in box-shaped magnetic nanoparticles. We have focused this work on this specific shape inspired by numerous experimental studies of iron nanocubes which are now available in well controlled cubic shapes and sizes. On the other hand, ferromagnetic resonance measurements on "isolated" nanoelements has now become possible with the necessary sensitivity for measuring the absorbed power. Accordingly, we have computed the absorbed power as a function of the excitation frequency and have shown that it is possible to attribute the different contributions of the surface and those of the core spins to the various peaks obtained in our calculations. In particular, the low-energy peak, corresponding to the k = 0 mode, consists of equal contributions from the surface and core spins. Furthermore, in the case of less symmetric box-shaped samples with Néel surface anisotropy, we observe an elliptic precession of the spins whose signature can be seen in a parametric resonance experiment, where a small signal should be detected at twice the frequency of the standard magnetic resonance response. Acknowledgments This work was partly supported by the French ANR/JC MARVEL. The work of DG was also supported by the U.S. National Science Foundation through Grant No. DMR-1161571. Appendix A. Energy Hessian in spherical coordinates At each site i of the cluster's lattice we may define the reference system with the spherical coordinate (θi, ϕi) and basis basis (si, eθi, eϕi) related to the Cartesian coordinates by  sin θi cos ϕi sin θi sin ϕi  , eθi = cos θi  cos θi cos ϕi cos θi sin ϕi − sin θi  , eϕi =  − sin ϕi cos ϕi  .(A.1) 0 si = eϕi From this we derive ∂ϕisi= sin θi eϕi, ∂θisi = eθi, ∂θieθi = − si, ∂ϕieθi = cos θi eϕi, ∂θieϕi = 0, ∂ϕi eϕi = − (sin θi si + cos θi eθi) . leading to δsi = δθi∂θisi+δϕi∂ϕisi=δθieθi+δϕi sin θi eϕi. Then using the gradient ∂si ≡ ∇i = eθi∂θi + eϕi 1 sin θi ∂ϕi , (A.2) we get δsi·∇i = δθi ∂θi + δϕi ∂ϕi. This implies for an arbitrary function f (θiϕi) ∂θif = eθi · ∇if, (A.3) Since the spin deviation δsk can be written in terms of δθk and δϕk Eq. (7) can be written in the basis {(eθi, eϕi)}i=1,···,N = {ξµ}µ=1,···,2N . Note, however, that in the general case these unit vectors are not orthogonal to each other i.e. ξµ · ξν (cid:54)= δµ,ν. In fact, δsk represents the ∂ϕif = sin θi eϕi · ∇if. Surface effects on ferromagnetic resonance in magnetic nanocubes 17 (cid:105) (cid:104) usual spin-wave deviations from the local equilibrium state of spin sk, which is denoted by s(0) k . The latter represents the quantization direction for the local algebra. It's well known that δsk can be written in terms of the spin operators S± k which form a local SU(2) algebra with the i , with εαβγ being the Levi-Civita tensor. usual commutation rules, i. e. In particular, spins operating on different sites commute with each other. This implies that the vectors δsk, or more precisely, the transverse vectors {(eθi, eϕi)}i=1,···,N = {ξµ}µ=1,···,2N can be represented by the vectors of the orthonormal canonical basis {(ei)}i=1,···,N with eα i=1 Ei, is given by a general Hamiltonian we obtain the second derivatives of Ei in terms of its derivative with respect to si. Assuming that the energy E = (cid:80)N = iεαβγδijSγ i = δi,α. i , Sβ Sα j E θiθk  ∂2 k ∇i =(cid:0) ∂θk sin θk 1 (cid:101)Hik(Ei) ≡ (cid:101)Hik = ∇T 1 sin θi ∂2 θkϕi E E ∂2 ϕkθi 1 sin θi sin θk 1 sin θk ∂ϕk  .  . E ∂2 ϕkϕi (cid:1) ∂θi (cid:110) sin θi 1 ∂ϕi (cid:111) (A.4) (A.5) This is the (pseudo-) Hessian of E resulting from the action of the (pseudo-) Hessian operator For a given nanocluster of given size, shape, anisotropy model and the applied field, one , where θi and ϕi first determines the equilibrium state, denoted by are the standard spherical angles defined with respect to the local basis (si, eθi, eϕi) at site i. The effective field is defined by Heff,i = −δsiE = −∇iE, such that the four second derivatives s(0) i = (θ(0) i=1,···,N , ϕ(0) ) i i read ∂2 θiθk E = δik [si · −eθi · (eθi · ∇i)] Heff,i − (1 − δik) eθi · [eθk · ∇k] Heff,i, ∂2 ϕkϕi E = δik sin θi [(sin θi si + cos θi eθi ) − sin θi eϕi · (eϕi · ∇i)] Heff,i − (1 − δik) sin θi sin θk eϕi · [eϕk · ∇k] Heff,i, ∂2 θkϕi E = − δik [cos θi eϕi · + sin θi eϕi · (eθi · ∇i)] Heff,i − (1 − δik) sin θi eϕi · [eθk · ∇k] Heff,i, ∂2 ϕkθi (A.6) It is understood that all these derivatives and the pseudo-Hessian have to be evaluated at − (1 − δik) sin θk eθi · (eϕk · ∇k) Heff,i. E = − δik [cos θi eϕi · + sin θi eθi · (eϕi · ∇i)] Heff,i (cid:110) (cid:111) the equilibrium state s(0) i = (θ(0) i , ϕ(0) ) i . i=1,···,N Appendix B. Toy model In order to achieve a simple physical picture of the contributions of core and surface spins to the spectral weight, together with a possible comparison with the numerical method developed in Subsection 2.2.1, we have built a toy model that captures the main feature we want to illustrate but which is analytically tractable. Accordingly, we consider a ferromagnet composed of 3 coupled layers as sketched in Fig. B1. Each layer is assumed to be infinite in x and y directions. 18 Surface effects on ferromagnetic resonance in magnetic nanocubes Figure B2. Surface and core spectral weights against the magnon energy for jc = jcs = 1 and with js = 0 (A), js = 0.5 (B). Figure B1. surface) and Jc (core). 2D Slab of 3 atomic layers with exchange couplings Js (surface), Jcs (core- H = −(cid:88) (cid:88) The spin Hamiltonian of such a system is the Heisenberg model Jl Si,l · (Si+x,l + Si+y,l) − Jcs Si,2 · (Si,1 + Si,3) , (B.1) where Si,l is the spin at site i within the layer l, and Jl=1,3 ≡ Js and J2 ≡ Jc. We restrict ourselves to the case of a ferromagnet with Jl > 0, Jcs > 0. In the spin-wave approach we choose z as the quantization axis and perform a Holstein-Primakoff transformation l=1,3 i i (cid:88) i,l = S− Sz † i,lS − a i,lai,l, (B.2) Then, we rewrite the Hamiltonian (B.1) in terms of the real-space magnon operators ai,l † and a i,l. The resulting expression can be partially diagonalized after a Fourier transformation with respect to the (x, y) directions 2Sai,l, i,l (cid:39) S− † 2Sa i,l. √ i,l (cid:39) S+ √  + Cte, H S = † q,1 a a † † q,2 a q,3 (cid:16) (cid:88)  J11 qx,qy −Jcs 0  aq,1 (cid:17)J (q) ·  , aq,2 aq,3 0 −Jcs J33 −Jcs J22 −Jcs where J (q) is the coupling matrix J (q) = (B.3) (B.4) Surface effects on ferromagnetic resonance in magnetic nanocubes with J11 = J33 = 2Js (1 − γq) + Jcs and J22 = 2Jc (1 − γq) + 2Jcs, and γq ≡ 1 2 (cos qx + cos qy). We use Jc as our energy scale and define the reduced couplings jcs ≡ Jcs/Jc and js ≡ Js/Jc. The three dispersions are then given by ω± (jcs, js, q) = 3 19 2 jcs + (1 + js) (1 − γq) (cid:110) ± 1 2 [3jcs + 2 (1 + js) (1 − γq)]2 −8 (1 − γq) [jcs (1 + js) + 2js (1 − γq)]}1/2 (B.5) ω0 (jcs, js, q) = jcs + 2js (1 − γq) . The spectral weights are then obtained as the squares of the projections of the eigenvectors i , i = 1, 2, 3; α = x, y, z. These weights depend on the physical onto the canonical basis eα parameters such as the exchange couplings and anisotropy constants. Upon summing over the wave vectors q within the first Brillouin zone, one can plot the spectral weights as functions of ω (q). i = δα In Fig. B2 we present the spectral weight of the surface and core spins as a function of the magnon energy for the three energy bands, along the path qx = qy, corresponding to the three dispersions (B.5). The circles and squares are the results for a finite cluster (Nx = Ny = 23) dealt with using the numerical method of Subsection 2.2.1, with periodic boundary conditions in the x and y directions. The full lines are the results obtained within the spin-wave approach presented above. The results in Fig. B2 exhibit a very good agreement between the numerical and analytical approaches for all values of the exchange parameters. In the spin-wave calculation we consider blocks of three spins, belonging to layers 1, 2, 3. These blocks are coupled to one another by lateral (in-plane) couplings. The spin-wave dispersion, as shown in the inset of Fig. B2, has three branches: the lowest branch corresponds to the ferromagnetic magnon excitations with the 3 spins precessing in phase. By computing the spectral weight associated with this branch, one finds that the surface contribution dominates (apart from the uniform mode at k = 0) because the corresponding modes require less energy to be excited. In contrast, the high-energy branch corresponds to the situation where the end spins (layers) precess with opposite phases. The spectral weight is then dominated by the core owing to a higher spin stiffness. For the particular case of js = 0, the magnon dispersion exhibits a non-dispersive branch at ωk = 1 [see inset of Fig. B2 (left)]. This intermediate branch follows from the fact that the bottom and top layer spins are not coupled within their respective planes. Therefore, creating an excitation within the top or bottom layer is costless, leading to a mode with constant energy in k-space. Obviously, this branch corresponds to excitations that are localized at the surface. This can be seen by examining the spectral weight for which the core contribution vanishes. As the surface exchange coupling increases (i.e. js > 0) more dispersion is observed and the branches start to merge for some magnon energies. Hence, the spectral weight changes both qualitatively and quantitatively: the gaps close and the surface and core contributions become more and more entangled. The calculation of the absorbed power for this system yields one absorption peak for the uniform mode corresponding to the lower energy band in Fig.B2. The eigenfunctions for the three energy bands at k = 0 are given by (φS1 + φC + φS2) , (φS1 − φS2) , (φS1 − 2φC + φS2) . (B.6)  Ψ1 = 1√ Ψ2 = 1√ Ψ3 = 1√ 3 2 6 20 REFERENCES Here φS1,2 corresponds to the surface spins and φC to the core spin. The coefficients Ck,(cid:96) of these vectors do not vanish (and are all equal) for the vector Ψ1 that corresponds to the uniform mode. In order to obtain more absorption peaks in the absorbed power we can introduce a core anisotropy kc but no surface anisotropy. In this case the eigenfunctions corresponding to k = 0 are  Ψ1 = 2 N1 Ψ2 = 1√ 2 Ψ3 = 2 N3 (cid:104) (cid:104) φS1 −(cid:16) φS1 −(cid:16) (φS1 − φS2) , 1 + kc −(cid:112)9 + 2kc + k2 1 + kc +(cid:112)9 + 2kc + k2 c c (cid:17) (cid:17) (cid:105) (cid:105) φC + φS2 φC + φS2 , , (B.7) where N1 and N3 are normalization factors of the wave-vectors Ψ1 and Ψ3 respectively. We can see that the modes corresponding to Ψ1 and Ψ3 can contribute to the absorbed power. References [1] Sun S, Murray C B, Weller D, Folks L and Moser A 2000 Science 287 1989–1992 (Preprint http://www.sciencemag.org/content/287/5460/1989.full.pdf) URL http: //www.sciencemag.org/content/287/5460/1989.abstract [2] Lisiecki I, Albouy P A and Pileni M P 2003 Advanced Materials 15 712–716 ISSN 1521-4095 URL http://dx.doi.org/10.1002/adma.200304417 [3] Tartaj P, del Puerto Morales M, Veintemillas-Verdaguer S, González-Carreño T and Serna C J 2003 Journal of Physics D Applied Physics 36 182 URL http://stacks.iop.org/ 0022-3727/36/i=13/a=202 [4] Lisiecki I and Nakamae S 2014 Journal of Physics Conference Series 521 012007 URL http://stacks.iop.org/0022-3727/41/i=13/a=134011 [5] Snoeck E, Gatel C, Lacroix L M, Blon T, Lachaize S, Carrey J, Respaud M and Chaudret B 2008 Nano Letters 8 4293–4298 URL http://dx.doi.org/10.1021/nl801998x [6] Mehdaoui B, Meffre A, Lacroix L M, Carrey J, Lachaize S, Gougeon M, Respaud M and Chaudret B 2010 Journal of Magnetism and Magnetic Materials 322 L49– L52 (Preprint 0907.4063) URL http://www.sciencedirect.com/science/article/pii/ S0304885310003057 [7] Vonsovskii S V 1966 Ferromagnetic Resonance: The Phenomenon of Resonant Absorption of a High-Frequency Magnetic Field in Ferromagnetic Substances (Pergamon Press, Oxford) [8] AG Gurevich and GA Melkov 1996 Magnetization oscillations and waves (Florida: CSC Press) [9] B Heinrich 1994 Ferromagnetic resonance in ultrathin film structures Ultrathin magnetic structures II ed Heinrich B and Bland J (Berlin: Springer-Verlag) p 195 [10] Tran M 2006 Structural and Magnetic properties of colloidal Fe-Pt and Fe cubic nanoparticles Master's thesis Institut National des Sciences Appliquees de Toulouse Toulouse [11] Lee I, Obukhov Y, Hauser A J, Yang F Y, Pelekhov D V and Hammel P C 2011 Journal of Applied Physics 109 07D313 URL http://scitation.aip.org/content/aip/journal/ jap/109/7/10.1063/1.3536821 [12] Kronast F, Friedenberger N, Ollefs K, Gliga S, Tati-Bismaths L, Thies R, Ney A, Weber R, Hassel C, Römer F M, Trunova A V, Wirtz C, Hertel R, Dürr H A and Farle M 2011 Nano Letters 11 1710–1715 pMID: 21391653 (Preprint http://dx.doi.org/10.1021/ nl200242c) URL http://dx.doi.org/10.1021/nl200242c REFERENCES 21 [13] Gonçalves A M, Barsukov I, Chen Y J, Yang L, Katine J A and Krivorotov I N 2013 Applied Physics Letters 103 172406 (Preprint 1310.7996) URL http://scitation.aip. org/content/aip/journal/apl/103/17/10.1063/1.4826927 [14] Schoeppner C, Wagner K, Stienen S, Meckenstock R, Farle M, Narkowicz R, Suter D and Lindner J 2014 Journal of Applied Physics 116 033913 URL http://scitation.aip.org/ content/aip/journal/jap/116/3/10.1063/1.4890515 [15] Ollefs K, Meckenstock R, Spoddig D, Römer F M, Hassel C, Schöppner C, Ney V, Farle M and Ney A 2015 Journal of Applied Physics 117 223906 URL http://scitation.aip. org/content/aip/journal/jap/117/22/10.1063/1.4922248 [16] Sidles J A, Garbini J L, Bruland K J, Rugar D, Züger O, Hoen S and Yannoni C S 1995 Rev. Mod. Phys. 67(1) 249–265 URL http://link.aps.org/doi/10.1103/RevModPhys.67.249 [17] Lavenant H, Naletov V V, Klein O, De Loubens G, Laura C and De Teresa J M 2014 Nanofabrication 1(1) 2299–680X (Preprint 1404.0492) [18] Trunova A V, Meckenstock R, Barsukov I, Hassel C, Margeat O, Spasova M, Lindner J and Farle M 2008 Journal of Applied Physics 104 093904 URL http://scitation.aip.org/ content/aip/journal/jap/104/9/10.1063/1.3005985 [19] Sukhova A, Usadel K D and Nowak U 2008 Journal of Magnetism and Magnetic Materials 320 31–35 URL http://www.sciencedirect.com/science/article/pii/ S0304885307006580 [20] Briático J, Maurice J L, Carrey J, Imhoff D, Petroff F and Vaurès A 1999 The European Physical Journal D - Atomic, Molecular, Optical and Plasma Physics 9 517–521 ISSN 1434- 6079 URL http://dx.doi.org/10.1007/s100530050491 [21] Ling T, Xie L, Zhu J, Yu H, Ye H, Yu R, Cheng Z, Liu L, Liu L, Yang G, Cheng Z, Wang Y and Ma X 2009 Nano Letters 9 1572–1576 (Preprint http://dx.doi.org/10. 1021/nl8037294) URL http://dx.doi.org/10.1021/nl8037294 [22] Lacroix L M, Huls N F, Ho D, Sun X, Cheng K and Sun S 2011 Nano Letters 11 1641– 1645 pMID: 21417366 (Preprint http://dx.doi.org/10.1021/nl200110t) URL http: //dx.doi.org/10.1021/nl200110t [23] Kachkachi H and Garanin D A 2001 Physica A Statistical Mechanics and its Applications 300 487–504 (Preprint cond-mat/0001278) [24] Kachkachi H and Garanin D A 2001 European Physical Journal B 22 291–300 (Preprint cond-mat/0012017) URL http://dx.doi.org/10.1007/s100510170106 [25] Krech M, Bunker A and Landau D P 1998 Computer Physics Communications 111 1–13 (Preprint cond-mat/9805214) [26] Steinigeweg R and Schmidt H J 2006 Computer Physics Communications 174 853–861 (Preprint cond-mat/0507262) [27] Grimsditch M, Leaf G K, Kaper H G, Karpeev D A and Camley R E 2004 Phys. Rev. B 69(17) 174428 URL http://link.aps.org/doi/10.1103/PhysRevB.69.174428 [28] Grimsditch M, Giovannini L, Montoncello F, Nizzoli F, Leaf G K and Kaper H G 2004 Phys. Rev. B 70(5) 054409 URL http://link.aps.org/doi/10.1103/PhysRevB.70.054409 [29] Usadel K D 2006 Phys. Rev. B 73(21) 212405 URL http://link.aps.org/doi/10.1103/ PhysRevB.73.212405 [30] Garanin D A and Kachkachi H 2003 Phys. Rev. Lett. 90(6) 065504 URL http://link.aps. org/doi/10.1103/PhysRevLett.90.065504 22 REFERENCES [31] Jiang F, Wang C, Fu Y and Liu R 2010 Journal of Alloys and Compounds 503 L31 – L33 ISSN 0925-8388 [32] O'Kelly C, Jung S J, Bell A P and Boland J J 2012 Nanotechnology 23 435604 URL http://stacks.iop.org/0957-4484/23/i=43/a=435604 

1311.0964

1

1311

Donor and Acceptor Levels in Semiconducting Transition Metal Dichalcogenides

Density functional theory calculations are used to show that it is possible to dope semiconducting transition metal dichalcogenides (TMD) such as MoS$_2$ and WS$_2$ with electrons and/or holes either by chemical substitution or by adsorption on the sulfur layer. Notably, the activation energies of Lithium and Phosphorus, a shallow donor and a shallow acceptor, respectively, are smaller than 0.1 eV. Substitutional halogens are also proposed as alternative donors adequate for different temperature regimes. All dopants proposed result in very little lattice relaxation and, hence, are expected to lead to minor scattering of the charge carriers. Doped MoS$_2$ and WS$_2$ monolayers are extrinsic in a much wider temperature range than 3D semiconductors, making them superior for high temperature electronic and optoelectronic applications.

cond-mat.mes-hall

cond-mat

Donor and Acceptor Levels in Semiconducting Transition Metal Dichalcogenides A. Carvalho1, A. H. Castro Neto1 1Graphene Research Center, National University of Singapore, 6 Science Drive 2, Singapore 117546∗ (Dated: November 6, 2013) Density functional theory calculations are used to show that it is possible to dope semiconduct- ing transition metal dichalcogenides (TMD) such as MoS2 and WS2 with electrons and/or holes either by chemical substitution or by adsorption on the sulfur layer. Notably, the activation ener- gies of Lithium and Phosphorus, a shallow donor and a shallow acceptor, respectively, are smaller than 0.1 eV. Substitutional halogens are also proposed as alternative donors adequate for different temperature regimes. All dopants proposed result in very little lattice relaxation and, hence, are expected to lead to minor scattering of the charge carriers. Doped MoS2 and WS2 monolayers are extrinsic in a much wider temperature range than 3D semiconductors, making them superior for high temperature electronic and optoelectronic applications. Advances in the fabrication and characterization of two-dimensional (2D) dichalcogenide semiconductors have reshaped the concept of thin transistor gate.[1, 2] Unlike thin fully-depleted silicon channels, physically limited by the oxide interface, single layer metal dichalco- genides are intrinsically 2D and, therefore, have no sur- face dangling bonds. The monolayer thickness is con- stant, and the scale of the variations of the electrostatic potential profile perpendicular to the plane is only lim- ited by the extent of the electronic wavefunctions. Hence, TMD can in principle be considered immune to channel thickness modulation close to the drain. Building on these fundamental advantages, numerous field-effect transistor (FET) designs employing MoS2 or WS2 channels have been proposed. These range from 2D adaptations of the traditional FET structure, where the 2D semiconductor is separated by a dielectric layer from a top gate electrode, to dual-gate heterolayer devices where the transition metal dichalcogenide is straddled between two graphene sheets[2]. Such FETs can be integrated into logic inversion circuits, providing the building blocks for all logical operations [3]. However, at present the success of TMD in electron- ics is limited by the difficulty in achieving high carrier concentrations and, by consequence, high electronic mo- bilities (current values range around 100 cm2/V.s)[4]. In the absence of a chemical doping technology, the con- trol of the carrier concentration relies solely on the ap- plication of a gate voltage perpendicular to the layer, which shifts the Fermi level position rendering the ma- terial n- or p-type[5]. But in practice the gate voltage drop across the insulator cannot exceed its electric break- down limit (about 1 V/nm for SiO2 , or lower for high-κ dielectrics[6]). A work-around demonstrated in graphene consists on gating with ferroelectric polymers[7], al- though at the expense of the thermal stability and switch- ing time. In this article we use first-principles calculations to show that MoS2 and WS2 can be doped both n- and p-type using substitutional impurities. This grants tran- sitional metal dichalcogenides an advantage over other chalcogenide semiconductor families where doping asym- metries are notorious: ZnS can be doped n-type but not p-type, while chalcopyrite CuInTe2 and CuGaSe2 can be p-type doped but not n-type doped[8], and SnTe has not yet been doped n-type[9]. In transition metal dichalco- genides, even though chemical doping is mostly unex- plored, there have already been some experimental re- ports of successful chemical doping[10, 11],as well as some electronic structure calculations for impurities[12, 13]. Further, we find both n- and p-type dopants substitut- ing in the S lattice site or adsorbed on top of the S layer. Leaving the transition metal layer nearly undisturbed, these substitutions promise less scattering to charge car- riers at the Mo-derived states at the bottom of the con- duction band (CBM) or at the top of the valence band (VBM). Having established that doping is possible, it follows that 2D doped semiconductors stand out as superior to 3D semiconductors for high temperature applications be- cause of fact that the electronic density of states, N (E ), close to the edge of the valence and conduction bands is, unlike the 3D case, energy independent. It is well-known that the intrinsic carrier concentration of a semiconduc- tor is given by: ni (T ) = pNc (T )Nv (T ) exp(−Eg /(2kT )), where Eg is the gap energy, and Nc(v) depend on N (E ) (and hence the dimensionality) of the semicondutor. In 2D we have: (1) Nc(v) = Mc(v)me(h) ln 2 π¯h2 kT , (2) Mc(v) is the degeneracy of the conduction (valence) band, me(h) is the effective mass of the conduction (valence) band electrons, and T the temperature (k and ¯h are the Boltzmann and Planck’s constants, respectively). Hence, in 2D we have ni,2D (T ) ∝ T which should be contrasted the 3D counterpart where ni,3D ∝ T 3/2 . Figure 1 illus- trates the relevance of the temperature dependence of the density of conduction electrons n(T ), by comparing the carrier density for n-type monolayer MoS2 and Si, 3 1 0 2 v o N 5 ] l l a h - s e m . t a m - d n o c [ 1 v 4 6 9 0 . 1 1 3 1 : v i X r a doped with the same dopant concentration and dopant activation energy, as a function of temperature. While Si leaves the extrinsic regime (that is, the region of temper- atures where ni (T ) becomes temperature independent) above 800 K, in MoS2 the n(T ) curve is flat beyond 1000 K. The temperature stability of ni ultimately re- flects on transistor characteristics, in particular the gate voltage threshold. We studied donor and acceptor impurities using first- principles calculations. These were based on density functional theory (DFT), as implemented in the Quan- tum ESPRESSO code.[14]. Geometry optimizations and total energy calculations are non-relativistic. A fully relativistic formalism was used for the bandstruc- ture calculations (see Supplementary Information). The exchange correlation energy was described by the gener- alized gradient approximation (GGA), in the scheme pro- posed by Perdew-Burke-Ernzerhof[15] (PBE). The Kohn- Sham bandgaps obtained in the non-relativistic calcu- lations are respectively 1.65 and 1.77 eV for MoS2 and WS2 . With spin orbit coupling, these values become 1.55 and 1.51 eV, respectively. We thus find that the GGA is a good approach for bandstructure calculations of these materials, and further exchange and correlation effects are likely to produce, in first approximation, only a rigid shift of the conduction band[16]. The energy cutoff used was 50 Ry. Further details of the calculation method can be found in Ref. 17 The supercell consisted of 4×4 unit cells of the sin- gle layer material, separated by a vacuum spacing with the thickness of two times the supercell lattice param- eter. For charged supercells, the electrostatic correc- tion of Komsa and Pasquarello was implemented.[18, 19] The Brillouin-zone (BZ) was sampled using a 4×4×1 Monkhorst-Pack grid.[20] We have considered five dopants: Si, P, Li, Br and Cl. Any of these can occupy substitutional positions or be adsorbed on the S layer. The point symmetry of the S site is C3v . When replaced by P or Si, the resulting de- fect keeps the trigonal symmetry and there is little associ- ated lattice distortion. In the case of neutral ClS and BrS however, the lowest energy configuration is a Cs geometry where the neutral ClS and BrS defects are displaced in the vertical plane, loosening one of the Cl/Br-Mo/W bonds (Fig. 2-a). This unusual configuration results from the fact that the halogen partially donates the unpaired elec- tron to the Mo/W d orbitals, whereas in most molecules Cl and Br receive an electron instead. Li is most stable at the S3 position[13], shown in Fig. 2- b, outside the S layer but on the top of a Mo atom. As for the adsorbed atoms, P, Si, Cl and Br take the S4 configuration as described in Ref. 13, on top of an S atom. A requirement for successful doping is that the im- purity must be stable at the lattice position where it is active, and comparatively unstable or electrically neutral at the competing positions. The equilibrium concentra- 2 Si extrinsic region M D N / n 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0 200 400 600 800 1000 T (K) FIG. 1. Electron density in n-type monolayer MoS2 and Si, with concentration ND = 1018 cm−3 of donors with ionization energy Ec−EI = 0.045 eV. An effective thickness of 6.46 Awas used for MoS2 . FIG. 2. Top: geometry of a distorted substitutional defect (BrS ), in top and side view. Bottom: geometry of Li ad- sorbed at the S3 position. TM and S atoms are represented as gray and white spheres, respectively. The broken bond is represented in dashed line. (3) (4) niµi + qµe , tion [D ] of a defect form D can be related to the defect formation energy ∆GD , kT (cid:19) , [D ] = gND exp (cid:18)− ∆GD where ND is the number of sites available to the defect. Since our calculations are for T = 0 entropy terms can be neglected, and the formation energy of the defect can be obtained from the total energies, ∆GD ≃ Ef (D) = E (D) − Xi where Ef (D) is the free energy of the system containing the defect, ni is the number of atoms of species i that it contains and µi is the respective chemical potential. Additionally, the formation energy of a charged system, in charge state q , depends on the chemical potential of the electrons (µe ). The chemical potentials are defined by the experimen- tal growth conditions, which can range from metal-rich to sulfur-rich. Bulk MoS2 and WS2 are often sulfur deficient,[21, 22] even though sulfur excess has been re- ported as well[23] Here, we will assume {µMo/W ,µS } are in the metal-rich extreme ie. the system is in equilibrium with a hypothetical reservoir of metallic Mo (or W). The chemical potentials for the impurities are taken to be the total energy of the respective isolated atoms, so that Ef (Dad ) is by definition the adsorption energy. The calculated formation energies are given in Table I. In the sulfur-poor limit, for both host materials, substitu- tional Si and P bind strongly to the lattice, and are more stable in the substitutional position. Li, on the contrary, is most stable at a surface adsorbed position. Br and Cl have comparable formation energies in both forms. The energy difference between adsorption and substitution at the S site is linear on the chemical poten- tial of sulfur, and independent on the chemical potential of the impurity itself: Ef (DS ) − Ef (Dad ) = E (DS ) − E (Dad ) + µS . (5) Thus, it is in principle possible to control the relative populations of Cl or Br in different sites by changing the sulfur abundance. Another way to enhance the incorporation ratio of Br and Cl at S sites by using material that has sulfur vacan- cies a priori (for example pre-irradiated material). The capture of an impurity atom adsorbed at the layer surface by a sulfur vacancy, VS + Xad → XS , (6) where VS is the sulfur vacancy and Xad is the adsorbed atom is isoenthalpic for Br and Cl. Furthermore, for Cl the respective energy gain is actually greater than the formation energy of the vacancy (1.3 and 1.7 eV in sulfur- poor MoS2 and WS2 , respectively). 3 We have so far considered the stability of the neu- tral defects. Now the most important requirement for a dopant is that its ionization energy ED is not greater I than a few kT . The thermodynamic transition level ED (q/q + 1) can be defined as the value of the Fermi level for which charge states q and q + 1 of the defect D have the same formation energy. The position of the ED (q/q + 1) level relative to the valence band top Ev can be found from the formation energies (see Eq. 4)[24] ED (q/q + 1) = Ef [X q ] − Ef [X q+1 ] − Ev . (7) and for donors Thus for acceptors ED (0/+) ≡ ED I ED (−/0) ≡ Eg − ED I . For comparison, we have also calculated the same defect levels using the marker method (MM). In this method, the ionisation energies/electron affinities of de- fective supercells are compared with those of the pris- tine supercell,[25] and the spurious electrostatic inter- actions are partially canceled. There is good agree- ment between the levels calculate using the two meth- ods, in most cases within about 0.1 eV. Another indi- cation of the quality of the method is the agreement between the gap obtained from total energy difference Eg = ES (+) + ES (−) − 2ES (0) − 2δE , where ES (q) is the energy of the pristine supercell in charge state q and δE is the electrostatic correction of Ref.18, and the Kohn- Sham gap. These are respectively Eg =1.64 and 1.87 eV for MoS2 and WS2 , and Eg =1.65 and 1.77 eV for MoS2 and WS2 . Adsorbed Li is a shallow donor with a small ionisa- tion energy <0.1 eV both in MoS2 and WS2 . This is mainly due to two effects. First, the relaxation of Li in the positive charge state, which is of the order of 30 meV and is a physical effect; second, a spurious band filling effect,[26, 27] which are larger in WS2 due to the greatest dispersion of the lowest conduction band. The bandstructure shows inequivocally that Liad is a shallow donor. In effect, it merely gives out an electron to the conduction band, changing little the matrix bandstruc- ture in the vicinity of the gap (Supplementary Figure 1). Substitutional Br and Cl are shallow donors only above room temperature. They contribute with an additional electron to populate a perturbed conduction band state. The shallowest of them is BrS , with a ionisation energy of about 0.1-0.2 eV both in MoS2 and WS2 (Table II). Even though this is higher than the ionisation energy of shallow dopants in bulk materials such as Si or GaAs, it is lower than the dopant ionisation energies in layered BN.[28] Substitutional P is found to be a very shallow acceptor, with activation energy ∼ 0.1 eV in MoS2 , and < 0.1 eV in WS2 , comparable to the uncertainty of the calculation. Si is also an acceptor, though deeper. It is noticeable that ionisation energies in WS2 are usu- ally smaller, despite its larger calculated bandgap, sug- gesting that this material is easier to dope. In summary, we have shown that it is possible to dope MoS2 and WS2 with electrons or holes by chemical sub- stitution at the S site or adsorption on the top of the layer. Amongst the shallow donors, Liad has the lowest ionisation energy. The donated electron is predominantly localized on the transition metal d states. However, Li diffuses extremely fast in most materials and therefore is not a good choice for high temperature applications. Besides, BrS and ClS are also donors, but have a higher ionisation energy. The higher temperature required to excite the carriers is a trade-off for the higher tempera- ture stability of the defects. Phosphorus is a shallow acceptor with a very low ion- isation energy, comparable to the uncertainty of the cal- culation. The wavefunction of the unpaired hole state is a valence-band like state, predominantly localized on the transition metal layer. This suggests that the ionized PS defect will be a weak scattering center. The combination between the high stability of P and the fact that it con- tributes with a very delocalized electron to the material, preserving the characteristics of a 2D electron gas, indi- cate that its extrinsic region would extend up to much higher temperatures than for Si, that readily becomes intrinsic at about 800 K (Fig. 1). These findings open the way to the control of the con- ductivity type in these two materials, offering a way to use MoS2 and WS2 for transistor parts other than the channel, or even to integrate different funtionalities in the same layer. This seems extremely promising for the design of electronic and optoelectronic devices for high temperature operation. ACKNOWLEDGMENTS We thank the financial support from NRF-CRP award ”Novel 2D materials with tailored properties: beyond graphene” (R-144-000-295-281). We thank RM Ribeiro for providing pseudopotentials. The calculations were performed in the GRC high performance computing fa- cilities. ∗ [email protected] [1] M. Chhowalla, H. S. Shin, G. Eda, L.-J. Li, K. P. Loh and H. Zhang, Nature Chemistry 5, 263 (2013). [2] Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman and M. S. Strano, Nature Nanotechnology 7, 699 (2012), and references therein. [3] B. Radisavljevic, M. B. Whitwick, A. Kis, ACS Nano. 5, 9934 (2011). [4] B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, Nature Nanotechnol. 6, 147150 (2011). [5] W. Bao, X. Cai, D. Kim, K. Sridhara, and M. S. Fuhrer , arxiv.org/1212.6292 4 [6] C. Sire, S. Blonkowski, M. J. Gordon, and T. Baron, Appl. Phys. Lett. 91, 242905 (2007). [7] G.-X. Ni, Y. Zheng, S. Bae, C. Y. Tan, O. Kahya, J. Wu, B. H. Hong, K. Yao, B. Ozyilmaz, ACS Nano 6 3935-3942 (2012). [8] Y.-J. Zhao, C. Persson, S. Lany, and A. Zunger, Appl. Phys. Lett. 85, 5860 (2004). [9] D. J. Singh, Functional Materials Letters 1, 4 (2008) [10] H. Fang, M. Tosun, G. Seol, T. C. Chang, K. Takei, J. Guo, and A. Javey, NanoLett. 13, 1991 (2013). [11] Y. Du, H. Liu, A. T. Neal, M. Si, and P. D. Ye, Electron Device Letters, IEEE (in press). [12] H.-P. Komsa, J. Kotakoski, S. Kurasch, O. Lehtinen, U. Kaiser, A. V. Krasheninnikov, Phys. Rev. Lett 109, 035503 (2012). [13] C. Ataca and S. Ciraci, J. Phys. Chem. C 115, 13303 (2011). [14] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococ- cioni, I. Dabo, A. Dal Corso, S. Fabris, G. Fratesi, S. de Gironcoli, R. Gebauer, U. Gerstmann, C. Gougous- sis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari, R. M. Wentzcovitch, J.Phys.:Condens.Matter 21, 395502 (2009). [15] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). [16] H.-P. Komsa and A. V. Krasheninnikov, Phys. Rev. B 86, 241201(R) (2012) [17] A. Carvalho, R. M. Ribeiro and A. H. Castro-Neto, Phys. Rev. B 88, 115205 (2013). [18] H.-P. Komsa and A. Pasquarello, Phys Rev. Lett. 110, 095505 (2013). [19] The values of the correction for a defect in the middle of the slab, at the interface and on the surface were 0.19, 0.21 and 0.19 eV respectively for MoS2 and 0.19, 0.21 and 0.16 eV respectively for WS2 . [20] H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13 , 5188 (1976). [21] T. Ito and K. Naka jima, Philosophical Magazine Part B 37, 773 (1978). [22] V. Wei, S. Seeger, K. Ellmer, and R. Mientus, J. Appl. Phys. 101, 103502 (2007). [23] M. Potoczek, Przybylski and M. Rekas, J. Phys. Chem. Solids 67, 2528 (2006). [24] The value of Ev obtained from the bandstructure of the pristine supercell is usually corrected by alignemento of the average electrostatic potentials [29]. However, since in this case a great fraction of the supercell was occu- pied by vaccuum in the alignment term was found to be negligible.. [25] J. Coutinho, V. J. B. Torres, R. Jones and P. R. Briddon, Phys. Rev. B 67, 035205 (2003) [26] A. Carvalho, A. Alkauskas, A. Pasquarello, A. K. Tagant- sev, and N. Setter, Phys. Rev. B 80, 195205 (2009). [27] S. Lany and A. Zunger, Phys. Rev. B 80, 085202 (2009). [28] F. Oba, A. Togo, I. Tanaka, K. Watanabe and T. Taniguchi, Phys. Rev. B 81, 075125 (2010) [29] C. G. van de Walle and J. Neugebauer, J. Appl. Phys. 95, 3851 (2004). 5 Isosurfaces of the unpaired electron state of Liad FIG. 3. (a) and of the unpaired hole state of PS (b), in MoS2 , as generated by fully relativistic calculations. The former is a donor, whereas the latter is an acceptor. The square of the wavefunction is represented. W and S are represented by cyan and yellow spheres, respectively. TABLE I. Formation energy of substitutional impurities (Ef ) along with adsorption energies. All values are in eV and refer to the neutral charge state. MoS2 Defect E S−poor (DS ) Ef (Dad ) E S−poor f f −0.3 −0.7 −1.0 BrS −0.9 −0.9 −1.5 ClS LiS −0.7 −2.0 −0.9 −2.7 −0.7 −2.9 PS SiS −2.6 −1.6 −2.0 WS2 (DS ) Ef (Dad ) −0.7 −0.9 −1.5 −0.6 −0.9 TABLE II. Defect-related levels in MoS2 and WS2 . E (−/0) is given relative to Ev and E (0/+) is given relative to Ec . FEM and MM stand for Formation Energy Method and Marker Method, respectively (see text). All values are in eV. (−/0) FEM – – – 0.11 0.39 (−/0) FEM – – – 0.02 0.23 MoS2 (−/0) MM – – – 0.06 0.34 WS2 (−/0) MM – – – −0.09 0.12 Method BrS ClS Liad PS SiS Method BrS ClS Liad PS SiS (0/+) FEM 0.15 0.18 −0.02 – – (0/+) FEM 0.14 0.18 −0.36 – – (0/+) MM 0.22 0.27 0.12 – – (0/+) MM 0.14 0.22 −0.16 – – Supplementary Information on: Donor and Acceptor Levels in Semiconducting Transition Metal Dichalcogenides A. Carvalho1, A. H. Castro Neto1 1Graphene Research Centre, National University of Singapore, 6 Science Drive 2, Singapore 117546∗ x ) V e ( v E - E 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 Γ a) ClS b) BrS c) Liad d) PS e) SiS K Γ M K M Γ K M Γ K M Γ K M FIG. 1. Fully relativistic Kohn-Sham bandstructures of defects in monolayer WS2 . The bandstructure of the pristine monolayer is represented shaded, in the same energy scale as the respective defect bandstructure. The calculations were performed in a supercell consisting of 4 × 4 primitive cells. ∗ [email protected] 

1110.4530

3

1110

Polarization and polarization induced electric field in nitrides - critical evaluation based on DFT studies

Density Functional Theory (DFT) calculations were used to evaluate polarity of group III nitrides, such as aluminum nitride (AlN), gallium nitride (GaN) and indium nitride (InN) providing physically sound quantitative measure of polarity of these materials. Two different approaches to polarization of nitride semiconductors were assessed and the conclusions have been used to develop models. It was shown that Berry phase formulation of the electron related polarization component provides a number of various solutions, different for various selection of the simulated volume. The electronic part gives saw-like pattern for polarization. A total number of these solutions, related to well known scaling of the geometric phase, is equal to the number of valence electrons in the system. Summation with similar pattern for ionic part provides several polarization values. Standard dipole density formulation depends on the selection of the simulation volume in periodic continuous way. Using a condition of continuous embedding into the infinite medium, and simultaneously, the zero surface charge representation at crystal boundary provides to physically sound solution. This solution is corresponding to maximal and minimal polarization values and also corresponds to different physical termination of the crystal surfaces, either bare or covered by complementary atoms. This change leads to polarization and electric field reversal. The polarization and related fields in finite size systems were obtained.

cond-mat.mes-hall

cond-mat

Polarization and polarization induced electric field in nitrides – critical evaluation based on DFT studies Pawel Strak*, Pawel Kempisty*, Konrad Sakowski*, and Stanislaw Krukowski*,† *Institute of High Pressure Physics, Polish Academy of Sciences, Sokolowska 29/37, 01-142 Warsaw, Poland †Interdisciplinary Centre for Materials Modeling, Warsaw University, Pawinskiego 5a, 02-106 Warsaw, Poland ABSTRACT It was shown that polarization could be defined only for finite slabs with specified boundary condition. In the case of nitrides in wurtzite structure two different polarizations occurred which differ by magnitude and direction of built-in electric field. Density Functional Theory (DFT) calculations were used to evaluate polarity of group III nitrides, such as aluminum nitride (AlN), gallium nitride (GaN) and indium nitride (InN). Two different approaches to polarization of nitride semiconductors were assessed. It was shown that Berry phase formulation of the electron related polarization component provides nonzero polarization even for single atom and additionally a number of various solutions, different for various selection of the simulated volume of the nitrides. The electronic part gives saw-like pattern for polarization. Alternative standard dipole density formulation of polarization depends on the selection of the simulation volume in periodic continuous way. A condition of continuous embedding into the infinite medium, and simultaneously, the zero surface charge representation at crystal boundary provides to physically sound solution. These solutions correspond to maximal and minimal polarization values. These solutions arise from different physical termination of the crystal surfaces, either bare or covered by complementary atoms. This change leads to polarization and electric field reversal. The polarization and related built-in electric fields were obtained. Keywords: polarization, polarity, gallium nitride, aluminum nitride, indium nitride, density functional theory PACS: 61.50.Ah, 81.10.Aj I. INTRODUCTION Macroscopic polarization of crystalline materials is physically important property of solids1. Polarization is related to symmetry of a crystalline lattice, which allows for the existence of vector property, in the point group symmetry of the lattice. It is therefore expected that microscopic 1 definition of the property is formulated, in direct relation to the atomic lattice structure. Unfortunately typical statement, frequently encountered in textbooks, defining polarization as dipole of a unit cell, divided by its volume, was criticized2-4. It was claimed that so defined quantity depends on the selection of the simulation volume so it could not be used to determine the physical property of the matter. The statement was related to the presence of the surface term, inherently involved in the definition of the polarization as an electric dipole density, which affects its value5-9. A direct modification, such as using of large volume, does not remove this deficiency2. It is still not clear whether additional requirement of independency of the selection of simulation volume should be applied to determination of the polarization. Recent results shed serious doubts whether such requirement could be met for infinite systems10-12. Macroscopic polarization is technically important physical property of pyro- and ferro- electrics, often used in many important technological applications. A natural use of permanent or induced electric dipole could be found in many applications13. Natural consequence of polarization is presence of an electric field in the crystal interior possibly affecting functionality of advanced electronic and optoelectronic devices. The built-in electric fields affect energy of quantum states that is known as Stark effect. In addition, strain induced field may contribute to this effect significantly, especially in strained quantum low dimensional structures frequently used in modern devices, built on polar GaN(0001) surface14-17. The electric field changes energy of quantum states of both types of carriers, electrons and holes, giving rise to phenomenon for long time known as Quantum Confined Stark Effect (QCSE) 17. A mere change of the energy of quantum states could be either beneficial or harmful; a really detrimental is spatial separation of electrons and holes that are shuffled to the opposite ends of the quantum well17-19. Spatial separation reduces an overlap of the hole-electron wavefunctions, their radiative recombination rates and lowers efficiency of photonic devices20-23. The negative influence of QCSE may be enhanced by Auger recombination or carrier leakage at high injection currents24-28. That could lead to decrease of the device efficiency for higher injection currents, the phenomenon nicknamed as “efficiency droop”29. A harmful influence of QCSE for optoelectronic devices is compensated by its beneficial contribution on electronic devices based on two-dimensional electron gas (2DEG), such as field effect transistors (FET’s) or molecular sensors. Electric field at AlN/GaN heterostructures stabilizes 2DEG leading to high carrier mobility which may be used for construction of fast electronic devices. The electric field, induced by dipoles of the molecules, attached to the surface, may contribute to the sensitivity of molecular sensors which opens new applications of such devices. Polarization is therefore a physical property that is an increasingly important in technology. Recently formulated approach was to divide polarization into ionic and delocalized charge and to calculate the change of polarization only8, 9. The ionic part may be calculated directly, the delocalized contribution may be obtained using Berry phase formulation2-7. This procedure provides required quantity modulo some factor. In the work presented below we will critically asses this formulation and 2 compare these results with reformulated standard definition of polarization as dipole density with additional condition imposed at boundaries. These conditions allow incorporation of the vacuum in the infinite crystal body and let to avoid generation of surface charge. Thus polarization is determined exactly without any inference from the boundary terms. As it is also presented below, polarization induced electric field in the bulk solid may be determined. That determination is compatible with the selection of the termination surfaces, thus provide base for accounting of the field influence on the properties of optoelectronic and electronic devices. II. CALCULATION METHODS In the calculations reported below three different DFT codes were used: commercially available VASP30-32, freely accessible SIESTA33-35 and commercially accessible Dmol36. In the first instance a standard plane wave functional basis set, as implemented in VASP with the energy cutoff of 29.40 Ry (400.0 eV), was used. As was shown by Lepkowski and Majewski, it gives good results in precise simulations of GaN properties by VASP code37. The Monkhorst-Pack grid: (7x7x7), was used for k-space integration38. For Ga, Al, In and N atoms, the Projector-Augemented Wave (PAW) potentials for Perdew, Burke and Ernzerhof (PBE) exchange-correlation functional, were used in Generalized Gradient Approximation (GGA) calculations39-41. Gallium 3d and Indium 4d electrons were accounted in the valence band explicitly. The energy error for the termination of electronic self- consistent (SCF) loop was set equal to 10-6. The obtained lattice constants were: GaN - a = 3.195 Å and c = 5.206 Å, AlN - a = 3.112 Å and c = 4.983 Å, and InN - a = 3.563 Å and c = 5.756 Å, which is in good agreement with the experimental data: GaN: a = 3.189 Å and c = 5.185 Å, for AlN: a = 3.111 Å and c = 4.981 Å and for InN: a = 3.537 Å and c = 5.706 Å. The second code, SIESTA uses norm conserving pseudopotentials with the numeric atomic orbitals local basis functions that have finite size support, determined by the user33-35. The pseudopotentials for Ga, Al, In and N atoms were generated, using ATOM program for all-electron calculations42,-43. Gallium 3d and Indium 4d electrons were included in the valence electron set explicitly and were represented by single zeta basis. For s and p type orbitals quadruple zeta basis set were used. Aluminum atom basis was represented by triple zeta function. Integrals in k-space were performed using 3x3x3 Monkhorst-Pack grid. The minimal equivalent of plane wave cutoff for grid was set to 275 Ry. As a convergence criterion terminating SCF loop, the maximum difference between the output and the input of each element of the density matrix was employed being equal or smaller than 10-4. For comparison, spontaneous polarization was also calculated using DMol3 commercial program36. In DMol3 package, full-electron Kohn-Sham eigenvalue problem with periodic boundary conditions (PBC) for wave function is solved, using basis of Linear Combination of Atomic Orbitals (LCAO) 44. The double zeta plus polarization (DZP) basis set was applied, in which required matrix elements are evaluated numerically on the properly chosen grid. For exchange energy, approximation 3 proposed by Becke45 and Lee, Yang, Parr46 (B88) was chosen, for correlation energy Tsuneda, Suzumura, Hirao functional was used47. The electric potential was determined by solution of Poisson equation which was expressed as sum of the multipole contributions, centered on each atom of the simulated volume with the optimized cutoff radii36, 48. In order to account long range interactions Ewald summation is employed49. The method is verified to assure solution tolerance of 10-6 hartree. Thus the DMol3 Poisson equation solution procedure is different than that of VASP and SIESTA. III. DENSITY DISTRIBUTION The electric fields in VASP and SIESTA were obtained by inverse Fast Fourier Transform (FFT) method based on periodicity of electrostatic potential at the edges of the simulated volume. Such periodicity is enforced by implicit addition of appropriate external field so that any potential changes related to polarization are compensated. DMol3 solution is based on multipole expansion method, employing Ewald summation procedure to account long range interaction. Since the DMol3 wavefunctions basis, are spherical harmonics, that are perfect electric multipoles, the summation employs coefficients only, assuring extremely fast convergence of the Poisson equation solution procedure. Siesta uses both FFT method and molecular orbitals basis set while VASP is combination of FFT and planewaves, therefore such combination of the codes allows exhaustive verification of these three approaches. In Fig. 1 the c-plane averaged density profiles were presented for AlN, GaN and InN. The diagrams presents the electronic density arising from simple superposition of atomic charges SAOρ , the density obtained by full solution of DFT Kohn-Sham equation KSρ and the difference of these two: ∆ ≡ − ρ ρ ρ SAO KS (1) Naturally, any superposition of charge of separated atoms has no electric dipole moment, therefore either Kohn-Sham density KSρ or the density difference ρ∆ may be used for calculation of the polarization and the polarization induced electric field. In principle the results obtained using both approaches should differ only by an error arising from intersection of the subtracted separated atom charge (i.e. effectively positive charge, representing atomic nuclei) by top and bottom boundaries. The coordinate shift, resulting from the periodic boundary conditions may affect magnitude of the dipole obtained from the integration over simulated volume. 4 N Al N Al KS SAO 0 1 2 3 4 5 z [Å] N Ga N Ga KS SAO 0 1 2 3 z [Å] 4 5 6 N In InN KS SAO 0 1 2 4 5 6 3 z [Å] 0,75 0,60 0,45 0,30 0,15 ] 3 Å / e [ ρ ] 3 Å / e [ ρ ] 3 Å / e [ ρ 1,8 1,5 1,2 0,9 0,6 0,3 0,0 1,2 1,0 0,8 0,6 0,4 0,2 0,0 ] 3 Å / e [ ρ ∆ 0,06 0,04 0,02 0,00 -0,02 -0,04 -0,06 VASP DMOL SIESTA 0 1 2 3 4 5 z [Å] ] 3 Å / e [ ρ ∆ ] 3 Å / e [ ρ ∆ 0,06 0,04 0,02 0,00 -0,02 -0,04 -0,06 0,06 0,04 0,02 0,00 -0,02 -0,04 -0,06 VASP DMOL SIESTA 0 1 2 3 z [Å] 4 5 VASP DMOL SIESTA 0 1 2 4 5 6 3 z [Å] Fig. 1. (Color online) Left column presents superposition of individual atom charges density ( SAOρ - red dashed line) and DFT Kohn-Sham solution ( KSρ - black solid line) obtained by VAPS code. Right column presents the density difference ρ∆ . The densities are averaged in the plane perpendicular to c- axis diagrams, along which they are plotted for: AlN(top); GaN(middle); InN(bottom). 5 It is worth noting that the density difference is relatively small. It is remarkable that these solutions are close for all three methods. That indicates on good convergence of the calculations and proper representation of the real density distribution by DFT solution. Boundary conditions enforce periodic density distribution which does not specify simulated volume entirely as its boundaries can be selected arbitrarily. Since the simulated system is electrically neutral, integral over the density should vanish, thus there are at least two different locations of the boundaries that of the zero averaged density difference. In principle any other choice could be adopted, giving rise to different values of the dipole moment of the sample, and consecutively different value of polarization. As it is shown below, the appropriate treatment of the boundary problem allows us to obtain well defined, physically sound magnitude of the dipole and consequently, the polarization. IV. POLARIZATION OF BULK AlN, GaN AND InN Polarization in the solids and in the molecules arises from the electronic charge transfer resulting from bonding that leads to emergence of electrical dipoles. The polarization may be obtained from its definition, being equal to the dipole density: (cid:1) P where Ω is the simulated volume, = r (cid:1) 3 d r ( ) rρ 1 Ω ∫ ( )rρ charge density, both electronic and nuclei. Naturally (2a) polarization of superposition of charges of individual atoms should have polarization equals zero, i.e. 1 Ω ∫ For the relation (2b) fulfilled, the polarization may be obtained equally from Kohn-Sham density ( ) rρ SAO 3 d r r (cid:1) (cid:1) P 0 = = (2b) KSρ or the density difference ρ ρ ρ ∆ = − SAO KS , which merely reflects above stated fact that it arises from electronic charge transfer: = = ∫ ∫ (cid:1) P r (cid:1) 3 d r 3 d r ( ) r ρ KS 1 1 Ω Ω It was recognized that for electrically neutral systems the result of such procedure does not depend on the choice of the coordinate system, nevertheless it depends on the choice of the simulated area4, 6, 7. It was a subject of long debate whether polarization may be uniquely defined as bulk properties of the solids, independent of their surfaces8,9. A simple model presented in Fig. 2. show that the polarization depends on the boundary conditions, even in the simplest case for the charge transfer represented by ( ) ρ ∆ r (2c) r (cid:1) point charges. 6 - + + - - + + - - + + - - + - + + - - + + - - + --- + - + P1 P2 Fig. 2. (Color online) Chain of metal (Al, Ga, In – large red circles) and N (small blue circles) atoms, representing polarization properties of nitride lattice. Top – superposition of individual atoms, electrically neutral, for which both polarization and average electric field vanish, center and bottom – two polarized (charged) lattices, differ by shift of the edge metal atom (bonding is denoted by dotted line). The edge atom, i.e. the one shifted, is represented in the top diagram by broken line. Polarization is represented by black arrows. As it is shown in Fig. 2, at least two equivalent polarization values could be defined, depending on the termination of lattice, i.e. on the position of the edge atoms. These two polarizations differ by both magnitude and the direction. Therefore, the definition of the polarization as purely bulk property, without reference to termination of the lattice cannot by physically justified. Polarization of the infinite medium cannot be uniquely determined, as at least the two different polarization values exist for the infinitely thick slab. The difference is not related to calculation method, it is physical in nature, as different electric field in the medium arises due to different boundaries. Therefore the polarization could be defined in finite slab only. In addition, the procedure determining polarization and the polarization related field has to enforce zero surface charge and zero external field for these two quantities, respectively. More recent approach is based on division of the polarization into the ionic and electronic contributions: where ionic contribution is treated in standard manner using summation over all point charges of the (cid:1) (cid:1) = P P ion + (cid:1) P el (2) nuclei Ze : 1 Ω and Berry phase formulation is applied for electronic part, based on linear response theory adiabatic ∑ (cid:1) ionP (3) = Ze ⋅ r (cid:1) change of the potential, controlled by parameter λ: 7 ∆ (cid:1) P el = − 2 e Ω ∑ ∫ n ( 1 (cid:1) r w n − ) 3 0 w d r n (4) where the sum runs over all occupied real space Wannier functions 0 nw and 1 nw , calculated for both terminations of the adiabatic path (superscripts correspond to λ = 0 and λ = 1, respectively) 4-7, 49. A. Polarization obtained from Berry phase formulation Berry phase formulae for polarization, given in Refs 4-7, modified for the application to USPP's and PAW datasets50, were used in the version implemented in VASP package. In order to obtain good approximation for band gap we have recalculated wavefunctions for PBE charge density with HSE03 functional51. Determination of polarization from geometric phase formulation relies on adiabatic change of the crystal potential. Derivation of polarization by Resta is limited to electronic part only2. The ionic and electronic contributions, obtained from Eq. 4 and 5 respectively without assumption of electric neutrality, depend on the coordinate system and the truncation of the integration area. As an initial test, the Berry phase polarization was determined using single Ga atom in the 20 Å long cell. ] Å e [ z d 150 100 50 0 -50 -100 -150 0 4 8 12 ∆z [Å] ion dz elec dz tot dz 16 20 Fig. 3. (Color online) Electric dipole of single gallium atom located in 20 Å long simulation cell as a function of the shift of the periodic cell related to the Ga atom along z-axis obtained from Berry phase formulation implemented in VASP: ionic part (Eq. 4) – black dash-dotted line, electronic 0 z∆ = corresponds to a system with Ga atom part (Eq. 5) – red dashed line, total – solid blue line. located in the center of a cell. z∆ = 10 corresponds to a system with Ga atom located in the boundary of a cell. 8 As shown in Fig.3. the Berry phase expression is not constant, it depends on the location of single atom in the simulation cell. Next, the solution obtained by the iteration procedure where the density is kept constant, equal to SAOρ . The SCF iteration loop converged to the wavefunction describing nonpolarized state of the system. Naturally, the Berry phase procedure should give the polarization of the system equal to zero. As shown in Fig. 4, the polarization of Ga-N system is not zero. Moreover, there is no such selection of the simulation volume for which the total polarization vanish. ] Å e [ z d 30 20 10 0 -10 -20 -30 -40 dion delec dtot 4 5 0 1 2 3 ∆z [Å] Fig. 4. (Color online) “Z” component of a Ga-N dipole as a function of shift of the periodic cell along c-axis obtained from Berry phase formulation implemented in VASP code: ionic part (Eq. 4) – black dash-dotted line, electronic part (Eq. 5) – red dashed line, total – solid blue line for the wavefunction compatible with the density being the sum of densities of separated atoms SAOρ of gallium and nitrogen atoms. Subsequently, the full solution of Kohn-Sham equation was obtained in which the relaxation procedure was performed for all three nitrides: AlN, GaN and InN. 9 ion dz ele dz tot dz 20 15 10 5 0 -5 -10 -15 ] g n A e [ z d 0 1 2 3 ∆ z [Ang] 4 5 ] Å e [ z d 30 20 10 0 -10 -20 -30 -40 ] Å e [ z d 40 30 20 10 0 -10 -20 -30 -40 -50 dion delec dtot 0 1 2 3 ∆z [Å] 4 5 dion delec dtot 0 1 2 4 5 6 3 ∆z [Å] Fig. 5. (Color online) “Z” component of a dipole of simulation cell as a function of shift of periodic cell along c-axis, obtained from Berry phase formulation VASP code: ionic part (Eq. 4) – black dash-dotted line, electronic part (Eq. 5) – red dashed line, total – solid blue line; (top) AlN; (middle) GaN; (bottom) InN. 10  As claimed by Resta et al, combination of electronic and ionic contributions should be independent of the system coordinates, and also of the truncation of the volume. As expected, the ionic contribution is linear function of the z coordinate that undergoes jumps when the N or Ga atoms cross the boundaries. For 4 atoms in total, four jumps in ionic contribution was obtained for all three nitrides. It was also argued that the electronic contribution is determined modulo periodic change of the geometric phase due to translation of the crystal as a whole, which leads to the following uncertainty: ∆ (cid:1) P el = fe Ω ∑ n (cid:1) R n (5) where f is the number of the electrons in valence band and R is the lattice period. Accordingly, the number of electrons is: f = 8 for AlN, f = 18 for GaN and InN, and the number of the jumps follows this prediction4. As expected the total polarization being combination of these two contributions is constant but it contains a number of jumps. B. Polarization obtained from dipole density As it was argued above, the polarization has to be simulated using finite region with appropriate boundary conditions, preventing emergence of surface charge which may affect the results. Nevertheless, in typical ab intio calculations small size cell with periodic boundary conditions for electronic density is used. It is therefore natural to verify whether the periodic boundary conditions affect the determined electric dipole moment. The results of the calculations for single gallium atom related dipole in 20 Å cell are presented in Fig. 6. elec dz ion dz tot dz ] Å e [ z d 200 100 0 -100 -200 0 4 12 16 20 8 ∆z [Å] 11 Fig.6 (Color online). Electric dipole of gallium atom in 20 Å long cell, in function of the shift of the periodic cell along z-axis: electronic part – black dash-dotted line, ionic contribution – red dashed 0 z∆ = corresponds to a system with Ga atom located in the center of a line, total – solid blue line. cell. z∆ = 10 corresponds to a system with Ga atom located in the boundary of a cell. As shown here, the obtained dipole is affected only for these locations where the boundary intersects electronic charge close to Ga atom. Otherwise, the result is zero as expected for nonpolarized system. Similar behavior was obtained for cell simulation of the system of gallium and nitrogen atoms. The dipole related polarization is calculated for the case of density obtained from superposition of atomic charges SAOρ and solution of Kohn-Sham equation KSρ , presented above in Fig.1. ] Å e [ d 150 100 50 0 -50 -100 SAO DFT 150 100 50 0 -50 -100 ] Å e [ d 0 1 2 3 z [Å] 4 5 0 1 2 3 z [Å] 4 5 Fig.7. (Color online). Electric dipole of GaN cell in function of the position of the cell boundary: left - determined for density obtained for superposition of atomic charges SAOρ ; right - solution of Kohn-Sham equation KSρ : electronic part – black dash-dotted line, ionic contribution – red dashed line, total – solid blue line. As shown in Fig. 7, direct determination of the dipole is strongly affected by boundary conditions, shifting part of the charge and completely changing the obtained results. In addition, the selection of arbitrary conditions is not compatible with the requirement of the zero surface charge, as the finite density cannot be smoothly terminated, i.e. the consistent method of simulation both the cell representing the volume, and the cell at the boundary of the slab, necessary for simulation of finite systems. For finite slab it is necessary to set zero surface charge in order to prevent additional, spontaneous polarization irrelevant contributions. As there should be no charge or dipole layer at the surfaces, both the electric potential and the electric field are continuous across the system boundaries. In addition, as the modeled sector should be naturally embedded into the infinite medium, the electric 12 potential, the field and its normal derivative should be continuous across the interface, amounting to the following conformity conditions lim 0 → ε lim 0 → ε ( ) φ ε = z = (cid:1) ( E z = ) ε = lim 0 → ε lim 0 → ε ( φ z ( ) ) 0 ε φ = − = (cid:1) ( E z ) ε = − = (cid:1) E (0) lim 0 → ε ( ∂ E z z ∂ z = ) ε ∂ = lim 0 → ε ) ε = − ( E z z ∂ z = ( ∂ E z z ∂ z = 0) (6) (7) (8) The considered field could be integrated over the area of the boundary to get average values, which are function of z coordinate only. Since the simulated area can be truncated at any position and the electric field follows the equation: z ( ) ∂ E z z ∂ = ( ) zρ ε (9) in which the averaged density profiles, plotted in Fig.1 could be used. Note that the boundary shift should entail appropriate rearrangement of the density distribution. In order to model the polarization exactly, it is required that the boundaries of the simulated regions should represent the surfaces of real crystal, without additional surface charge. The physically sound approach is that the electronic density difference vanishes outside. Abrupt termination in nonphysical, and any modification of Kohn-Sham density KSρ to assure continuous could in principle provide additional surface charge. In contrast to that this condition can be imposed for the density difference ρ∆ so this quantity is used in polarization determination below. Application of the condition (Eq. 9) to assumption of continuity (Eq. 8) entails constant density outside the simulated region as shown in Fig. 6. ] 3 Å / e [ ρ ∆ 0,04 0,00 -0,04 ρ0 ρ1 ρ2 ρ3 ρ4 4 5 6 7 8 9 10 11 Fig. 8. (Color online) Several choices of the truncation of AlN density difference profiles obtained z [Å] from VASP and its continuation in accordance to the conformity conditions in Eqs. 6-9. 13 Therefore, only the two selected terminations, those plotted by solid lines in Fig. 8 for which the density difference is zero outside are suitable for calculation of the polarization. It has to be stressed out that this conclusion results from simultaneous requirement that the embedded region is smoothly incorporated into the crystal body and their edge should represent the surface of real crystal without any surface charge. Due to electric neutrality condition, at least two such terminations could be found for any polarized system. In fact the difference in the selection of the simulated volume directly affects the resulting value of the electric dipole. Using obtained surface averaged electron density difference profiles, the AlN electric dipole magnitude was calculated directly by integration according to Eq. 2c. The result is presented in Fig. 9. ] Å e [ e l o p i d 1,0 0,8 0,6 0,4 0,2 0,0 -0,2 -0,4 -0,6 -0,8 ∆ρ4 ∆ρ3 Berry Phase ∆ρ2 ∆ρ1 ∆ρ = 0 0 1 2 3 4 5 z [Å] Fig.9. (Color online) Electric dipole of the AlN simulated volume, in function of the shift along c-axis. The results obtained by integration of the profiles truncated in the way presented in Fig. 2 are denoted by red squares. For real slab zero density condition outside is valid which selects from all possible truncations only the two that are denoted by solid lines in Fig. 8. As expected these two selections correspond to maximal and minimal values of the dipole plotted in Fig. 9. In fact, these two selections correspond to two possible termination of the top polar surface, i.e. by nitrogen or aluminum atoms. These two cases correspond to the top surface having aluminum triply bonded atoms that are either bare or nitrogen covered. In order to fulfill chemical stoichiometry slab criterion, such change of the top surface and position of the layer of N atoms has to be accompanied by the appropriate rearrangement of the bottom surface. In summary this leads to reversal of the polarization dipole, as shown in Fig. 3. 14 It has to be noted that the modeled area and the interfaces does not describe the real atomic configuration of any nitride surface. This is a simulation model devised to obtain the polarization and electric field in uniformly polarized nitrides (cid:1) 0E , corresponding to minimal energy of the nitrides under no external field. The magnitude of such field will be determined in the next Section. C. Polarization - summary It was postulated that the Berry phase results should give the physically meaningful value of polarization by proper selection of the additional constant value given by Eq. 10 (Ref. 4). This is not as straightforward as it was supposed to be. The number of the possible selection of arbitrary constant is large, even for simple structure of the nitrides. Naturally, the Berry phase result should be within the interval obtained from dipole calculations. This is not the case, but it could be reconciled by additional contribution which may bring the Berry phase result to this interval. In order to compare these results, the obtained polarization values are summarized in Table I. Table I. Polarization of nitrides; AlN, GaN and InN, obtained from dipole and Berry phase calculations (in e/Å2) AlN GaN InN Berry phase -0.3220 -0.2028 -0.1432 -0.0836 0.0357 0.0952 0.1549 0.3934 -0.7666 -0.5404 -0.3142 -0.2572 -0.0880 -0.0335 0.0250 0.1382 0.2513 0.3082 0.3644 0.4774 0.5905 -0.7205 -0.5386 -0.3568 -0.1749 -0.1275 0.0070 0.0980 0.1889 0.2798 0.3256 0.3708 0.4617 0.5526 ∆ρ -0.01485 0.01979 -0.0128 0.0153 -0.0114 0.0123 15 These polarization values were obtained using the following cell volumes: ΩAlN = 41.79 Å, ΩGaN = 46.03 Å and ΩInN = 63.30 Å. Using Eq. 10 and the lattice constants c AlN = 4.983 Å, c GaN = 5.206 Å and c InN = 5.756 Å, the following additive constants for polarization were obtained: ∆Pel-AlN = 0.954 e/Å2, ∆Pel-GaN = 2.036 e/Å2 and ∆Pel-InN = 1.637 e/Å2. These values are relatively high compared to the dipole results obtained with method based on density difference Eq. 1. It is worth mentioning that the jumps in the electronic part follow Eq. 10 for f = 1, i.e. accounting single electron contribution only. Such jumps could be translated into the following additional constant values: ∆Pel-AlN-1 = 0.1192 e/Å2, ∆Pel-GaN-1 = 0.1131 e/Å2 and ∆Pel-InN-1 = 0.0909 e/Å2. In fact these polarization values obtained in dipole model may be approximated by appropriate subtraction of single electron values. Notably, the nitrides spontaneous polarization obtained by Bernardini et al.14 using Berry phase approach were: AlN: -0.081 C/m2 (-5.05 x 10-3 e/Å2), GaN: -0.029 C/m2 (-1.81 x 10-3 e/Å2), and InN: -0.032 C/m2 (-1.99 x 10-3 e/Å2). These results are slightly lower that our results obtained from dipole calculations. Additional difference stems from the fact that, the dipole formulation presented above, gives two different polarization values. V. POLARIZATION INDUCED FIELD IN BULK AlN, GaN AND InN. In addition to the polarization, the physically relevant quantity, directly affecting the performance of MQW based devices, is polarization induced electric field. Finite systems could be subjected to arbitrarily selected boundary conditions for potential, giving rise to different field inside. A standard case of a flat parallel plate capacitor demonstrates the dilemma. For the plates uniformly charged, the solution of Poisson equation having zero field outside is routinely selected by invoking additional argument that the potential should be finite at infinity. Formally, the second solution, that of the zero field inside and uniform nonzero field outside, fulfills Poisson equation as well as any normalized linear combination of these two solutions. Thus a plethora of the solutions exists, in which an additional argument of potential at infinity is applied to find physically sound solution. . In fact, this freedom is used in solution of Poisson equation by FFT method. Nevertheless, the physically sound solution is found using the same criterion as for capacitor with zero field at infinity. This solution was obtained for zero density outside, i.e. for the two selected cases above. This formulation allows obtaining the two possible values of the field, which corresponds to two selection of the crystal termination. In Fig. 5, the electric fields obtained for selected cases shown in Fig.2. The physically sound solutions are plotted using solid lines. 16  Fig. 10. (Color online) Electric field, averaged in c-plane, along c-axis, in single periodic cell of: (a) AlN; (b) GaN; (c) InN. 17 In addition, Fig. 11 presents the physically sound electric potential distributions plotted for AlN, GaN and InN. From these potential distributions the following potential differences were obtained: for AlN: ∆VN AlN = -18.00 V and ∆VAl AlN = 13.59 V; for GaN: ∆VN GaN = -14.49 V and ∆VGa InN: ∆VN InN = -12.85V and ∆VIn InN = 11.75 V. Since the following lattice constants were adopted for the modeling: c AlN = 4.983 Å, c GaN = 5.206 Å and c InN = 5.756 Å, the average fields are as follows: for AlN: EN 0,AlN = -3.612 V/Å and EAl 0,AlN = 2.727 V/Å; for GaN: EN 0,GaN = -2.783 V/Å and 0,GAN = 2.317 V/Å; for InN: EN 0,InN = -2.232 V/Å and EIn EGa 0,InN = 2.041 V/Å. These values are relatively high and they should be relatively easy to detect in nitride based structures. Note GaN = 12.06 V; for that screening could decrease or remove its influence, in the case when the size of the structures is comparable or larger than the characteristic screening lengths. (a) (b) 18  (c) Fig. 11. (Color online) Electric potential, averaged in c-plane, along c-axis, in single periodic cell of: (a) AlN; (b) GaN; (c) InN. Black (solid) and red (broken) lines correspond to two different possible electric field directions, related to different polarization values. These fields are to be used from in any geometrical arrangement of nitrides, encountered in electronic or optoelectronic devices by minimization of the electrostatic energy functional: )2 ( (cid:1) (cid:1) ε − E E 0 2 ∆ = ∫ W 3 d r (5) Naturally, the uniform polarization field oE(cid:1) polarization in the single finite size crystal without any surface or external contributions. These fields is the minimal energy solution, corresponding to uniform are the maximal field induced in finite size polarized semiconductors. In any real quantum structure, the fields should be lower, nevertheless that should affect properties of these structures considerably. VI. SUMMARY Two different approaches to polarization of nitride semiconductors were assessed. It was shown that Berry phase formulation of the electron related polarization component provides a nonzero polarization for nonpolarized system. The electronic part gives saw-like pattern for polarization. Additionally a number of various solutions, different for various selection of the simulated volume could be obtained. A total number of these solutions, related to well known scaling of the geometric phase is equal to the number of valence electrons in the system. Summed with similar pattern for ionic part, provides several polarization values. Standard dipole density formulation depends on the selection of the simulation volume in periodic continuous way. Using the condition of continuous embedding into the infinite medium, and simultaneously, the zero surface charge representation at crystal boundary, physically sound solution 19 could be identified. This solution corresponds to maximal and minimal polarization values and corresponds to different physical termination of the crystal surfaces, either bare or covered by complementary atoms. This change leads to polarization and electric field reversal. Values of the fields are maximal values possible for finite size polarized nitrides without any surface charges or externals fields. ACKNOWLEDGEMENTS The calculations reported in this paper were performed using computing facilities of the Interdisciplinary Centre for Modelling of Warsaw University (ICM UW). The research was partially supported by the European Union within European Regional Development Fund, through grant Innovative Economy (POIG.01.01.02-00-008/08). We would like to thank Michael Springborg from University of Saarland and Bernard Kirtman from University of California Santa Barbara for p discussion of importance of boundary conditions for polarization determination which has changed the present publication considerably. REFERENCES 1 L.D. Lanadau, and E.M. Lifsic, Electrodynamics of Continuous Media (Pergamon, Oxford, 1984). 2 R. Resta, Rev. Mod. Phys. 66, 899 (1994). 3 R. Resta, Ferroelectrics 136, 51 (1992). 4 R. M. Martin, Phys. Rev. B 9, 1998 (1974). 5 R. D. King-Smith, and D. Vanderbilt, Phys. Rev. B 47 1651 (1993). 6 D. Vanderbilt, and R.D. King-Smith, Phys. Rev. B 48 4442 (1993). 7 R. Resta, in Berry Phase in Electronic Wavefunctions, Troisième Cycle de la Physique en Suisse Romande, (Année Academique, 1996). 8 M. Posternak, A. Baldereschi, A. Catellani, and R. Resta, Phys. Rev. Lett. 64 1777 (1990). 9 A. K. Tagantsev, Phys. Rev. Lett. 69 389 (1992). 10 M. Springborg, B. Kirtman, Phys. Rev B 77, 045102 (2008). 11 M. Springborg, V. Tevekeliyska, B. Kirtman, Phys. Rev B 82, 165442 (2010). 12 M. Springborg, B. Kirtman, Theor. Chem. Accounts submitted (2011) . 13 M. E. Lines, and A.M. Glass, Principles and Applications of Ferroelectrics and Related Materials, (Clarendon, Oxford, 1974) 14 F. Bernardini, V. Fiorentini, and D. Vanderbilt, Phys. Rev. Lett. 79, 3958 (1997) 15 F. Bernardini, V. Fiorentini, D. Vanderbilt, Phys. Rev. B 56, R10024 (1997). 16 F. Bernardini, V. Fiorentini, Phys. Stat. Solidi B 216, 391 (1999). 17 M. P. Halsall, J. E. Nicholls, J. J.Davies, B. Cockayne, and P.J. Wright, J. Appl. Phys. 71, 907 (1992). 18 M. Buongiorno Nardelli, K. Rapcewicz, and J. Bernholc, Appl. Phys. Lett. 71, 3135 (1997). 20 19 S. P. Łepkowski, H. Teissyere, T. Suski, P. Perlin, N. Grandjean, J. Massies, Appl. Phys. Lett. 79, 1483 (2001). 20 T. Takeuchi, C. Wetzel, S. Yamaguchi, H. Sakai, H. Amano, I. Akasaki, Y. Kaneko, S. Nakagawa, Y. Yamaoka, and N. Yamada, Appl. Phys. Lett. 73, 1691 (1998). 21 R. Langer, J. Simon, V. Ortiz, N. T. Pelekanos, A. Barski, R. Ander, and M. Godlewski, Appl. Phys. Lett. 74, 3827 (1999). 22 J. S. Im, H. Kollmer, J. Off, A. Sohmer, F. Scholz, and A. Hangleiter, Phys. Rev. B 57, R9435 (1998). 23 P. Lefebvre, J. Allegre, B. Gil, H. Mathieu, N. Grandjean, M. Leroux, J. Massies, and P. Bigenwald, Phys. Rev. B 59, 15363 (1999). 24 Y. C. Shen, G. O. Muller, S. Watanabe, N. F. Gardner, A. Munkholm, and M. R. Kramers, Appl. Phys. Lett. 91, 141101 (2007). 25 I. V. Rozhansky, and D. A. Zakeheim, Phys. Stat. Solidi C 3, 2160 (2006). 26 M. F. Schubert, S. Chajed, J. K. Kim, E. F. Schubert, D. D. Koleske, M. F. Crawford, S. R. Lee, A. J. Fisher, G. Thaler, and M. A. Banas, Appl. Phys. Lett. 91, 231114 (2007). 27 J. Xie, X. Ni, Q. Fan, R. Schimada, U. Ozgor, and H. Morkoc, Appl. Phys. Lett. 93, 121107 (2008). 28 H. Teisseyre, A. Kaminska, G. Franssen, A. Dussaigne, N. Grandjean, I. Grzegory, B. Lucznik, and T. Suski, J. Appl. Phys. 105, 63104 (2009). 29 H. Morkoc, Handbook of Nitride Semiconductors and Devices (Wiley-VCH, Berlin, 2008), vol. 3. 30 G. Kresse, and J. Hafner, Phys. Rev. B 47, 558 (1993); ibid. 49, 14251 (1994). 31 G. Kresse, and J. Furthmüller, Comput. Mat. Sci. 6 15 (1996). 32 G. Kresse, and J. Furthmüller, Phys. Rev. B 54, 11169 (1996). 33 P. Ordejón, D. A. Drabold, M. P. Grumbach, and R. M. Martin, Phys. Rev. B 48, 14646 (1993). 34 P. Ordejón, D. A. Drabold, R. M. Martin, and M. P. Grumbach, Phys. Rev. B 51, 1456 (1995). 35 J. M. Soler, E. Artacho, J. D. Gale, A. García, J. Junquera, P. Ordejón, and D. Sánchez-Portal, J. Phys.: Condens. Matter 14, 2745 (2002). 36 B. J. Delley, J. Chem. Phys. 92, 508 (1990); ibid 113, 7756 (2000). 37 S. P. Łepkowski, and J. A. Majewski, Phys. Rev. B 74 35336 (2006). 38 H.J Monkhorst, and J. D. Pack, Phys. Rev. B 13 5188 (1976). 39 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77 3865 (1996). 40 P.E. Blöchl, Phys. Rev. B 50 17953 (1994). 41 G. Kresse, and D. Joubert, Phys. Rev. B 59 1758 (1999). 42 N. Troullier, and J. L. Martins, Phys. Rev. B 43, 1993 (1991), ibid 43, 8861 (1991). 43 Z. Wu, and R.E. Cohen, Phys. Rev. B 73 235116 (2006). 44 C. C. Roothaan, J. Rev. Mod. Phys. 32, 179 (1960); ibid 23, 69 (1951). 45 A. D. Becke, J. Chem. Phys. 88, 2547 (1988). 46 C. Lee, W. Yang, and R. G. Parr, Phys. Rev. B 37, 785 (1988). 21 47 T. Tsuneda, T. Suzumura, and K. Hirao, J. Chem. Phys. 110(22), 10664 (1999). 48 B. Delley, J. Phys. Chem. 100, 6107 (1996). 49 E. P. Ewald, Ann. Phys. 64, 253 (1921). 50 D. Vanderbilt, and R. D. King-Smith, in Electronic polarization in the ultrasoft pseudopotential formalism (Unpublished report, 1998). 51 J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003). 52 R.W. Nunes, and X. Gonze, Phys. Rev. B 63 155107 (2001). 22 

1802.01816

1

1802

Strong photon antibunching in weakly nonlinear two-dimensional exciton-polaritons

A deterministic and scalable array of single photon nonlinearities in the solid state holds great potential for both fundamental physics and technological applications, but its realization has proved extremely challenging. Despite significant advances, leading candidates such as quantum dots and group III-V quantum wells have yet to overcome their respective bottlenecks in random positioning and weak nonlinearity. Here we consider a hybrid light-matter platform, marrying an atomically thin two-dimensional material to a photonic crystal cavity, and analyze its second-order coherence function. We identify several mechanisms for photon antibunching under different system parameters, including one characterized by large dissipation and weak nonlinearity. Finally, we show that by patterning the two-dimensional material into different sizes, we can drive our system dynamics from a coherent state into a regime of strong antibunching with $g^{(2)}(0) \sim 10^{-3}$, opening a possible route to building scalable, on-chip quantum simulators.

cond-mat.mes-hall

cond-mat

Strong photon antibunching in weakly nonlinear two-dimensional exciton-polaritons Albert Ryou1, David Rosser2, Abhi Saxena3, Taylor Fryett1, Arka Majumdar1,2 1 Department of Electrical Engineering, University of Washington, Seattle, WA 98195, USA 2 Department of Physics, University of Washington, Seattle, WA 98195, USA 3 Department of Electrical Engineering, Indian Institute of Technology, Delhi, Hauz Khas, New Delhi 110016, India (Dated: February 7, 2018) A deterministic and scalable array of single photon nonlinearities in the solid state holds great potential for both fundamental physics and technological applications, but its realization has proved extremely challenging. Despite significant advances, leading candidates such as quantum dots and group III-V quantum wells have yet to overcome their respective bottlenecks in random positioning and weak nonlinearity. Here we consider a hybrid light-matter platform, marrying an atomically thin two-dimensional material to a photonic crystal cavity, and analyze its second-order coherence function. We identify several mechanisms for photon antibunching under different system parame- ters, including one characterized by large dissipation and weak nonlinearity. Finally, we show that by patterning the two-dimensional material into different sizes, we can drive our system dynamics from a coherent state into a regime of strong antibunching with g(2)(0) ∼ 10−3, opening a possible route to building scalable, on-chip quantum simulators. I. INTRODUCTION Quantum optical nonlinearities have received grow- ing interest for their key role in quantum informa- tion science1, quantum simulations2, and other quan- tum technologies3. While nonlinear effects with individ- ual emitters have been demonstrated across a range of platforms, including ultracold atoms4, superconducting qubits5, and semiconductor quantum dots6,7, realizing a deterministic and scalable array of such nonlinearities has proved a far more challenging task. For quantum dots, which are particularly attractive due to their ver- satility and on-chip compatibility8, random positioning and inhomogeneous broadening of the emitters remain formidable bottlenecks9,10. Another solid-state candidate for quantum nonlinear optics is the exciton-polariton, a quasiparticle made of a semiconductor exciton strongly coupled to a microcav- ity photon. Inheriting strong interactions from the mat- ter component and fast dynamics and state observabil- ity from the photonic component, exciton-polaritons are particularly well-suited as building blocks for photonic quantum simulations11–13. A host of many-body corre- lated phenomena with exciton-polaritons have been ob- served, including Bose-Einstein condensation14 and po- lariton lasing15. Nevertheless, there has been no report of a strong polariton-polariton interaction at a single quan- tum level. To increase the interaction strength, several researchers tried shrinking the size of the polariton wave- function. Besga et al. decreased the cavity mode volume by employing a fiber-tip cavity16, and recently Munoz- Matutano et al., using a similar setup, reported a weak nonlinearity17. Researchers have also tried decreasing the effective size of group III-V quantum wells, albeit with limited success18,19. Recent advances in atomically thin two-dimensional (2D) materials point to a new potential platform for scalable quantum optical nonlinearities. These materi- als, including graphene, hexagonal boron nitride, and transition-metal dichalcogenides (TMDCs), boast ex- ceptional light-emitting and light-harvesting properties, along with an unprecedented ability to be fabricated and transferred onto other photonic structures20. The TMDCs, in particular, hold great promise for inte- grated photonics due to their large, direct bandgap21. TMDCs embedded in microcavities have been employed to observe optically pumped lasing22,23, cavity-enhanced second harmonic generation? , electroluminescence24, and strong coupling25,26. Finally, Wei et al. showed that TMDCs patterned via electron beam lithography into cir- cular nanodots with radii down to 15 nm could still host long-lived excitons27. In this paper, we analyze the optical nonlinearity of a 2D-material monolayer coupled to a low mode-volume photonic crystal defect cavity. The strength of the quan- tum interaction can be revealed by its second-order co- herence function g(2)(τ ). We identify different mecha- nisms that give rise to non-classical photon distributions and arrive at a robust regime, characterized by large dis- sipation and weak nonlinearity, whose second-order co- herence at zero time delay is much less than unity. Fi- nally, we consider the effect of the size of the monolayer on the system parameters. We numerically show that by physically patterning the monolayer into different sizes, it is possible to drive its dynamics from a coherent state into a non-classical regime with g(2)(0) ∼ 10−3. An observa- tion of such strong photon antibunching in this hybrid platform would open the door to further experiments in coupled nonlinear cavities and scalable quantum simula- tors. II. SYSTEM DESCRIPTION Our system consists of a patterned 2D-material mono- layer placed on top of a photonic crystal nanobeam cav- ity (see Fig. 1a)28. The choice of a nanobeam has been motivated by its small cavity mode volume. The simu- 8 1 0 2 b e F 6 ] l l a h - s e m . t a m - d n o c [ 1 v 6 1 8 1 0 . 2 0 8 1 : v i X r a exciton-polariton system is given by (setting  = 1) H = ∆ca†a + ∆eb†b + g(cid:0)a†b + ab†(cid:1) + U b†b†bb + E(a† + a) 2 (II.1) where a†(a) and b†(b) are the creation (annihilation) op- erators for the cavity photon and the monolayer exciton, respectively; ∆c = ωc − ωpump and ∆e = ωe − ωpump are their frequency detunings relative to the pump laser; g is the exciton-photon coupling strength; U is the on- site Kerr nonlinearity representing the exciton-exciton repulsion29; and E is the strength of the pump laser. The system dynamics is given by the evolution of the density matrix according to the master equation30: i ρ = [H, ρ] + i + i Γ 2 (cid:0)2aρa† − a†aρ − ρa†a(cid:1) κ 2 (cid:0)2bρb† − b†bρ − ρb†b(cid:1) (II.2) (Color online) Patterned 2D FIG. 1: material-embedded cavity. (a) Schematic illustration of the proposed experimental platform. A patterned 2D-material (tungsten diselenide, WSe2) monolayer is placed on top of a photonic crystal nanobeam cavity. The radius of the monolayer is on the order of tens of nanometers. The top view of the cavity with a simulated field profile of the fundamental mode is shown below. The calculated mode volume is about 2.5(λ/n)3. (b) Energy level diagram. The dressed states are labeled by the number of energy quanta, or Fock manifold, followed by a symbol: 1,−(cid:105) and 1, +(cid:105) are the first-manifold states representing the lower and upper polaritons; 2, e1(cid:105), 2, e2(cid:105), and 2, e3(cid:105) are the second-manifold states. The solid lines represent the eigenenergies of the Hamiltonian with nonzero nonlinearity, whereas the dotted lines represent the eigenenergies with zero nonlinearity. The arrows represent the pump laser frequency that is resonant with either 1, +(cid:105) (blue) or 2, e3(cid:105) (red). (c) Eigenenergies as a function of the nonlinearity U, calculated via exact matrix diagonalization. All parameters are normalized by the exciton-photon coupling strength g. lated field profile of the fundamental mode of the cav- ity is shown below the schematic. Unlike the conven- tional semiconductor-embedded distributed Bragg reflec- tor cavity, whose excitons couple to a continuum of in- plane momenta, the monolayer-embedded photonic crys- tal cavity only supports a narrow band in the momen- tum space. Thus, in our model we consider only those excitons whose momenta match that of the fundamental cavity mode17. In a frame rotating at the frequency of an exter- nal pump laser, the Hamiltonian of a strongly coupled where κ and Γ are the inverse lifetimes of the cavity pho- ton and the exciton, respectively. The energy level diagram of the system containing up to two energy quanta is shown in Fig. 1b, where we have taken ωc = ωe. The degeneracy of the bare states is lifted by the exciton-photon coupling. The dressed states 1,−(cid:105) and 1, +(cid:105), containing one energy quan- tum and collectively known as the first Fock manifold of the Hamiltonian, represent the lower and upper polari- tons, respectively. Similarly, the second-manifold states, 2, e1(cid:105), 2, e2(cid:105), 2, e3(cid:105), containing two energy quanta, be- come nondegenerate. For zero exciton-exciton repulsion (U = 0), their eigenenergies are −2g, 0, and 2g (dotted lines), forming a harmonic energy ladder for two coupled oscillators. For U > 0, however, the eigenenergies shift (solid lines). The eigenenergies of the first (blue) and the second (red) manifold as a function of U are plotted in the rotating frame in Fig. 1c. The shifting of the second-manifold eigenenergies due to the exciton-exciton repulsion is the source of the quan- tum optical nonlinearity. Consider tuning the pump laser so that it resonantly excites the upper polariton 1, +(cid:105) (blue arrows in Fig. 1b). Whereas the first photon from the laser drives the system from 0(cid:105) to 1, +(cid:105), a second photon cannot subsequently drive the system from 1, +(cid:105) to 2, e3(cid:105) because the eigenenergy of 2, e3(cid:105) has shifted out of resonance. On the other hand, if the pump laser is tuned to half the energy of 2, e3(cid:105) (red arrows), it can no longer excite 1, +(cid:105), while at the same time, it can excite 2, e3(cid:105) via two-photon resonance. Thus, by mea- suring the photonic content of the state of the system, we can determine the strength of the nonlinearity. The photonic content, in turn, can be measured by detecting the light that leaks out of the cavity and an- alyzing its temporal distribution. The second-order co- herence function g(2)(τ ) yields the ratio of the detection rate of photon pairs separated by a delay τ to that of 3 (Color online) g(2)(0) vs. pump laser frequency for different U. (a) A 2D plot of g(2)(0) versus pump FIG. 2: laser frequency detuning (x-axis) for different values of U (y-axis). The color corresponds to the base-10 logarithm of g(2)(0). Four strong bunching peaks (red) are observed, three of which come from the second-manifold eigenstates. The remaining bunching peak at ωpump = 0 is due to photon-induced tunneling7. Also observed are three strong antibunching dips (blue): the first-manifold eigenstates (lower and upper polaritons) and a quantum-interference dip. The other parameters are ωe = ωc = 0 and κ = Γ = 0.01g. (b) Horizontal cross-sections of (a) for U/g = 0.3, 0.67, and 1.5. When U/g is near 2/3, the location of the quantum interference dip overlaps with that of the upper polariton at ωpump = g, yielding an extremely strong antibunching with g(2)(0) ∼ 10−7. (Color online) g(2)(0) vs pump laser frequency for different Γ and U. (a-c) Γ/g = 0.1, 0.5, and 1.0, FIG. 3: with U/g ranging from 0.1 to 0.5. The pump laser frequency is relative to the exciton resonance, and ωe = ωc = 0. (a) For small Γ, g(2)(0) resembles that in Fig. 2b, with the strong quantum interference-induced antibunching appearing near ωpump = g. (b) For intermediate Γ, the antibunching dip at ωpump = g becomes shallow while a new antibunching dip appears at a slightly negative ωpump. (c) This new antibunching dip, also due the destructive quantum interference, can be significantly large with g(2)(0) ∼ 10−2. single photons: III. PARAMETER STUDY OF g(2)(0) g(2)(τ ) = (cid:104)a†(0)a†(τ )a(τ )a(0)(cid:105) (cid:104)a†(0)a(0)(cid:105)2 (II.3) In particular, for zero time delay, g(2)(0) = 1 indicates a Poissonian distribution typical of classical light, whereas g(2)(0) < 1 is a sub-Poissonian distribution and an exper- imental smoking gun of a distinctly quantum process. In the following section, we will investigate g(2)(0) in various parameter spaces. We first consider g(2)(0) for Γ (cid:28) g (cid:28) U. We assume proach ±√ κ is equal to Γ. The second-manifold eigenenergies ap- 2g and 2U, the former pair resembling the well-known anharmonic Jaynes-Cummings ladder for a two-level qubit. The observation of photon antibunching dips (g(2)(0) < 1) at the polariton resonances as well as the bunching peaks (g(2)(0) > 1) at the energies of the two second-manifold states has been extensively explored in atomic31 and solid-state systems7. 4 (Color online) Minimum g(2)(0) for FIG. 4: different Γ and U. A 2D plot of the minimum value of the g(2)(0) that appears at negative ωpump (see Fig. 3) versus U (x-axis) and Γ (y-axis). The color represents the base-10 logarithm of g(2)(0). For a given value of Γ, strong antibunching is observed for a range of U. As Γ increases, the optimal value of U as well as its width increase. White dashed lines mark where g(2)(0) = 0.1. For this simulation, κ is set at 0.5g. The dotted appearance for strong antibunching is a numerical artifact. (Color online) g(2)(0) vs pump laser FIG. 5: frequency for different S. A plot of g(2)(0) versus pump laser frequency detuning for different monolayer area, with radius R ranging from 30 nm to 60 nm. The strong antibunching appears for R = 42 nm. (inset) The effect of R on the other parameters g, U, and Γ, and consequently g(2)(0), can be seen by plotting the parameters (black dotted line) on top of Fig. 4. As R is changed, the set of parameters cuts across the region of strong antibunching, making the system dynamics tunable. When U becomes comparable to g, there appears an- other energy, separate from the polaritons, that produces antibunching. As explained by Bamba et al.32, this anti- bunching dip is a result of destructive quantum interfer- ence between the first and the second manifolds, and its energy is given by 2ω(cid:48)3 + 2U ω(cid:48)2 + g2U = 0 (III.1) where ω(cid:48) = ω − i Γ 2 . Figure 2 shows a plot of g(2)(0) versus the pump laser frequency detuned from the cavity resonance at multiple values of U. In addition to the first and the second-manifold eigenenergies plotted in Fig. 1c, the interference-induced antibunching is clearly observed in Fig. 2a (the color represents the base-10 logarithm of g(2)(0)). As U increases, the interference dip passes through the upper polariton dip at ωpump = g. Figure. 2b shows the cross-sections of Fig. 2a for U/g = 0.3, 0.67, and 1.5. For U/g = 0.67 (shown in green), the interfer- ence dip coincides with the upper polariton dip, yielding an extremely strong antibunching (g(2)(0) ∼ 10−7) . Having explored Γ (cid:28) g ∼ U , we increase the dissi- pation in our system until it becomes comparable to g, which is more representative of typical solid-state envi- ronments. In Fig. 3, we explore three separate values of Γ/g: 0.1, 0.5, and 1.0. For each one, we plot g(2)(0) ver- sus the pump laser detuning for a range of U values. As Γ increases, previously sharp features become rounded, and what used to be a strong antibunching dip at ωpump = g becomes gradually shallower (Fig. 3a). For large Γ, on the other hand, an additional anti- bunching dip appears. As seen in Fig. 3b and c, this dip only appears for U < Γ, and the value of U at which it appears depends on how close Γ/g is to unity. The origin of this antibunching is once again the destructive quantum interference32, which has been extensively in- vestigated by Liew et al. in the context of "polariton boxes"33. For a given Γ, Eq. III.1 gives the optimum U and ω that produce the smallest g(2)(0). Figure 4 displays a two-dimensional color plot of min- imum g(2)(0) as a function of Γ and U. Here we set κ = 0.5g. The color represents the base-10 logarithm of g(2)(0), ranging from red (g(2)(0) ≈ 1) to blue (g(2)(0) ≈ 10−6). We have indicated on the plot with white dotted lines where g(2)(0) = 0.1, showing that the domain of U that produces strong antibunching increases with Γ. IV. PROPOSED EXPERIMENTAL DESIGN To observe the strong, interference-induced antibunch- ing, we propose to pattern a 2D-material monolayer into a circular island with radius R and place it on a thin pho- tonic crystal cavity (see Fig. 1a). We assume that the 5 While the cavity loss for a typical nanobeam is fixed (κ = 2π× 150 GHz)28, the exact dependence of Γ on R is unknown and remains an open problem. It has been re- ported that patterned monolayers on the order of tens of nanometers in radii can suffer from linewidth broadening due to the presence of edge states. Since the length of the edge scales linearly with R and the loss has been seen to increase for smaller monolayers, for our simulations, we have chosen to fix Γ = 2π × 300 GHz at R = 50 nm, an experimentally measured value, and vary it as 1/R28. Figure 5 shows the effect of changing R on g(2)(0). As R increases from 30 nm to 60 nm, an antibunching dip ap- pears, becomes sharper, and then recedes. The strongest antibunching occurs at R = 42 nm. The inset shows how the appearance of the dip compares to the general an- tibunching behavior in Fig. 4. The black dotted line, representing changing R, cuts across the region of strong antibunching, exhibiting the system's tunability. Finally, we explore the robustness of the antibunching dip for unequal cavity and exciton detunings, i.e., ωc (cid:54)= ωe. Figure 6 shows a plot of g(2)(0) as a function of ωc and ωe for the optimal paramters (R = 42 nm, g = 2π × 560 GHz, Γ = 2π× 360 GHz, κ = 2π× 150 GHz, U = 2π× 40 GHz), where the color represents the base-10 logarithm of g(2)(0). While the antibunching behavior is observed only for a narrow range of the exciton detuning (x-axis), it survives for a much larger range of the cavity detuning (y-axis), giving us substantial leeway in the fabrication precision of the nanobeam cavity. V. CONCLUSION We have explored the second-order coherence of a 2D- material monolayer embedded in a photonic crystal cav- ity and identified a range of system parameters that yield strong photon antibunching. We have shown that by pat- terning the monolayer into different sizes, we can tune the system dynamics, driving it from a weak to a strong pho- ton antibunching regime. The successful implementation of the experimental design will open the door to a new regime of quantum interference-based quantum simula- tions on a scalable, on-chip platform. VI. ACKNOWLEDGEMENTS This work was supported by the National Science Foundation under grants NSF-EFRI-1433496 and NSF- 1708579 and the Air Force Office of Scientific Research- Young Investigator Program under grant FA9550-15-1- 0150. (Color online) g(2)(0) vs ∆e vs ∆c. A 2D plot FIG. 6: of g(2)(0) versus exciton detuning (x-axis) and cavity detuning (y-axis) for monolayer radius R = 42 nm. The color represents the base-10 logarithm of g(2)(0). Clearly, the variance is much greater for the exciton detuning compared to that for the cavity detuning. area of the patterned monolayer is much smaller than that of the cavity mode, i.e., R (cid:28) Rmode. We also as- sume that the monolayer is free of any defect such that the excitons are delocalized over the entire monolayer area. Hence, the spatial extent of the exciton wavefunc- tion is equal to the physical size of the monolayer. Both the exciton-photon coupling g and the nonlinear- ity U depend on the size of the monolayer. The former is given by34 (cid:115) dcvφ(0)√ωc √ g = πR2 πR2 mode (IV.1) 2
0Lc φ(0) = (cid:112)2/(πaB)2 is the amplitude of the exciton is the interband dipole matrix element, where dcv wavefunction (aB is the exciton Bohr radius), ωc is the cavity resonance frequency,
0 is the permittivity of free space, and Lc is the effective length of the cav- ity mode. The nonlinear interaction strength is given by U = 6Eba2 B/(πR2), where Eb is the exciton binding energy? . Thus, g ∼ R and U ∼ 1/R2, allowing us to tune the system dynamics by patterning the monolayer into dif- ferent areas via, for instance, electron beam lithography. For a WSe2 monolayer with R = 5 nm coupled to a SiN nanobeam cavity with Rmode = 1 µm, g ≈ 2π × 700 GHz and U ≈ 2π × 30 GHz34. 1 A. Kiraz, M. Atature, and A. Imamoglu, Phys. Rev. A 2 A. Kuhn, M. Hennrich, and G. Rempe, Phys. Rev. Lett. 69, 032305 (2004). 89, 067901 (2002). 6 3 G. J. Milburn, Phys. Rev. Lett. 62, 2124 (1989). 4 W. S. Bakr, J. I. Gillen, A. Peng, S. Folling, 20 T. Fryett, A. Zhan, and A. Majumdar, Nanophotonics and (2017). M. Greiner, Nature (London) 462, 74 (2009). 5 M. H. Devoret and R. J. Schoelkopf, Science 339, 1169 (2013). 6 A. Majumdar, M. Bajcsy, and J. Vuckovi´c, Phys. Rev. A 85, 041801 (2012). 7 A. Faraon, I. Fushman, D. Englund, N. Stoltz, P. Petroff, and J. Vuckovi´c, Nat. Phys. 4, 859 (2008). 8 A. Reinhard, T. Volz, M. Winger, A. Badolato, K. J. Hen- nessy, E. L. Hu, and A. Imamoglu, Nat. Photonics 6, 93 (2012). 9 V. Negoita, D. W. Snoke, and K. Eberl, Applied Physics Letters 75, 2059 (1999). 10 M. J. Hartmann, Journal of Optics 18, 104005 (2016). 11 N. Y. Kim and Y. Yamamoto, "Exciton-polariton quan- tum simulators," in Quantum Simulations with Photons and Polaritons: Merging Quantum Optics with Condensed Matter Physics, edited by D. G. Angelakis (Springer In- ternational Publishing, Cham, 2017) pp. 91–121. 12 I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85, 299 (2013). 13 I. Carusotto, D. Gerace, H. E. Tureci, S. De Liberato, C. Ciuti, and A. Imamoglu, Phys. Rev. Lett. 103, 033601 (2009). 14 H. Deng, H. Haug, and Y. Yamamoto, Rev. Mod. Phys. 82, 1489 (2010). 15 D. Sanvitto and S. K´ena-Cohen, Nat. Mater. 15, 1061 (2016). 16 B. Besga, C. Vaneph, J. Reichel, J. Est`eve, A. Reinhard, J. Miguel-S´anchez, A. Imamoglu, and T. Volz, Phys. Rev. Applied 3, 014008 (2015). 17 G. Munoz-Matutano, A. Wood, M. Johnson, X. V. Asen- sio, B. Baragiola, A. Reinhard, A. Lemaitre, J. Bloch, A. Amo, B. Besga, et al., arXiv preprint arXiv:1712.05551 (2017). 18 V. Verma and J. Coleman, Applied Physics Letters 93, 111117 (2008). 21 Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman, and M. S. Strano, Nat. Nanotechnol. 7, 699 (2012). 22 S. Wu, S. Buckley, J. R. Schaibley, L. Feng, J. Yan, D. G. Mandrus, F. Hatami, W. Yao, J. Vuckovic, A. Majumdar, and X. Xu, Nature (London) 520, 69 (2015). 23 Y. Ye, Z. J. Wong, X. Lu, X. Ni, H. Zhu, X. Chen, Y. Wang, and X. Zhang, Nat. Photonics 9, 733 (2015). 24 C.-H. Liu, G. Clark, T. Fryett, S. Wu, J. Zheng, F. Hatami, X. Xu, and A. Majumdar, Nano Letters 17, 200 (2017), pMID: 27936763. 25 X. Liu, T. Galfsky, Z. Sun, F. Xia, E. chen Lin, Y.-H. Lee, S. Kena-Cohen, and V. M. Menon, 9, 30 (2014). 26 S. Dufferwiel, S. Schwarz, F. Withers, A. A. P. Trichet, F. Li, M. Sich, O. D. Pozo-Zamudio, C. Clark, A. Nalitov, D. D. Solnyshkov, G. Malpuech, K. S. Novoselov, J. M. Smith, M. S. Skolnick, D. N. Krizhanovskii, and A. I. Tar- takovskii, Nat. Commun. 6 (2015), 10.1038/ncomms9579. 27 G. Wei, D. A. Czaplewski, E. J. Lenferink, T. K. Stanev, and N. P. Stern, Sci. Rep. 7 (2017), I. W. Jung, 10.1038/s41598-017-03594-z. 28 T. K. Fryett, Y. Chen, J. Whitehead, Z. M. Peycke, X. Xu, and A. Majumdar, arXiv preprint arXiv:1709.02032 (2017). 29 Y. Sun, Y. Yoon, M. Steger, G. Liu, L. N. Pfeiffer, K. West, and K. A. Nelson, Nat. Phys. 13, 870 EP D. Snoke, (2017), article. 30 A. Kavokin, J. Baumberg, G. Malpuech, and F. Laussy, Microcavities, Series on Semiconductor Science and Tech- nology (Oxford University Press New York, 2007). 31 K. M. Birnbaum, A. Boca, R. Miller, A. D. Boozer, T. E. and H. J. Kimble, Nature (London) 436, 87 Northup, (2005). 32 M. Bamba, A. Imamoglu, I. Carusotto, and C. Ciuti, Phys. Rev. A 83, 021802 (2011). 33 T. C. H. Liew and V. Savona, Phys. Rev. Lett. 104, 183601 19 L. Lee, L. Zhang, H. Deng, and P.-C. Ku, Applied Physics (2010). Letters 99, 263105 (2011). 34 H.-X. Wang, A. Zhan, Y.-D. Xu, H.-Y. Chen, W.-L. You, A. Majumdar, and J.-H. Jiang, J. Phys. Condens. Matter 29, 445703 (2017). 

1807.01728

3

1807

Nonlinear Chiral Transport in Dirac Semimetals

We study the current of chiral charge density in a Dirac semimetal with two Dirac points in momentum space, subjected to an externally applied time dependent electric field and in the presence of a magnetic field. Based on the kinetic equation approach, we find contributions to the chiral charge current, that are proportional to the second power of the electric field and to the first and second powers of the magnetic field, describing the interplay of the chiral anomaly and the drift motion of electrons moving under the action of electric and magnetic fields.

cond-mat.mes-hall

cond-mat

a Nonlinear Chiral Transport in Dirac Semimetals Alexander A. Zyuzin,1, 2 Mihail Silaev,3 and Vladimir A. Zyuzin4 1Department of Applied Physics, Aalto University, P. O. Box 15100, FI-00076 AALTO, Finland 2Ioffe Physical -- Technical Institute, 194021 St. Petersburg, Russia 3Department of Physics and Nanoscience Center, University of Jyvaskyla, P.O. Box 35 (YFL), FI-40014 University of Jyvaskyla, Finland 4Department of Physics and Astronomy, Texas A&M University, College Station, Texas 77843-4242, USA We study the current of chiral charge density in a Dirac semimetal with two Dirac points in momentum space, subjected to an externally applied time dependent electric field and in the presence of a magnetic field. Based on the kinetic equation approach, we find contributions to the chiral charge current, that are proportional to the second power of the electric field and to the first and second powers of the magnetic field, describing the interplay of the chiral anomaly and the drift motion of electrons moving under the action of electric and magnetic fields. I. INTRODUCTION The Weyl and Dirac semimetals are recently discov- ered materials, whose conduction and valence bands with linear energy dispersion touch at a number of Weyl or Dirac points in the Brillouin zone [1 -- 6]. These systems belong to the Fermi point universality class of fermionic vacua [2] and possess nontrivial topology of the electronic band structure. The non degenerate Weyl point might be described as a monopole sink or source of the Berry curvature and assigned with a topological charge, an in- tegral of the Berry curvature over the surface enclosing the point. Since the net topological charge is zero, Weyl points always appear in pairs of opposite charge. The Dirac point might be composed of two Weyl points with topological charges of opposite sign. In certain classes of three-dimensional semimetals such Dirac points occur in pairs separated along a rotation axis of the crystal pro- vided both time-reversal and inversion symmetries are not broken [7 -- 10]. One of the distinct properties of Weyl and Dirac semimetals is the chiral anomaly, which is a non- conservation of chiral charge induced by the externally applied parallel electric and magnetic fields [11, 12]. The presence of the chiral charge imbalance leads to a num- ber of phenomena such as for example the chiral mag- netic effect - charge current driven along the magnetic field [13], chiral electric separation effect - the flow of chiral charge imbalance along the electric field [14], the quantum and classical negative magnetoresistance [15 -- 20], and contributions to the nonlinear optical response [21 -- 28]. Another anomalous transport phenomena, al- though unrelated to the chiral anomaly, is the chiral sep- aration effect, which describes the flow of fermions with opposite chiral charges in opposite directions along with the external magnetic field [29, 30]. The progress in the topological semimetals is reviewed in Ref. [31]. Recently, a question of the interplay of the chiral anomaly and the nonlinear chiral transport was ad- dressed for a ferromagnetic Weyl semimetal [32]. Based on the kinetic equation approach [33, 34], it was shown that the chiral anomaly might lead to quadratic in elec- tric field corrections to the chiral charge current. Here, we study the chiral charge current driven by a time-dependent electric field in the presence of a mag- netic field in the Dirac semimetal, with a pair of Dirac points in it's band structure. Besides the chiral charge imbalance, the chiral anomaly generates a spin imbalance in each Dirac valley, such that the total spin polarization in the system is zero, although the staggered spin polar- ization is induced. We show that the chiral charge cur- rent as well as the current of staggered spin polarization is proportional to the second power of the electric field and is described by joint action of the chiral anomaly and the electron motion in the presence of the electric and magnetic fields. II. MODEL Let us consider a model of the inversion and time re- versal symmetric gapless Dirac semimetal with two Dirac points separated in momentum space on the crystal rota- tion axis (one might have in mind Cd3As2 and Na3Bi as particular material candidates). The system is described by the Hamiltonian H(k) = v(σxszkx − σyky) + m(kz)σz + δH(k), (1) −+s−k2 where m(kz) = m1k2 z − m0, in which m0m1 > 0, and σ and s are the vectors composed of the three Pauli matrices denoting the pseudo-spin and spin degrees of freedom (we set  = 1). The Hamiltonian δH(k) = +) ∝ O(k3) is a small correction, which γσxkz(s+k2 is off-diagonal in spin space. Two Dirac points are sepa- rated by a distance 2pm0/m1 along z-axis in momentum space. Provided δH(k) = 0 the Hamiltonian in Eq. 1 is block diagonal and one can introduce a sign, s = ±, to label the eigenvalues of sz. To proceed, we consider a spherical Fermi surface, set 2√m0m1 ≡ v, and linearize the Hamiltonian around each Dirac point as Hη,s(k) = v(sσxkx − σyky + ησzkz), where the momentum in each valley is now measured relatively to the corresponding point, which is labeled by η = ±, as kz → kz − ηpm0/m1. We note that each Dirac point is composed of two Weyl points of opposite topological charge, which are related by the time reversal symmetry and determined by the spin eigenvalues. The Berry curvature for each of four Weyl points is given by Ωη,s = ηsk/2k2, where k = k/k is the unit vector in the direction of momentum. In the absence of the spin-flip processes, the sz- component of the spin is conserved, allowing one to in- troduce the topological charge for the spin-up and spin- where the integral is taken over the surface S enclos- ing the Weyl node. While the total topological charge down electrons Cη,+ − Cη,−, with Cη,s =RS dS· Ωη,s/2π, Pη(Cη,+ + Cη,−) is zero, the staggered spin charge is finite, Pη η(Cη,+ − Cη,−)/2 = 2; for a more detailed dis- cussion of Z2 topological charge in the Dirac semimetals, see Refs. [7 -- 10]. In the situation where a magnetic field is applied to the semimetal, one naturally expects the chiral separation ef- fect. Turning on an electric field in addition to the mag- netic field gives rise to a chiral anomaly with pronounced nonlinear corrections to the chiral charge current. This is in contrast to the chiral electric separation effect studied in Refs. [14, 29, 30, 35], being linear in powers of electric field. A. Kinetic equation Having established the model of the Dirac semimetal, let us analyze the chiral charge current within the chiral kinetic equation approach focusing on the zero tempera- ture limit. This approach has been described extensively in the literature and here we briefly outline the key points [33, 34, 36 -- 38]. We assume a spatially homogeneous time-dependent electric field E(t) = E0(ω)e−iωt + E∗ 0(ω)eiωt (2) and a magnetic field B applied to the system (we will comment on the effect of the wave-vector dependence of the electromagnetic field later in the conclusions). We consider the case of electron doped semimetal, in which the chemical potential is in the conduction band µ > 0, neglect the Zeeman effect of a magnetic field compared to its orbital effect, and focus on the response, which is quadratic in powers of electric field. e The kinetic equation for the distribution function fη,s(t, k) of the wave-packet with energy εη,s = ε(1 − c B · Ωη,s), where ε = vk, reads ∂fη,s ∂k = I[fη,s]. + k · ∂fη,s (3) ∂t The electric and magnetic field dependent higher order corrections to the energy and to the Berry curvature do not change the result and will be neglected [39]. The ki- netic equation is supplemented by the solutions of equa- tions of motion, which contain contributions from the Berry curvature and orbital magnetic moment 2 1 c r = D−1 k = eD−1 η,s(cid:26)E + [vη,s × B] + η,snvη,s + e[E × Ωη,s] + (E · B)Ωη,s(cid:27) , (vη,s · Ωη,s)Bo , (4b) where vη,s = ∂εη,s/∂k is the wave-packet velocity, I[fη,s] is the collision integral, Dη,s = 1 + e c (B· Ωη,s), and e < 0. e c e c (4a) B. Collision integral The chiral charge is not strictly conserved when terms nonlinear in momentum and spin-flip scattering processes are included in the Hamiltonian [40]. For the collision in- tegral in Eq. 3, we assume that the inter-valley scatter- ing rate is exponentially suppressed with respect to the intra-valley scattering rate. We then note that the Hamil- tonian in Eq. 1 is block-diagonal in spin-space and the z-component of the particle's spin is a conserved quan- tity provided δH(k) is neglected. Turning on the spin- flip processes, we adopt a model in which the intra-valley spin-flip relaxation time is much longer than the intra- valley spin-conserving relaxation time. We also assume the magnetic length v/√ωcµ, where ωc = −ev2B/cµ is the cyclotron frequency, to be much larger than the cor- relation radius of the scattering potential. Hence the spin-flip and valley-flip relaxation times can be consid- ered magnetic field independent [41]. These assumptions allow us to simplify the collision integral and separate the intra-valley spin-conserving contribution I[fη,s] = hfη,si − fη,s(t, k) + Λ[fη,s] + Iin[fη,s], (5) where Iin[fη,s] describes the energy relaxation processes, the valley-flip and spin-flip elastic scattering processes are described by the functional τ (ε) Λ[fη,s] ≡ hf−η,s − fη,si + hfη,−s − fη,si τV (ε) , + hf−η,−s − fη,si τ ′ V (ε) in which τV (ε) and τ ′ conserving and spin-flip scattering times, τ ′ intra-valley spin-flip scattering time, and τ ′ 0(ε) V (ε) are the inter-valley spin- 0(ε) is the (6) 1/τ (ε) = 1/τ0(ε) + 1/τV (ε) + 1/τ ′ 0(ε) + 1/τ ′ V (ε) (7) is the momentum relaxation rate of a particle with en- ergy ε, in which τ0(ε) describes the intra-valley spin- conserving scattering processes. We consider the hierar- chy of the elastic scattering times τ0(ε) < (τ ′ 0(ε), τV (ε)) < τ ′ V (ε), where the inter-valley spin-flip length vτ ′ V (µ) is assumed to be smaller than the system size. Different inter-node relaxation processes in the Dirac semimetal were studied in detail in Ref. [42]. The triangle brackets h...i mean integration over the directions of momentum, taking into account the change scatterings Λ[fη,s] field [33, 34], such that hfη,si ≡ R dΘ of the phase space in the presence of the magnetic 4π Dη,s(k)fη,s(t, k). The term describing the valley and spin flip elas- tic and can be related to the distribution function as satisfies Pη,s Λ[fη,s] = 0 Pη,s ηs{Λ [fη,s] + hfη,si/τf (ε)} = 0, where 1/2τf (ε) ≡ 1/τV (ε)+1/τ ′ 0(ε) is an effective relaxation rate, which in- cludes inter-valley spin-conserving and intra-valley spin- flip scattering processes. III. SOLUTION OF KINETIC EQUATION Let us now study the chiral charge response of the Dirac semimetal taking into account the inter and in- tra valley scattering processes. To reveal the topolog- ically nontrivial contributions we consider the limit of weak magnetic field, in which the cyclotron frequency is smaller than the electron momentum relaxation rate. We search for the approximate solution of Eq. 3, keep- ing contributions to the distribution function up to sec- ond order of the electric field. Similarly to Ref. [43], we the expand distribution function in powers of the incident electric field, fη,s(t, k) = f (0) η,s (ε) + + f (2) η,s (k) + 1 1 2hf (1) 2hf (2) η,s (ω, k)e−iωt + c.c.i η,s (2ω, k)e−2iωt + c.c.i , (8) 3 η,s (ε) ≈ f (0)(ε) − e where f (0) c (B · Ωη,s)ε∂εf (0)(ε) and f (0)(ε) = θ(µ − ε) is the distribution function at zero temperature, f (1) η,s (ω, k) is the first order correction, and η,s (k), f (2) f (2) η,s (2ω, k) are the second order corrections at the zeroth and double frequencies, respectively. Follow- ing Perel' and Pinskii [44], we set an additional constrain on the second order correction Xη,s Z d3kDη,sf (2) η,s = 0, (9) meaning that the second order solution does not change the concentration of particles. The kinetic equation for the correction to the distribu- tion function in n-th (n > 0) power of the electric field is given by (cid:18)inω − e c D−1 η,s[vη,s × B] · ∂ ∂k(cid:19) f (n) (E0 · B)Ωη,so · e c η,s + I[f (n) η,s ] ∂f (n−1) η,s ∂k . = eD−1 η,snE0 + (10) Physically, the solution to the equation linear in elec- tric field [with n = 1 in Eq. 10] describes the elastic scattering in the system, while the nonlinear solution should also account for the energy relaxation. Hence, neglecting the inelastic scattering in the collision inte- gral, the solution of the first order differential equation 10 for f (1) η,s (ω, k) is given by f (1) η,s (ω, k) = τ (ε) 1 − iωτ (ε)(cid:20)hf (1) η,si τ (ε) + Λ[f (1) η,s ] − ev Dη,s(cid:26) ηse 2ck2 (E0 · B) + E0 · k − κη,s [E0 × B] + κη,sE0 1 + κ2 η,s · k⊥(cid:27)∂εf (0) η,s(cid:21), (11) τ (ε) ε ωcDη,s where parameter κη,s(ε, ω) = µ 1−iωτ (ε) is intro- duced for brevity, k⊥ is the unit vector in the direction of momentum lying in the plane transverse to the direc- tion of magnetic field B = B/B. The last term in Eq. 11 absorbs the contribution from the cyclotron part of the Lorentz force. It is worth to note that the cyclotron frequency in κη,s(ε, ω) is renormalized with the Berry curvature. Multiplying Eq. 11 with Dη,s and integrating over the directions of momentum, one obtains an equation for hf (1) η,si in the form −iωhf (1) e2v 2ck2 (E0 · B)∂εf (0) η,si = Λ[f (1) η,s ] − ηs e2v 6ck2 (E0 · B)h∂εf (0) + ε∂2 + ηs ε f (0)i , (12) where the second line describes the contribution of the orbital magnetic moment. We then find that the relax- ation of the chiral imbalance ηshf (1) η,si = − 2e2v 3ck2 τf (ε)(E0 · B) 1 − iωτf (ε) Xη,s [2∂εf (0) − ε∂2 ε f (0)], (13) comes from the inter-valley spin-conserving and intra- valley spin-flip scattering processes and describes the emergence of non-equilibrium chiral charge as well as staggered spin accumulations. This quantity is as- sociated with the chiral anomaly. The staggered spin polarization in the Dirac semimetal, P(t) = N −1Pη,s ηsR (dk)Dη,sfη,s(t, k), where N = 2µ3/3π2v3 is the electron density, in the lowest order in the electric and magnetic fields P(ω) = 3ωc is determined by the (E0 · B) product. Finally, the sumPη,s f (1) satisfies Pη,shf (1) η,s gives the standard expression for the first-order solution, which The exact solution in the second order is rather cum- bersome. However, we are interested in the limit of weak magnetic field ωc < ω and keep up to quadratic in pow- η,si =0. (E0 · B) 1−iωτf evτf 2µ µ ers of ωcτ corrections to the distribution function. The solution can be formally written as e c (14) τ (ε) τ (ε)Dη,s f (2) η,s (2ω, k) = + Iin[f (2) η,s ] v[k × B] 1 − 2iωτ (ε) · 1 − 2iωτ (ε)(cid:26)hf (2) η,si τ (ε) η,s 1 − − eD−1 ×hE0 + ∂k! η,s (ω, k)(cid:27), η,si ∝ ω2 where we neglect Λ[f (2) µ2 (E0 · B)2 ≪ 1 is beyond the validity of our assumptions. The difference between the momentum relaxation rates of the first and second harmonics is also neglected. (E0 · B)Ωη,si · η,s ], since Pη,s ηshf (2) ∂ ∂k f (1) e c ∂ c To determine the contribution to hf (2) 0 one has to take into account the inelastic processes [44] and apply a condition of a constant concentration of particles in the presence of the electric field. For the collision integral, describing the inelastic processes, we consider the simplest form η,s (2ω)i ∝ E2 Iin[f (2) η,s ] = − f (2) η,s τin , (15) where τin is the energy relaxation time of the electron due to coupling to some thermal bath. Taking into account only the terms that give dominant contribution to the chiral charge current, we find hf (2) η,s (2ω)i = e2v2 3 1 − 2iωτin (cid:20)∂ε τin(E0 · E0) τ (ε) ∂εf (0) + τ (ε) 1 − iωτ (ε) (∂2 1 − iωτ (ε) ε f (0) + 2ε−1∂εf (0))(cid:21). k ·(cid:26)E0 η,s (ω, k)i, evτ (ε) (16) (17) We then obtain the second order correction in the form f (2) η,s (2ω, k) = hf (2) η,s (2ω)i − 1 − 2iωτ (ε) 1 − 2iωτ (ε) (cid:27)∂εhf (1) ev τ (ε)[E0 × B] ck + where we substitute f (1) η,s (ω, k) with a solution of Eq. 12. The expression for f (2) η,s (k) can be found from Eq. 17 with the formal substitutions ω = 0, E · E → E2, and E0hf (1) η,s (ω, k)i. We are now in the position to calculate the chiral charge current density. η,s (ω, k)i → ReE∗ 0hf (1) 4 The first term gives a finite contribution provided the staggered spin polarization is induced. The second and third terms are the corrections to the motion of the wave- packet due to the nontrivial Berry curvature and describe the chiral separation and inverse Faraday effects, respec- tively. The terms on the second line describe corrections from the orbital magnetic moment of the wave-packet. We first consider the chiral transport in the collision- less limit and neglect I[fη,s] in Eq. 3. The chiral charge current density can be expanded in powers of the electric field amplitude j5(t) = e2µ π2c B +(cid:2)j5(2ω)e−2iωt + c.c.(cid:3) /2, (19) where the first term describes the chiral separation effect (for a review see Ref. [14]). It is a non-dissipative chi- ral current, which exists in the equilibrium state of chiral fermions provided a magnetic field is applied [29, 30], and doesn't depend on the orbital magnetic moment. The other terms describe second-order correction at the dou- ble frequency of the electric field. In the collisionless limit at frequencies ω ≫ ωc, we obtain j5(2ω) = − e3ωc 6π2ω2 [E0 × [ B × E0]]. (20) We find that j5(2ω) vanishes for the parallel orientation of electric and magnetic fields (the field dependence in Eq. 20 matches the one derived for the strain induced non-equilibrium spin current in the Dirac semimetal [45]) and does not depend on the orbital magnetic moment. Although, turing on the scattering processes, different inter-valley and intra-valley relaxation rates give rise to a finite chiral current for parallel orientation of the fields. At ωτ < 1, where τ ≡ τ (µ) is the momentum re- laxation rate at the Fermi energy, we find a linear in magnetic field contribution to the chiral charge current density in the form j5(2ω) = 2e3 3π2 ωcτ 1 − iωτ(cid:26) τin 1 − 2iωτin τf − G(ω) 1 − 2iωτ 1 − iωτf E0(E0 · B)(cid:27), (E0 · E0) B (21) where the model dependent coefficient G(ω) is deter- mined by the scattering potential ε2τ (ε) 1 − 2iωτ (ε)(cid:12)(cid:12)(cid:12)(cid:12)ε=µ . (22) In the model of intra-valley short range potential, where τ0(ε) ∝ ε−2, assuming other scattering times to be energy independent, one estimates G(0) ∝ 1 − τ /τ0. Although, for the model of the Coulomb impurities τ0(ε) ∝ ε2 one gets G(0) = 2. The first term in Eq. 21 describes the ac-field ∝ E2 0 induced change of the electron distribution function, be- ing sensitive to the inelastic relaxation time. Similarly IV. NONLINEAR CHIRAL CHARGE CURRENT G(ω) = (1 − 2iωτ ) 1 2µτ ∂ε Let us define the chiral charge current density in the (2π)3 Dη,s r(t, k)fη,s(t, k), where explicitly semimetal j5(t) = ePη,s ηsR d3k (2π)3(cid:26)ηsvk + j5(t) = eXη,s Z d3k + ev ck2 k(B · k) − evB 2ck2 + 2ck2(cid:27)fη,s(t, k). evB e 2k2 [E(t) × k] (18) to the chiral separation effect, this contribution to the chiral current is parallel to the magnetic field. The sec- ond term in Eq. 21 describes the nonlinear chiral electric separation effect, being more pronounced in the case of collinear electric and magnetic fields. It is smaller than the first term at ω → 0, provided τin ≫ τf . It is also worth noting that the signs of two terms in Eq. 21 are opposite. This describes the suppression of the chiral current in the case of collinear fields being stronger at frequencies much larger than the relaxation rates. At ωτ < 1 we also find a correction to the chiral current from the interplay of the chiral anomaly and the Hall effect j5,Hall(2ω) ∝ e3 (ωcτ )2 1 − iωτ τf (E0 · B) 1 − iωτf [ B × E0] (1 − 2iωτ )2 . (23) This correction is smaller than the drift contribution given in Eq. 21 provided ωcτ < 1. The physical meaning of the contributions to j5(2ω) can be understood if we consider a two-step process. The first step contains long inter-valley scattering, which equilibrates the spin imbalance generated due to chiral anomaly ∝ (E0 · B). The second step is the drift motion of particles within the valleys under the joint action of the electric and magnetic fields, described by the Lorentz force eE + e c [v × B]. This is in contrast to the chiral anomaly generated charge current, which is determined by the inter-valley scattering processes [16, 17]. The contributions to the axial current in chiral plasma produced by time-dependent electric and magnetic fields and gradients of the chemical potentials were also con- [35]. While the first term in Eq. 21 sidered in the Ref. is new, the second term and the Hall contribution given in Eq. 23 are compatible with the results of Ref. [35], provided the chiral chemical potential imbalance is field induced and proportional to τf (E · B). V. DISCUSSION AND CONCLUSIONS So far, we have studied the effect of spatially homoge- neous electric field. Briefly, we would like to comment on the case when the electromagnetic wave has a spatial de- pendence E(t, r) = E0 exp(−iωt+iq·r)+c.c. In this case, a finite, linear in electric field contribution to the chiral charge current is allowed. In order to evaluate it, the spa- tial derivatives in the chiral kinetic equation have to be taken into account, see, for example, Ref. [38]. Indeed, at ω = 0 and using τ ≪ τf , which determines the sit- uation where the chiral anomaly is the dominant source [1] A. A. Abrikosov and S. D. Beneslavskii, "Possible Ex- istence of Substances Intermediate Between Metals and Dielectrics," J. Exp. Theor. Phys. 32, 699 (1971). 5 to the chiral current, we find that the linear response contribution is given by j5(ω, q) ∝ −ie2ωcτ µτf q(E0 · B). Naturally, it is proportional to the wave-vector q of the field, such that the chiral current remains invariant with respect to a spatial inversion. 3π 2 E2 Let us now discuss the experimental feasibility of the proposed effect. The amplitude of the first term in Eq. 21 α can be written at zero frequency as j5 = − 16e  Iωcτ τin, where I = cǫ0 0, α = e2/4πǫ0c is the fine structure constant, and ǫ0 is the permittivity of free space (we have restored  here). Taking numerical values consistent with experiment Ref. [46], µ ∼ 220 meV, v ∼ 9.3 × 107 cm/s, and τ ∼ 0.5 ps, we estimate ωc ∼ 1 ps−1 at B = 0.2T, which satisfies ωcτ < 1. Although the values of the in- elastic relaxation time is not known, we might roughly estimate it as τin ∼ 10τ . We obtain j5 ∼ 50I[ A/cm2 W/cm2 ], which might be of the order of the quantized circular photogalvanic effect in a Weyl semimetal studied in Ref. [27]. The chiral charge current in a Dirac semimetal could be probed indirectly via the interplay between the elec- tric and chiral charge currents, which gives rise to the chiral magnetic waves [38]. Although in ferromagnetic Weyl semimetals, where the fields induce a finite spin polarization, probing the chiral charge current might be straightforward via nonlocal measurements similar to the spin polarization in metals [47, 48] (a similar idea pro- posed for the Weyl and Dirac semimetals was discussed in Ref. [42]). However, one needs to be able to extract it from the total signal, which also contains large but electric field and frequency independent contribution, de- scribed by the chiral separation effect. To conclude, we have calculated the nonlinear in the electric field corrections to the chiral charge current in the Dirac semimetal. These are proportional to the second power of the externally applied electric field and consist of contributions that are proportional to the first and second powers of the magnetic field. We have also com- mented on the chiral anomaly generated staggered spin accumulation, i.e. the nonequilibrium spin polarization in each Dirac valley of the semimetal, with vanishing net spin polarization. ACKNOWLEDGMENTS A.Z. and M.S. are supported by the Academy of Fin- land. tions Lect. Notes Phys. 718, 31 (2007). from topology in momentum space," [2] G. E. Volovik, "Quantum phase transi- quantum spin Hall and insulator phases [3] S. Murakami, "Phase transition between the in gapless phase," semimetals," Phys. Rev. B 94, 245121 (2016). 6 [11] S. L. Adler, "Axial-Vector Vertex in Spinor Electrody- [32] V. A. Zyuzin, "Chiral electric separation effect in Weyl → γγ in [33] D. Xiao, M.-C. Chang, semimetals," Arxiv: 1803.00723. "Berry Rev. Mod. Phys. 82, 1959 (2010). effects phase on and Q. Niu, properties," electronic emergence 3D: topological New Journal of Physics 9, 356 (2007). of a [4] X. Wan, A. M. Turner, A. Vishwanath, and S. Y. Savrasov, "Topological semimetal and Fermi-arc surface states in the electronic structure of pyrochlore iridates," Phys. Rev. B 83, 205101 (2011). [5] K.-Y. Yang, Y.-M. Lu, and Y. Ran, "Quantum Hall effects in a Weyl semimetal: Possible application in py- rochlore iridates," Phys. Rev. B 84, 075129 (2011). [6] A. A. Burkov and L. Balents, "Weyl Semimetal Multilayer," in Phys. Rev. Lett. 107, 127205 (2011). Topological Insulator a [7] T. Morimoto and A. Furusaki, "Weyl Dirac semimetals with Z2 Phys. Rev. B 89, 235127 (2014). topological and charge," [8] B.-J. Yang and N. Nagaosa, "Classification of sta- ble three-dimensional Dirac semimetals with nontrivial topology," Nat. Com. 5, 4898 (2014). [9] E. V. Gorbar, V. A. Miransky, I. A. Shovkovy, and P. O. Sukhachov, "Dirac semimetals A3Bi (A = Na, K, Rb) as Z2 Weyl semimetals," Phys. Rev. B 91, 121101 (2015). [10] B.-J. Yang, T. Morimoto, and A. Furusaki, "Topolog- ical charges of three-dimensional Dirac semimetals with rotation symmetry," Phys. Rev. B 92, 165120 (2015). namics," Phys. Rev. 177, 2426 (1969). [12] J. S. Bell and R. Jackiw, "A PCAC Puzzle: π0 the σ-Model," Nuovo Cimento A 60, 47 (1969). [13] K. H. Phys. Rev. D 78, 074033 (2008). Fukushima, J. Warringa, D. E. Kharzeev, "Chiral magnetic and effect," [14] D. E. Kharzeev, J. Liao, S. A. Voloshine, and G. Wang, "Chiral magnetic and vortical effects in high-energy nuclear collisions -- A status report," Progr. Part. Nucl. Phys. 88, 1 (2016). [15] H. B. Nielsen and M. Ninomiya, "The Adler-Bell- in a crystal," fermions Jackiw anomaly and Weyl Phys. Lett. B 130, 389 (1983). [16] D. T. Son and B. Z. Spivak, "Chiral anomaly and classical negative magnetoresistance of Weyl metals," Phys. Rev. B 88, 104412 (2013). [17] A. A. Burkov, "Chiral Anomaly fusive Phys. Rev. Lett. 113, 247203 (2014). Magnetotransport in Weyl and Dif- Metals," [18] Q. Li, D. E. Kharzeev, C. Zhang, Y. Huang, I. Pletikosi´c, A. V. Fedorov, R. D. Zhong, J. A. Schneeloch, G. D. Gu, and T. Valla, "Chiral magnetic effect in ZrTe5," Nature Physics 12, 550 (2016). [19] V. A. Zyuzin, "Magnetotransport of Weyl semimetals due to the chiral anomaly," Phys. Rev. B 95, 245128 (2017). [20] B. Z. Spivak and A. V. Andreev, "Magnetotransport phe- nomena related to the chiral anomaly in Weyl semimet- als," Phys. Rev. B 93, 085107 (2016). [21] L. Wu, S. Patankar, T. Morimoto, N. L. Nair, E. The- walt, A. Little, J. G. Analytis, J. E. Moore, and J. Orenstein, "Giant anisotropic nonlinear optical re- sponse in transition metal monopnictide Weyl semimet- als," Nat. Phys. 13, 350 (2017). [22] K. Taguchi, T. Imaeda, M. Sato, and Y. Tanaka, "Pho- tovoltaic chiral magnetic effect in Weyl semimetals," Phys. Rev. B 93, 201202 (2016). [23] T. Morimoto, S. Zhong, J. Orenstein, and J. E. Moore, "Semiclassical theory of nonlinear magneto-optical re- sponses with applications to topological Dirac/Weyl [24] A. Cortijo, "Magnetic-field-induced nonlinear optical responses in inversion symmetric Dirac semimetals," Phys. Rev. B 94, 235123 (2016). [25] A. A. Zyuzin and A. Yu. Zyuzin, "Chiral anomaly and second-harmonic generation in Weyl semimetals," Phys. Rev. B 95, 085127 (2017). [26] E. J. Konig, H.-Y. Xie, D. A. Pesin, and A. Levchenko, "Photogalvanic effect in Weyl semimet- als," Phys. Rev. B 96, 075123 (2017). [27] F. de Juan, A. G. Grushin, T. Morimoto, and J. E. Moore, "Quantized circular photogalvanic effect in Weyl semimetals," Nat. Com. 8, 15995 (2017). [28] H. Rostami and M. Polini, "Nonlinear anoma- semimetals," lous Phys. Rev. B 97, 195151 (2018). photocurrents in Weyl [29] D. T. Son and A. R. Zhitnitsky, "Quantum anomalies in dense matter," Phys. Rev. D 70, 074018 (2004). [30] M. A. Metlitski and A. R. Zhitnitsky, "Anomalous axion interactions and topological currents in dense matter," Phys. Rev. D 72, 045011 (2005). [31] N. P. Armitage, E. J. Mele, and A. Vishwanath, "Weyl and Dirac semimetals in three-dimensional solids," Rev. Mod. Phys. 90, 015001 (2018). [34] D. T. Son and N. Yamamoto, "Kinetic theory with Berry curvature from quantum field theories," Phys. Rev. D 87, 085016 (2013). [35] E. V. Gorbar, I. A. Shovkovy, S. Vilchinskii, I. Rude- nok, A. Boyarsky, and O. Ruchayskiy, "Anomalous Maxwell equations for inhomogeneous chiral plasma," Phys. Rev. D 93, 105028 (2016). [36] Y. Gao, S. A. Yang, and Q. Niu, "Field Induced Posi- tional Shift of Bloch Electrons and Its Dynamical Impli- cations," Phys. Rev. Lett. 112, 166601 (2014). [37] Y. Gao, S. A. Yang, effects rical Phys. Rev. B 91, 214405 (2015). orbital magnetic in and Q. Niu, "Geomet- susceptibility," [38] E. V. Gorbar, V. A. Miransky, I. A. Shovkovy, and P. O. Sukhachov, "Second-order chiral kinetic the- ory: Chiral magnetic and pseudomagnetic waves," Phys. Rev. B 95, 205141 (2017). [39] We neglect quadratic in the fields terms to the energy of the wave-packet, Refs. [36 -- 38]. These terms do not contribute to Eq. 21, while corrections to the Hall contri- bution to the chiral current 23 arising from these terms are small τ µ ≪ 1. [40] A. A. Burkov, in "Negative Dirac longitudinal mag- and Weyl metals," netoresistance Phys. Rev. B 91, 245157 (2015). [41] J. Behrends and J. H. Bardarson, "Strongly angle- dependent magnetoresistance in Weyl semimetals with long-range disorder," Phys. Rev. B 96, 060201 (2017). [42] S. A. Parameswaran, T. Grover, D. A. Abanin, D. A. Pesin, and A. Vishwanath, "Probing the Chiral Anomaly with Nonlocal Transport in Three-Dimensional Topolog- ical Semimetals," Phys. Rev. X 4, 031035 (2014). [43] M. M. Glazov, "Second Harmonic Generation in Graphene," JETP Letters 93, 366 (2011). [44] V. I. Perel and Ya. M. Pinskii, "Direct-current in con- ducting medium induced by high-frequency electromag- netic field," Sov. Phys. Solid State 15, 688 (1973). [45] Y. Araki, "Strain-induced nonlinear spin Hall effect in topological Dirac semimetal," Arxiv: 1803.01693. [46] T. Liang, Q. Gibson, M. N. Ali, M. Liu, R. J. Cava, and N. P. Ong, "Ultrahigh mobility and gi- ant magnetoresistance in the Dirac semimetal Cd3As2," Nat. Mat. 14, 280 (2014). [47] M. Johnson and R. H. Silsbee, "Interfacial charge-spin coupling: Injection and detection of spin magnetization in metals," Phys. Rev. Lett. 55, 1790 (1985). [48] D. A. Abanin, S. V. Morozov, L. A. Ponomarenko, R. V. Gorbachev, A. S. Mayorov, M. I. Katsnelson, K. Watan- abe, T. Taniguchi, K. S. Novoselov, L. S. Levitov, and A. K. Geim, "Giant Nonlocality Near the Dirac Point in Graphene," Science 332, 328 (2011). 7 

1511.06073

1

1511

Anderson transitions in disordered two-dimensional lattices

We numerically analyze the energy level statistics of the Anderson model with Gaussian site disorder and constant hopping. The model is realized on different two-dimensional lattices, namely, the honeycomb, the kagom\'e, the square, and the triangular lattice. By calculating the well-known statistical measures viz., nearest neighbor spacing distribution, number variance, the partition number and the dc electrical conductivity from Kubo-Greenwood formula, we show that there is clearly a delocalization to localization transition with increasing disorder. Though the statistics in different lattice systems differs when compared with respect to the change in the disorder strength only, we find there exists a single complexity parameter, a function of the disorder strength, coordination number, localization length, and the local mean level spacing, in terms of which the statistics of the fluctuations matches for all lattice systems at least when the Fermi energy is selected from the bulk of the energy levels.

cond-mat.mes-hall

cond-mat

a Anderson transitions in disordered two-dimensional lattices Dayasindhu Dey,1, ∗ Manoranjan Kumar,1, † and Pragya Shukla2 1S. N. Bose National Centre for Basic Sciences, Block - JD, Sector - III, Salt Lake, Kolkata - 700098, India 2Department of Physics, Indian Institute of Technology Kharagpur, Kharagpur - 721302, India. (Dated: August 26, 2018) We numerically analyze the energy level statistics of the Anderson model with Gaussian site dis- order and constant hopping. The model is realized on different two-dimensional lattices, namely, the honeycomb, the kagom´e, the square, and the triangular lattice. By calculating the well-known statistical measures viz., nearest neighbor spacing distribution, number variance, the partition num- ber and the dc electrical conductivity from Kubo-Greenwood formula, we show that there is clearly a delocalization to localization transition with increasing disorder. Though the statistics in different lattice systems differs when compared with respect to the change in the disorder strength only, we find there exists a single complexity parameter, a function of the disorder strength, coordination number, localization length, and the local mean level spacing, in terms of which the statistics of the fluctuations matches for all lattice systems at least when the Fermi energy is selected from the bulk of the energy levels. PACS numbers: 71.23.An, 72.15.Rn, 72.80.Ng, 05.60.Gg I. INTRODUCTION The effect of impurities in the low dimensional sys- tems such as Graphene1 -- 4, nano flakes5, metallic thin films6,7,arrays of quantum dots etc., has been an in- tense area of research in the last decade. Some of these low dimensional structures have potential applications in electronic devices, because of their finite size and the performance under varying degree of disorder and dimensionality7,8. In disordered systems delocalization to localization transition is one of the intersting phe- nomenon, therefore to caracterize this transition, analysis of the electronic wavefunction is necessary. The extended to localized transition, also known as the 'Anderson tran- sition', and its dependence on various system parameters such as, disorder, dimensionality, lattice structure, sys- tem size etc., in the finite systems are a frontier area of research. This motivates us to seek an estimation of the critical disorder for the transition in finite size lattices with different geometry. In 1958, Anderson suggested that an electron inside a material can be fully localized in the presence of a large disorder9, whereas Edwards et al. showed that a transition from an extended to a localized states in the square lattice can occur at a disorder strength of 5 or 6 times higher than the band width of the Anderson model Hamiltonian10. In 1979, Abrahams et al. conjectured, based on a scaling hypothesis, that the electronic states are localized in less than three dimensional (3D) systems in the presence of any amount of disorder11,12. The scal- ing hypothesis was later on supported by many studies of the Anderson model on square lattices13. Further, Altshuler et al. showed that weak electron interaction enhances the localization in these systems, whereas un- der strong electron interaction, 2D electrons behave like a Wigner crystal and, as shown by Tanatar et al.14, even a small amount of disorder can make the system insulat- ing at zero temperature. The theoretical analysis15 sug- gested that under strong electron interactions and small disordered regime, 2D systems can be conducting. Most of these studies are done for infinite systems and square lattices, except some of the recent studies on the hon- eycomb lattices4,16. A detailed review on the Anderson transition is given in the Ref. 17. The study of Anderson transition is not only important for material sciences but also relevant to understand the influence of the wavefunction dynamics on the physical properties of the disordered systems. The presence of disorder and/or interaction leads to a randomization of the Hamiltonian, resulting in a random matrix represen- tation in a physically suitable basis e.g., site basis. The structure of the matrix e.g., degree of sparsity is sensitive to various system conditions viz., dimensionality, shape, size, and boundary conditions. The statistical behaviour of the system can therefore be analysed by an ensem- ble of the disordered Hamiltonians. Such analysis has been a subject of extensive study during the past decade. It is now well-known that, in the weak disorder regime, the statistics can be well-modelled by the Wigner-Dyson universality classes of random matrix ensembles which correspond to extended, featureless eigenfunctions and a strong level-repulsion with statistics of the eigenval- ues and the eigenfunction independent from each other. Increasing the disorder in finite size systems causes the statistics to crossover from Wigner-Dyson to the Pois- son universality class (with no level-repulsion and fully localized eigenfunctions in strong disorder limit). The statistics in the intermediate regime e.g., near critical disorder is sensitive to the degree of eigenfunction lo- calization which in turn is expected to depend on the system conditions besides disorder. The study17,18 how- ever, showed that the statistics can be well-modeled by the single parametric power-law random banded matrix (PRBM) ensembles. Another theoretical study19 later (a) (b) (c) (d) FIG. 1. (Color online) Two dimensional lattice structures considered in this paper: (a) the honeycomb lattice with co- ordination number 3, (b) the kagom´e and (c) the square lat- tices both having 4 nearest neighbors and (d) the triangular lattice with coordination number 6 on indicated the application of a wide range of random matrix ensembles (besides PRBM) as the model for the intermediate statistics; this study was based on the com- mon mathematical formulation of the energy level statis- tics of a broad class of random matrix ensembles (with varying degree of sparsity and disorder but same sym- metry class). The formulation is governed by a single parameter, referred as the complexity parameter, a func- tion of all system parameters including energy range of interest and therefore different ensembles are expected to show analogous statistics if their complexity parameters are same19. As the theoretical claim about the existence of a complexity parameter is in clear agreement with the single parameter scaling conjecture of Ref. 11, it is highly desirable to seek its numerical validity in disordered sys- tems. In this paper, we study the transitions from extended to localized state of Anderson Hamiltonian on finite 2D systems with different geometry in presence of the disor- der. As shown in Fig. 1, we consider four lattice geome- tries, namely, square, triangular, honeycomb and kagom´e lattice which differ from each other in the coordination number and their bond connectivity. The coordination numbers of honeycomb, square, kagom´e and triangular, lattice are three, four, four, and six respectively. We note that the square and the kagom´e both have four nearest neighbour but their spectrum is completely different be- cause of the bond connectivity. The paper is organized as follows. In section II, we introduce the model dis- ordered Hamiltonian; here the theory leading to single parameter based formulation of the spectral statistics is also briefly reviewed. The section III briefly describes the statistical measures used in our numerical analysis. The section IV presents the details of the numerical tech- niques as well as results; this section is divided into four 2 sub-sections one for each lattice system. The compari- son of the statistics of physical parameters with respect to the single complexity parameter for various lattices is given in section V. In section VI, results are discussed and compared with the existing literature. II. THE MODEL The Hamiltonian: We consider the standard Ander- son model9 under tight binding approximation realized on four different 2D lattice systems. The tight binding model Hamiltonian is well suited for a metallic system, where electron-electron interactions are screened. The Hamiltonian can be written in a single particle basis as H =Xi ǫiiihi +Xhi,ji tijiihj, (1) where ǫi, and tij are random site energies and the near- est neighbour hopping energies. In this work, the sites energies are considered Gaussian distributed for each lat- tice system and the hopping energies are kept constant: tij = −1.0. The Gaussian type site disorder has been widely studied for the square lattice system and has a close analogy with real thin film materials. The ensemble: The presence of the disorder in the sys- tem makes it inevitable to consider an ensemble of the Hamiltonians. For the Hamiltonian in Eq.(1) with on- site random energies, the probability density ρ(H) of the ensemble including all possibilities can be written in gen- eral as ρ(H) = C exp"−Xk 1 2hkk (Hkk − bkk)2# × Yk,l;k6=l δ(Hkl − fkl), (2) where C is the normalization constant, fkl = t if {k, l} correspond to nearest neighbour sites and otherwise fkl = 0. Writing the delta function as a limiting Gaussian (δ(x) = limv→0 ), reduces Eq.(2) to the following form 1√2πv2 e−x2/2v2 1 hkl ρ(H, h, b) = C exp Xk≤l (Hkl − bkl)2  kli − hHkli2, where h is the set of all variances hkl = hH 2 and b is the set of all mean values bkl = hHkli. Here, for constant hopping between nearest neighbor sites and an onsite Gaussian disorder W in Eq.(1), one has , (3) hkk = W 2/12, hkl = 0, bkk = 0, bkl = −f (kl)t, (4) (5) where f (kl) = 1 for {k, l} pairs representing nearest neighbors and zero otherwise. The statistical behavior of the Hamiltonian H, Eq.(1), can now be analyzed using the ensemble (3). An important aspect of Eq.(3) is that it can represent the ensemble density for a wider class of lattice systems under different conditions e.g., random, anisotropic, variable range hopping, dimensionality and boundary conditions; the only constraint on these system is to preserve the time-reversal symmetry which allows H to be real-symmetric. Its form is therefore appropriate for the verification of single parameter scaling conjecture. Single parameter formulation: Before proceeding to numerical analysis, it is helpful to briefly review the complexity parametric formulation of the spectral statis- tics. The variation of system conditions in general re- sult in a variation of the distribution parameters h and b and therefore an evolution of the H-ensemble. As dis- cussed in Ref. 19 and 20, under a change of parameters hkl → hkl + δhkl and bkl → bkl + δbkl, ρ(H) undergoes a diffusion dynamics along with a finite drift, and, using Gaussian nature of ρ(H), it can exactly be shown that T ρ = Lρ, (6) 3 eigenvalues are uncorrelated in insulator limit, thus im- plying P ({En}) = Qi<j Pn(En)21. The distribution for the intermediate states of localization can be derived by integrating ρ over the associated eigenvector space. As shown in Ref. 19, an integration of Eq.(9) leads to a sin- gle parametric diffusion of the eigenvalue distribution of the ensemble (3) P. (12) + En  ∂P ∂Y =Xn ∂ ∂En   ∂ ∂En + Xm6=n 1 Em − En The above equation can be used to obtain the ensemble averaged level density as well as its local fluctuations19. As discussed in Ref. 22, while the diffusion of the average level density is governed by Y , the diffusion of its fluc- tuations occurs at a scale determined by (Y − Y0) ∼ ∆2 l where Y0 is the value of Y at the beginning of evolution and ∆l is the local mean level spacing: ∆local(E) ≈ Ld ξ−d ∆(E) (13) where T is a combination of parametric derivatives, and L is a diffusion operator in matrix space and are given by with ξ as the localization length and d as the dimension (here d = 2). The statistics other than mean level density is therefore governed by a rescaled parameter Λ(E): T =Xk≤l(cid:20)(gkl − 2hkl) ∂ ∂hkl − bkl ∂ ∂bkl(cid:21) and L =Xkl ∂ ∂Hkl (cid:20) gkl 2 ∂ ∂Hkl + Hkl(cid:21) (7) (8) with gkl = 1 + δkl. A suitable transformation of para- metric space maps T to a single parametric derivative, T ρ = ∂ρ ∂Y , which in turn reduces Eq. (6) to a single para- metric diffusion equation19,20 ∂ρ ∂Y = Lρ. (9) Here Y = −1 N 2 Xk≤lh ln1 − (2 − δkl)hkl + ln bkl + δb02i+Cy (10) with Cy as an arbitrary constant and δb0 = 1 if bkl = 0 else δb0 = 0. For the ensemble described by the set of parameters in Eq.(5), this leads to Λ(E) = Y − Y0 ∆2 l . (14) The solution of Eq.(12) at Λ → ∞ corresponds to the Wigner-Dyson statistics. The Λ → 0 limit correspond to the distribution at the initial state of the evolution; for an insulator initial state, the spectral statistics reduces to the statistics of uncorrelated energy levels. The transition parameter Λ is in general a function of various parameters e.g., disorder, system size, dimension- ality, energy range of interest, lattice topology. Although both Y and ∆l contribute to the system dependence of Λ, the crucial influence comes from ∆l due to its dependence on the localization length ξ. For finite system sizes N , a variation of system conditions e.g., disorder leads to a smooth crossover of statistics between the stationary lim- its Λ → 0 and Λ → ∞. In infinite size limit, the statistics abruptly changes from one stationary limit to the other. If however, the limit Λ∗ = limN→∞ Λ(N ) exists, the cor- responding statistics would belong to a universality class different from the two stationary limits. The existence of Λ∗ therefore is a criteria for the existence of critical spectral statistics19. .t + δt0z/2(cid:21) + Cy, (11) III. FLUCTUATION MEASURES: THE DEFINITIONS Y = −1 N 1 − ln(cid:20)(cid:12)(cid:12)(cid:12)(cid:12) W 2 12 (cid:12)(cid:12)(cid:12)(cid:12) where z is the coordination number of the lattices. The joint probability distribution of the eigenvalues P ({En}) ≡ P (E1, . . . , EN ) for a metal (fully extended eigenfunctions limit) is given by the Wigner-Dyson distri- bution, P ({En}) =Qi<j Ei − Ej exp(− 1 k). The 2Pk E2 Our main objectives in this paper is to study the influ- ence of the system conditions on the statistical behavior and identify the critical regime. For this purpose, we consider four different fluctuation measures namely, den- sity of states (DOS), reduced partition number (P/L), the peak position of the NNSD and the dc electrical con- ductivity which can briefly be described as follows. Spectral measures: A well-known measure to analyse the short range correlations among energy levels is the nearest neighbour spacing distribution (NNSD) which de- scribes the probability P (s) of two nearest neighbour en- ergy levels to be found at a distance s measured in the units of the mean level spacing around the desired en- ergy regime. For the weak disorder regime with extended eigenfunctions, P (s) is given by the Wigner surmise PW (s) = π 2 s exp(cid:16)− π 4 s2(cid:17) . (15) For the opposite limit of strong disorder with fully local- ized wavefunctions, P (s) follows Poisson distribution PP (s) = exp (−s) . (16) To compare the level spacing distribution for the entire transition within an arbitrary energy regime for different lattices, we use a traditional measure, namely, the cu- mulative nearest neighbor spacing distribution ηi which depends on the tail behaviour of the nearest neighbour spacing distribution and is defined as si ηi = R ∞ P (s, Λ)ds −R ∞ PP (s)ds −R ∞ R ∞ si si si PW (s)ds PW (s)ds , (17) where si, i = 1, 2 refer to the two crossing points of PW (s) and PP (s): s1 = 0.473 and s2 = 2.00221,23,24. As the system makes a transition from delocalized to localized state, ηi changes from 0 to 1. While NNSD gives the short range correlations of the energy levels, there is another measure which gives the long range correlations, namely, the number variance. It is defined as the variance of the number n of unfolded energy levels in an energy interval r centered at the energy regime of interest i.e., the number variance Σ2(E, r) = h(n(E, r) − hn(E, r)i)2i. Participation number: The dependence of Λ on the localization length ξ through ∆l results in sensitivity of the energy level statistics to eigenfunction behaviour. This motivates to analyse a standard measure for the eigenfunction localization, namely, the reduced partici- pation number, referred here as P/L, which character- ize the spread of eigenfunctions in the site basis. The partition number P for a wavefunction ψn is defined as ψin4 and is proportional to the localization DC conductivity: The localization of electronic wave- function can be characterized by the DC electrical con- ductivity σ which can be measured also experimentally for real systems. Here we numerically calculate σ using Kubo-Greenwood formula (see Ref. 25 for details) P −1 =PN length. i σ(EF ) = 2π Ω Tr [Jδ(EF I − H)Jδ(EF I − H)] (18) where EF is the Fermi energy, H is the Hamiltonian, Ω is the volume of the system and J is the one electron 4 current operator ~J = ieV  Xhi,ji (Ri − Rj )(iihj − jihi) (19) with Ri as the position vector of site i. IV. NUMERICAL ANALYSIS AND RESULTS To study the influence of system conditions on the sta- tistical behavior, we apply exact diagonalization tech- nique to numerically obtain the eigenvalues and the eigenfunctions of the Hamiltonian (1) for four different lattice types. For each lattice type, an ensemble of ap- proximately 1000 realizations is considered to attain sta- tistical accuracy. To explore the size-dependence of the statistics and its sensitivity to the energy range, many system sizes, varying from L = 40 to L = 100, are analysed for each lattice type and for two energy ranges. The latter correspond to two filling of the Hamiltonian: firstly, the half-filling which is a more natural choice of the systems like graphene, gold etc., and secondly, a fill- ing where the density of states is significantly high, mak- ing a rigorous statistical analysis of energy levels more feasible. For the local fluctuations analysis, it is neces- sary to first rescale the energy eigenvalues by the local mean level spacing26 (also known as unfolding of the lev- els). The average density of states for this purpose is calculated using the binning method. To analyse the local spectral fluctuations in the desired energy-range, approximately ∼ 3% of the rescaled energy levels are taken around the chosen Fermi energy. The rescaled en- ergy levels and normalized eigenfunctions are then used to calculate the density of states (DOS), reduced parti- tion number (P/L), the peak position of the NNSD and the dc electrical conductivity. The results are arranged in order of increase in the coordination numbers in the sub- sections A, B, C and D containing honeycomb, kagom´e, square and triangular lattices respectively. Here we de- scribe three different methods of calculation in detail for the honeycomb lattice only; the calculations for other lat- tice systems are carried out following similar methods. A. Honeycomb lattice The 2D honeycomb structure is one of the most in- teresting lattice types which occurs in many systems of industrial importance e.g., graphene. In this case, the non-interacting electronic models like uniform Anderson model or tight binding model have two unique Dirac cones in a brillouin zone (BZ). The DOS ρ(ǫ) for a 2D honeycomb lattice in the clean limit (i.e. absence of dis- order) is given by ρ(ǫ) = − 1 π Im(cid:20) lim ∆ǫ→0+Z1stBZ dk V V 2 − µ2t2(cid:21) , (20) W = 0 3.0 3.6 4.0 25.0 0.3 0.2 ) F E ( r 0.1 0 EF = -1.5 EF = 0 10 W 20 0.2 0.1 e) ( r 0 -8 -4 0 e 4 8 20 10 r e b m u N n o i t a p i c i t r a P 0 0 5 5 4 x W 3 2 0 0.01 0.02 1/L 0.03 2 4 W 6 FIG. 2. (Color online) The DOS of the honeycomb lattice. The solid line shows the DOS in the absence of disorder. Three disorders strength are chosen around the critical dis- order (see table I). The DOS at very large disorder is also shown. The inset shows the variation of DOS with disorder at two energy regimes. FIG. 3. (Color online) The partition number per unit length with disorder for different system sizes are shown at (a) EF = −1.5t and (b) EF = 0. The linear fits of the curves cut the disorder axis at Wx. The insets of both (a) and (b) show the critical disorder Wx with 1/L which are fitted with the lines: Wx = 3.13 + 68.6/L for inset (a) and Wx = 2.64 + 72.6/L for inset (b). where V = ǫ + i∆ǫ and µ = exp(ikxa) + exp[i(−kxa/2 + ky√3a/2)] + exp[i(−kxa/2 − ky√3a/2)] with kx and ky as the components of the wave vector27; ρ(ǫ) in this case vanishes at ǫ = ±3t. In half-filled limit, the sys- tem shows a gapless state and a linear dispersion relation which can be mapped to that of a massless Dirac state. In real materials, the vacancies can lead to coupling of cones but coupling strength remains very small, the two cones being separated by a large momentum vector28. The microscopic calculation for the two valley Hamilto- nian indicates a crossover from anti-localization to weak localization28. A recent experiment also confirms the anti-localization and weak localization of states at low and high carrier concentration respectively (for defects like charge impurity etc.). To observe the effect of disorder, the DOS ρ(ǫ) is cal- culated numerically for different disorders; the results are shown in Fig. 2. In the clean system (W = 0), ρ(ǫ) is vanishingly small near energy ǫ = 0 (the Fermi-energy at half filling) and has van-Hove singularities at ǫ = ±t . The effect of varying disorder on ρ(ǫ) at ǫ = −1.5t (i.e., at the bulk of DOS) and at ǫ = 0 (i.e., at the half filling) is shown in the inset of Fig. 2. At the half filled energy regime, ρ(ǫ) increases till W = 7.0t and then decrease ex- ponentially as shown in the inset of the Fig. 2. In large W limit, DOS becomes flat as shown in the Fig. 2. As clear from the Fig. 2, the disorder has a significant im- pact on the DOS of this lattice type in the energy regime near ǫ = 0 (with DOS showing a prominent dip for small disorder) and ǫ = ±1.5t. For the fluctuations analysis we therefore, choose Fermi energy from these two regimes, namely, ǫ = 0 (half filling) and ǫ = −1.5t (bulk). Our next step is to seek the critical disorder for the extended to localized state transition in the honeycomb lattice by three different routes. While the transition in principle takes place in the thermodynamics limit L → ∞ 0.8 (a) 5 (i) * W 4 0.6 3 0 0.01 0.02 0.03 L-1 L = 40 50 70 100 W = 2.0 3.0 3.6 4.0 25.0 (ii) 0.8 ) S P ( 0.4 k p S 0.4 0.2 0 0 0.8 (b) 0.6 k p S 0.4 0.2 0 0 0 0 1 2 S 3 0.1 0.2 W -1 0.3 0.4 6 (iii) * W 4 0 L = 40 50 70 100 0.01 0.02 0.03 L-1 (iv) 0.8 ) S P ( 0.4 W = 2.0 3.6 4.0 4.5 25.0 0.1 0 0 0.2 W -1 1 2 S 0.3 3 0.4 FIG. 4. (Color online) The peak-position of the NNSD with the inverse of the disorder (a) at EF = −1.5t and (b) at EF = 0. The lines in the main part of (a) and (b) are the fitted curves with the function 0.77 tanh((W ∗/W )2). The critical disorder for various system sizes are depicted in insets (i) and (iii). The insets (ii) and (iv) show the NNSD at different disorders for the system size L = 100 (a) 3 (b) ln Wc 0 1 ln W 2 0 s n l -2 -4 -1 -2 -3 -4 s n l 0 1 ln W 2 3 FIG. 5. (Color online) The dc conductivity with disorder at (a) ǫ = −1.5t and (b) ǫ = 0. The upper part in (b) is fitted with: ln σ = 0.99 − 1.84 ln W and the lower part is fitted with ln σ = 1.577 − 2.265 ln W which gives Wc = 3.9t. at a critical disorder (or critical energy), the finite sys- tems undergo a smooth crossover within a critical regime (around critical disorder or energy). It is therefore imper- ative to analyze the critical disorder Wcrit(L) for many finite system sizes, with exact critical disorder given by limL→∞ Wcrit(L) = Wcrit. For this purpose, we first analyse the disorder dependence of the reduced partic- ipation number P/L. Fig. 3 elucidates the behavior of the reduced partition number at two Fermi energies at (a) the bulk and (b) at the half filling respectively. For each case P/L varies linearly with the disorder W in the small disorder regime while it varies exponentially for strong disorders. The linear portion of the curves are fit- ted with lines (P/L = a−bW ) which cut the disorder axis at Wx = a/b. For W > Wx, the localization length of the systems is less than the system size. At W = Wx, the lo- calization length become equal to the system size L, thus suggesting Wx as the finite size critical disorder (more clearly W pn crit(L) = Wx with superscript referring to the method applied) . (Note, for later reference, here we use different notations for the critical disorder obtained by different methods). The critical disorder measured from this method for different system sizes of the honeycomb lattice are given in the table I. As shown in the insets 6 of Fig. 3(a) and (b), Wx for different system sizes vary linearly with the inverse of the system size. The numer- ics reveals the sensitivity of Wx to Fermi energy too: for L = 100, the value of Wx is 3.4 at the half filling and 3.8 at the bulk of the DOS. Continuing with our quest for critical disorder, our next step is to analyze the Nearest neighbor level spac- ing distribution (NNSD) defined in section III. The de- crease of disorder in the finite systems causes the NNSD to crossover from the Poisson distribution (eq.(16)) to the Wigner surmise (eq.(15)). To analyse the NNSD for different disorder strengths, the energy eigenvalues are apriori unfolded using the local mean level spacing. The numerically obtained NNSD for a given disorder and sys- tem size is compared with the Brody distribution29: PB(S) = α(1 + ω)Sω exp(−bS1+ω) (21) which gives the Brody parameters ω and b as a function of W and L. The fitted Brody distribution is then used to calculate the peak position dependence on the disorder strengths W for a fixed size L. The dependence is plot- ted with respect to inverse of the disorder in Fig. 4 for many system sizes and at two Fermi energies (the bulk and the half filling). The curve depicting peak positions with respect to disorder for each L is now fitted with the function F (W ) = 0.77 tanh((W ∗/W )n) (22) where n = 2 and W ∗ is the critical disorder strength (in this method) at which localization occurs for a fi- nite system size L (more clearly W nnsd crit (L) = W ∗). This type of functional dependence of the peak position of NNSD was suggested by A J Millis et al. in context of the energy level statistics of one dimensional spin-1/2 chain to find the critical value Jc indicating integrable and non-integrable boundary30. As shown in the main Fig. 4 (b), there is large deviation in the peak position- disorder curve from the function F (w) in the small dis- order regime for the half filled case. F (W ) with n = 2 however fits well in small W limit for the case where the Fermi energy is chosen away from the half filling. As shown in the insets (i) and (iii) of Fig. 4 (a) and (b) respectively, the crossover disorder W ∗ decreases with system size L. Again W ∗ shows an energy dependence for a fixed L: for L = 100, W ∗ = 3.9 and 3.6 for ǫ = 0 and ǫ = −1.5t respectively (see table I). The insets (ii) and (iv) of Fig. 4 (a) and (b) also display the NNSD behavior for ǫ = 0 and ǫ = −1.5t respectively. We notice that for ǫ = 0 or half filled case the NNSD vary signif- icantly in small impurity limit but the variation in the bulk limit ǫ = −1.5t is weak below W = W ∗. The un- usual behaviour can be explained from the rapid change in the DOS with disorder at the half filling. The half fill- ing case suggests that the system is still in the ballistic limit for small disorder. The electronic conductivity is an important character- istic of the localization to delocalization transition, with high conductivity an indicator of the large localization ǫ −1.5t L 40 50 70 100 Wx 4.9 4.4 4.2 3.8 W ∗ 5.2 4.8 4.3 3.9 Wc 3.9 TABLE I. Critical disorders calculated from all three methods for the honeycomb lattice at two different Fermi energies for four system sizes length. The dc electrical conductivity for various system sizes for the two energy ranges are calculated using the Kubo-Greenwood formula. The minimum conductivity σm of a clean sample at T = 0 and half-filled limit is an- alytically predicted to be 4e2/πh; this is also confirmed by our analysis. The effect of changing disorder on the behavior of conductivity (σ) in honeycomb lattice is dis- played in the figure 5 (a) and (b) at the bulk of DOS and at the half filling respectively. In the bulk limit, the con- ductivity decreases with increasing disorder following a power law dependence with exponents −1.84 in the weak disorder regime and −2.265 in the strong disorder regime. The intersection of these two lines gives the crossover dis- order W cond. crit (L) = Wc = 3.9t which is almost in agree- ment with the values calculated from the other two meth- ods. The disorder-dependence of conductivity at the half filling however shows some atypical behavior: it remains almost constant at very small disorders up to W = 0.4t, decreases thereafter up to W < 2.8t, then increases till W = 5.0 and decrease afterwards for further increase in W . A possible explanation of this flip-flop behavior may lie in poor statistics due to small number of states avail- able at this filling or due to dominant finite size effect. We intend to probe this behavior with more numerical rigor separately. Next we compare the crossover values of Wcrit(L) cal- culated from three different methods (i.e Wx, W ∗, Wc) for four different L. Due to statistical reliability, the comparison is shown in table I (only for the bulk energy limit). As given in the table I, Wx and W ∗ decrease with an increase in the system size L, varying as L−1 (As the size dependence on the conductivity is negligibly small for this case, we present the data for L = 100 case only). This clearly indicates conductivity as a better measure to find the critical disorder for the transition (due to faster convergence with L). The analysis suggests the critical disorder Wcrit ≈ 3.8. B. Kagom´e lattice 0.2 0.2 ( r e) ( r 0.1 ) F E 0.1 0 10 W 20 7 W = 0 4.5 5.0 6.0 21.0 0 -8 -4 0 e 4 8 FIG. 6. (Color online) The DOS of the kagom´e lattice for different disorders. The solid line is the DOS for the clean system. The inset shows the DOS at half filling with the disorder. weight 1/3. The DOS diverges at three values of energy (at ǫ = 0 and ±2t) and is zero for ǫ = −t for the clean system31. The band width in the clean limit of this lattice is spanned from ǫ = −4t to ǫ = 2t. Effect of the onsite disorder on the DOS of the kagom´e lattice Hamiltonian of system size L2 with L ∼ 100 is shown in the Fig. 6 for three disorder strengths alongwith the clean limit. As clear from the figure, the singularities in the DOS vanish with the increase in disorder, with DOS approaching a Gaussian distribution in energy at very large disorder. The inset in Fig 6 displays the behavior of the DOS at the Fermi energy for varying disorders which turns out to be an exponential decay in large disorder limit. To determine the critical disorder Wcrit in this case, we again apply the three methods mentioned in detail for the honeycomb lattice. The results for reduced par- 20 15 / L P 10 5 0 0 7 6 x W 5 4 0 0.01 0.02 1/L 0.03 L = 97 69 48 41 2 4 6 W 8 10 12 The DOS ρ(ǫ) of the Anderson model with uniform site energy for a kagom´e lattice is exactly solvable in the clean limit, and is composed of the two bands of the honeycomb lattice shifted in energy by t with amplitude reduced by a factor of 2/3 and a delta peak at ǫ = 2t with FIG. 7. (Color online) The reduced partition number for dif- ferent system sizes of the kagom´e lattice. The linear portion of the curves are fitted with lines which cut the disorder axis at Wx. The inset shows the critical disorder Wx with the inverse of the system size and which follows the fitted line: Wx = 4.0 + 91.3/L.  0.8 0.6 * 6 W 4 0 k p S 0.4 0.2 0 0 0.01 0.02 0.03 L-1 0.8 ) S P ( 0.4 L = 40 50 70 100 W = 2.0 4.0 4.5 5.0 21.0 0.1 0 0 0.2 W -1 1 2 S 3 0.3 FIG. 8. (Color online) The peak-position of the NNSD vs the inverse of the disorder. The lines in the main figure are fitted curves with 0.77 tanh((W ∗/W )2). The top left inset shows the critical disorder with the system size which follows: W ∗ = 3.67+108.5/L. The bottom right inset shows the NNSD of the kagom´e lattice of system size L ∼ 100 for different disorders ticipation number P/L are displayed in Fig. 7 for differ- ent system sizes in the kagom´e lattice. As shown in the inset of Fig. 7, the finite size critical disorder Wx calcu- lated from this method varies linearly with the inverse of the system size. The results for the search of crit- ical disorder through the NNSD peak position analysis are displayed in Fig. 8, for four different system sizes of kogom´e lattice. Here again the curve describing peak position-disorder dependence is fitted with the function F (W ) given by Eq.(22) with n = 2. As shown in the inset of Fig. 8, W ∗ decrease with system size L. The behavior of NNSD shown in the inset of Fig. 8 indicates a very small variance in NNSD for W < W ∗. The disorder dependence of the electronic conductivity for kagom´e lattice is displayed in Fig. 9 (shown here only for L = 100); it reveals a decreasing conductivity with 0 -2 -4 -6 s n l ln Wc 8 increase of disorder, along with two different power law behaviors in small and large disorder ranges (indicated by two linear regimes with different slopes in the log σ- log W plot). For a system size L ≈ 100, the crossing point of the two lines lies at Wc ∼ 5.9t. Table II displays the critical disorder Wcrit(L) calcu- lated from three different methods for four different sys- tem sizes. Similar to the honeycomb lattice case; the size dependence of Wc is not presented as it is negligibly small for this system also. whereas Wx and W ∗ calculated from participation number and shift in peak height of NNSD is linear with L−1. Similar to the case of honeycomb lat- tice, here again the calculations suggest a convergence of Wc to its critical value lattice, once again confirming a weaker sensitivity of the conductivity approach to finite- ness of the kagom´e lattice. L 40 50 70 100 Wx 6.2 5.9 5.3 4.9 W ∗ 6.4 5.9 5.2 4.8 Wc 5.9 TABLE II. Critical disorders calculated from all three meth- ods for the kagom´e lattice for four system sizes C. Square lattice The DOS ρ(ǫ) of an Anderson model with uniform site energy for a square lattice at the clean limit can be cal- culated from the band dispersion relation ǫ(k) = 2t(cos kx + cos ky). (23) In the clean limit, the DOS has a van Hove singularity at W = 0 W = 5.0 W = 5.6 W = 6.5 W = 25.0 0.3 0.2 ) F E ( r 0.1 0 4 8 12 16 20 24 W 0.2 e) ( r 0.1 0 1 ln W 2 3 0 -12 -8 -4 0 e 4 8 12 FIG. 9. (Color online) DC conductivity of the kagom´e lattice of linear size ≈ 100 with disorder. The upper part of the plot follows: ln σ = 0.938 − 1.9 ln W and the lower part follows: ln σ = 3.514 − 3.2 ln W and the critical disorder Wc = 5.9. FIG. 10. (Color online) The DOS of the square lattice at different disorders. The solid line is the DOS for a clean system. The inset shows the DOS at the half filling with the disorder. ǫ = 0 and vanishes for ǫ > 4t. Effect of on-site disorder on the DOS for four system sizes of square lattices is shown in the Fig. 10. As shown in the inset of Fig 10 the DOS at half filling decays exponentially with disorder which is consistent with the Ref. 32. Following the same procedure as mentioned in the case of honeycomb lattice, here again we analyze the disorder dependence of the reduced partition number P/L, NNSD and DC conductivity for different system sizes but only in bulk energy regime; the results are shown in Fig. 11, Fig. 12 and Fig. 13 respectively. As clear from the figures, the qualitative behavior in this case is same as in the bulk energy regimes of honeycomb and kagome lattices; a quantitative difference however shows up in the fitted line Wx = 5.39 + 106.8/L for P/L, with power n for fit F (W ) as 3 (instead of 2 as in previous two cases), with W ∗ = 5.38 + 118.4/L and Wc = 6.5t. The critical disorders Wx, W ∗ and Wc from the three methods are given in table III. Wx, W ∗ and Wc for this case show an inverse linear dependence on the system size. L 40 50 70 100 Wx 8.0 7.6 6.9 6.5 W ∗ 8.3 7.8 7.0 6.6 Wc 7.6 7.0 6.5 TABLE III. Critical disorders calculated from all three meth- ods for the square lattice at half filling for four system sizes 0.8 0.6 * 8 W 6 0 0.01 0.02 0.03 L-1 k p S 0.4 0.2 0 0 9 L = 40 50 70 100 W = 2.0 6.0 6.6 7.0 25.0 0.8 ) S ( I 0.4 0 0 1 2 S 3 0.1 W -1 0.2 FIG. 12. (Color online) The peak position of the NNSD of the square lattices of various system sizes. The curves in the main plot are fitted with: 0.77 tanh((W ∗/W )3). The top left inset W ∗ are plotted with inverse of the system size which follows the line: W ∗ = 5.38 + 118.4/L. The bottom right inset depicts the NNSD at different disorder for the square lattice of size L = 100. 2 0 -2 -4 -6 s n l ln Wc 0 1 2 ln W 3 (Color online) DC conductivity of the disordered FIG. 13. square lattice at the half filling. The upper part is fitted with: ln σ = 1.807 − 1.723 ln W and the lower part is fitted with: ln σ = 5.845 − 3.879 ln W which gives Wc = 6.5t. 9 8 x W 7 6 5 D. Triangular lattice 0 0.01 0.02 1/L 0.03 In the clean limit, the DOS ρ(ǫ) for a triangular lattice can exactly be obtained from the band dispersion relation L = 100 L = 70 L = 50 L = 40 ǫ(k) = −t"2 cos(kx) + 4 cos(cid:18) kx 2 (cid:19) cos √3 2 ky!# , (24) 25 20 15 / L P 10 5 0 0 2 4 6 W 8 10 12 FIG. 11. (Color online) The reduced partition number of the square lattice of different system sizes. The upper part of the curves are fitted with lines which cut the disorder axis at Wx. The inset shows the critical disorder Wx with the inverse of the system size and the fitted line: Wx = 5.39 + 106.8/L. where k is confined to the first Brillouin Zone31. The DOS obtained numerically in the clean limit along with the presence of disorder is shown in the Fig 14. The analysis indicates a van Hove singularity in the DOS at energy E = 2t and it approaches zero for ǫ < −6t and ǫ > 3t. The energy dependence of the DOS for four dis- orders is shown in the Fig. 14. In the inset of Fig 14, the disorder dependence of the DOS at a fixed energy (half W = 0 7.0 9.0 11.0 25.0 12 * W 9 6 0 0.01 0.02 0.03 L-1 0.8 0.6 k p S 0.4 0.2 0.8 ) S P ( 0.4 10 L = 40 50 70 100 W = 2.0 7.0 9.0 11.0 25.0 ) F E 0.1 0 10 W 20 0.2 ( r 0.2 e) ( r 0.1 0 -8 -4 0 e 4 8 FIG. 14. (Color online) The DOS of the triangular lattice in presence of different disorder strengths. The solid line is the DOS for the system without any disorder. The inset shows the variation of DOS at the half filling with disorder. filled energy) is displayed which indicates an exponential decrease in DOS with increasing disorder. The results for the search of critical disorder in tri- angular lattice by the three methods (same as men- tioned above for previous lattices) are shown in Fig. 15, Fig. 16 and Fig. 17 respectively. The figures again in- dicate the same qualitative behavior as in the bulk en- ergy regimes of honeycomb, kagom´e and square lattices but the quantitative difference shows up in the fitted line Wx = 6.88 + 169.3/L for P/L, with power n for fit F (W ) as 3 (same as in square lattice but different from hon- eycomb and kagom´e), with W ∗ = 7.10 + 165.8/L and Wc = 8.7t. A comparison of critical disorders Wx, W ∗ and Wc (for Fermi energy in bulk) from the three meth- ods is displayed in table IV which confirms, as for previ- ous three lattice types, an inverse linear dependence for 30 20 / L P 10 12 10 x W 8 6 0 0.01 0.02 1/L 0.03 L = 100 70 50 40 0 0 2 4 6 8 W 10 12 14 FIG. 15. (Color online) The reduced partition number of triangular lattice of different system sizes with disorder. The are the linear fit for the reduced partition number curve which cut the W axis at Wx. The finite size dependence of the critical disorder Wx is shown in the inset which follows the line: Wx = 6.88 + 169.3/L. 0 0 0.04 0.08 W -1 0 0 1 2 S 3 0.12 0.16 FIG. 16. (Color online) The peak position of the NNSD with the inverse disorder for the triangular lattice of dif- ferent system sizes. The data is fitted with the function 0.77 tanh((W ∗/W )2). The top left inset shows the varia- tion of W ∗ with system size which follows the fitted line: W ∗ = 7.10 + 165.8/L. The bottom right inset shows the NNSD of the triangular lattice of system size L = 100 calcu- lated at the half filling. Wx, W ∗ and Wc in this case too. 2 0 s n l -2 -4 -6 0 ln Wc 1 2 ln W 3 FIG. 17. (Color online) DC conductivity of the disordered triangular lattice at the half filling. The upper part is fitted with: ln σ = 2.126 − 1.902 ln W and the lower part is fitted with: ln σ = 5.812 − 3.606 ln W . The crossing point of these two fitted curves gives Wc = 8.7. L 40 50 70 100 Wx 11.1 10.3 9.2 8.6 W ∗ 11.2 10.4 9.5 8.7 Wc 10.4 9.4 8.7 TABLE IV. Critical disorders calculated from all three meth- ods for the triangular lattice at half filling for four system sizes V. COMPLEXITY PARAMETER FORMULATION OF THE TRANSITION (a) 1 As discussed in previous section, the numerical analy- sis reveals the qualitative insensitivity of the local fluc- tuations in physical properties to system parameters (within a fixed energy range) although a quantitative dependence is indicated. More clearly, for each lattice type with Fermi energy in the bulk we observe the fol- lowing behavior: (i) the reduced particiapation ratio has a linear/exponential dependence on the disorder in weak/strong disorder regime, respectively, (ii) the disor- der dependence of the peak positions of NNSD can be described by the function F (w) (with different n values), (iii) the conductivity has a power law dependence on the disorder with different exponents in weak and strong dis- order regime. The observed behavior therefore strongly suggests the possibility of a common mathematical for- mulation of the statistical properties where system de- pendence enters through a single function of all system parameters. The theoretical steps for such a formulation are briefly reviewed in section II, with the function re- ferred as the complexity parameter. As defined in Eq. 11, (a) 1 0.8 0.6 1 h 0.4 0.2 0 10-4 (b) 1 0.8 0.6 1 h 0.4 0.2 0 0 Honeycomb Kagome Square Triangular 10-3 10-2 10-1 100 L 101 102 103 Honeycomb Kagome Square Triangular 5 10 15 W 20 25 FIG. 18. (Color online) The variation of η1 with (a) the rescaled complexity parameter Λ for four different two di- mensional lattices and (b) the disorder W . As obvious, the behaviour for different lattices coincide in terms of Λ but not in terms of disorder. 11 Honeycomb Kagome Square Triangular 10-3 10-2 10-1 100 L 101 102 103 Honeycomb Kagome Square Triangular 5 10 15 W 20 25 0.8 0.6 2 h 0.4 0.2 0 10-4 (b) 1 0.8 0.6 2 h 0.4 0.2 0 0 (Color online) The variation of η2 with (a) the FIG. 19. rescaled complexity parameter Λ for four different two di- mensional lattices and (b) the disorder W . As obvious, the behaviour for different lattices coincide in terms of Λ but not in terms of disorder. the function Y is a combination of various system param- eters, with explicit dependence on the disorder strength, hopping and system size. The information about dimen- sionality and boundary conditions is implicitly contained in the sparsity of the matrix H as well matrix element identities and therefore in the summation in Eq. 10. The necessary rescaling of energy levels for comparison of the fluctuations however leads to Λ (Eq. 14) as the relevant transition parameter; the rescaling therefore introduces the crucial dependence on the dimensionality as well as on the Fermi energy. The obvious relevance of the theo- retically obtained single parameter governing the transi- tion renders its numerical/ experimental verification very desirable. For this purpose, we consider here three well- known spectral fluctuation measures namely, the cumula- tive NNSD η1, η2 and the number variance Σ2(r) (defined in section III) of the four lattices and analyze their evo- lution in terms of the complexity parameter Λ instead of disorder W . The verification in context of the eigen- function fluctuations is yet to be carried out and will be reported elsewhere. To calculate Λ for our analysis, we use the fact that the localization length ξ is proportional to the average partic- (a) 10 8 6 ) r ( 2 S 4 2 0 10-4 (b) 10 8 6 ) r ( 2 S 4 2 0 0 Honeycomb Kagome Square Triangular 10-3 10-2 10-1 100 L 101 102 103 Honeycomb Kagome Square Triangular 5 10 15 W 20 25 FIG. 20. (Color online) The number variance Σ2(r) at r = 10 with (a) the rescaled complexity parameter Λ for four different two dimensional lattices and (b) the disorder W . As obvious, the behaviour for different lattices coincide in terms of Λ but not in terms of disorder. ipation number P . Fig. 18, 19 and 20 show the results for η1, η2 and Σ2(r), respectively, for four lattices for Fermi energy at e = 0. As clear from the part (a) of these fig- ures, Λ-governed evolution of each of these measure for all four lattices falls almost on the same curve for en- tire crossover from the localization to delocalization; the difference of connectivity of the lattices does not affect their behavior. Note a disorder (W )-dependent evolu- tion of η1 and η2 for honeycomb and kagome lattices is expected to differ from that of square and triangular ones (as suggested by the observed difference in disorder gov- erned evolution of the NNSD-peaks of the lattices, with η1 and η2 being cumulative NNSDs); the deviation of the results for four lattices is clearly visible from part (b) of Fig. 18, 19 and 20). This clearly reveals Λ as the param- eter in terms of which the transition in spectral statistics in Anderson lattices follows a universal route. VI. DISCUSSION From our results it is clear that the statistical behav- ior of two dimensional finite size lattices depends on the 12 coordination number, the lattice connectivity and the system size. The statistics is different for the lattices with same coordination number with different connectiv- ity which is evident from the spectral statistics of the square lattice and the kagom´e lattice (both having same coordination number). The spectral averaged density of states for all the lattices, considered in this paper, show strong disorder sensitivity in weak disorder regime, and, at least one van-Hove singularity in clean limit. The po- sition of the singularity is sensitive to the lattice type; it occurs at energy ǫ = 0 for the square lattice, at ǫ = 2.0t for the triangular lattice, at ǫ = ±t for the honeycomb lattice and at ǫ = 0, −2t for the kagom´e lattice. The den- sity of states for the honeycomb lattice and the kagom´e lattice are related to that of the triangular lattice31 and can be written in terms of the DOS of the triangular lattice in the clean limit. The DOS at ǫ = ǫF varies differently for different lattice systems at small disorder region whereas they all decay exponentially for strong disorder region. At very large disorder, the DOS for all the lattice system lead to a Gaussian distribution. Three methods are used to estimate the critical disor- der for delocalization to localization transition in two di- mensional finite size lattices. First, the reduced partition number (P/L) are used to find the critical disorder Wx. P/L varies linearly in the weak disorder regime whereas, it varies exponential in the strong disorder regime. Next, we study the peak position of the NNSD of each lat- tice type for four system sizes; which fits with function 0.77 tanh((W ∗/W )α) where α = 3 for the square and the triangular lattices whereas, α = 2 for the honeycomb and the kagom´e lattices. Last, the Kubo-Greenwood dc conductivity is to find the critical disorder (Wc) as the conductivity follows two power law decay in the weak and strong disorder regimes. The critical disorders calculated from all the methods are in agreement with each other. We also analyze dependence of the critical disorder on finite size and lattice structure considering four system sizes for all the cases. Our results indicate (i) a linear varaition of critcal disorder with 1/L, (ii) it is smallest for the honeycomb lattice and largest for the triangu- lar lattice and increases with the coordination number for a finite lattice of size L. Although the coordination number is same for the kagom´e lattice and the square lat- tice the critical disorders are different for the two cases. Therefore, the critical disorder depends not only on the coordination number but also on the lattice connectivity. Finally, we compare three fluctuation measures namely, the cumulative NNSD measures η1, η2 and the number variance Σ2(r) for the four lattices and study their evolution with the single complexity parameter Λ. The reults confirm the single parameter dependence of the localization to delocalization transition in Anderson Hamiltonian in context of the spectral statistics. 13 VII. ACKNOWLEDGMENTS M. K. thanks DST for Ramanujan fellowship grant vide No. SERB/F/3290/2013-2014 and DST Nanomission for the CRAY computational facility. 15 E. Abrahams, S. V. Kravchenko, and M. P. Sarachik, Rev. Mod. Phys. 73, 251 (2001). 16 P. M. Ostrovsky, I. V. Gornyi, and A. D. Mirlin, Phys. Rev. Lett. 98, 256801 (2007); S. Ryu, C. Mudry, H. Obuse, and A. Furusaki, 99, 116601 (2007); K. No- mura, M. Koshino, and S. Ryu, 99, 146806 (2007); P. San- Jose, E. Prada, and D. S. Golubev, Phys. Rev. B 76, 195445 (2007); C. H. Lewenkopf, E. R. Mucciolo, and A. H. Castro Neto, 77, 081410 (2008). 17 F. Evers and A. D. Mirlin, Rev. Mod. Phys. 80, 1355 (2008). 18 A. D. Mirlin, Phys. Rep. 326, 259 (2000). 19 P. Shukla, J. Phys.: Condens. Matter 17, 1653 (2005). 20 P. Shukla, Phys. Rev. E 71, 026226 (2005). 21 B. I. Shklovskii, B. Shapiro, B. R. Sears, P. Lambrianides, and H. B. Shore, Phys. Rev. B 47, 11487 (1993). 22 A. Pandey, Chaos, Solitons & Fractals 5, 1275 (1995), quantum Chaos: Present and Future. 23 R. Berkovits and Y. Avishai, Phys. Rev. Lett. 80, 568 (1998). 24 E. Cuevas, Phys. Rev. Lett. 83, 140 (1999). 25 K. Niizeki, Prog. Theor. Phys. 62, 1 (1979). 26 F. Haake, Quantum Signatures of Chaos, 3rd ed. (Springer, 2010). 27 B. A. McKinnon and T. C. Choy, Aust. J. Phys. 46, 601 ∗ [email protected] † [email protected] 1 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 2 A. K. Geim and K. S. Novoselov, Nature Materials 6, 183 (2007). 3 A. K. Geim, Science 324, 1530 (2009). 4 D. Geng, B. Wu, Y. Guo, L. Huang, Y. Xue, J. Chen, G. Yu, L. Jiang, W. Hu, and Y. Liu, PNAS 109, 7992 (2012); J. H. Bardarson, J. Tworzyd lo, P. W. Brouwer, and C. W. J. Beenakker, Phys. Rev. Lett. 99, 106801 (2007). 5 C. Gonz´alez-Santander, F. Dom´ınguez-Adame, M. Hilke, and R. A. Romer, EPL (Europhysics Letters) 104, 17012 (2013). 6 A. Biswas, K.-S. Kim, and Y. H. Jeong, Journal of Applied Physics 116, 213704 (2014). 7 J. Liao, Y. Ou, X. Feng, S. Yang, C. Lin, W. Yang, K. Wu, and Y. Li, ArXiv e-prints K. He, X. Ma, Q.-K. Xue, (2015), arXiv:1504.01847 [cond-mat.mes-hall]. 8 H.-Z. Lu, J. Shi, and S.-Q. Shen, Phys. Rev. Lett. 107, 076801 (2011). 9 P. W. Anderson, Phys. Rev. 109, 1492 (1958). 10 J. T. Edwards and D. J. Thouless, J. Phys. C: Solid State Phys. 5, 807 (1972). (1993). 11 E. Abrahams, P. W. Anderson, D. C. Licciardello, and 28 H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603 T. V. Ramakrishnan, Phys. Rev. Lett. 42, 673 (1979). 12 P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys. 57, 287 (1985). 13 A. MacKinnon and B. Kramer, Phys. Rev. Lett. 47, 1546 (1981); Zeitschrift fur Physik B Condensed Matter 53, 1 (1983); A. MacKinnon, 59, 385 (1985); J. L. Pichard and G. Sarma, Journal of Physics C: Solid State Physics 14, L127 (1981); 14, L617 (1981). 14 B. L. Altshuler, A. G. Aronov, and P. A. Lee, Phys. Rev. Lett. 44, 1288 (1980). (2002). 29 T. A. Brody, J. Flores, J. B. French, P. A. Mello, A. Pandey, and S. S. M. Wong, Rev. Mod. Phys. 53, 385 (1981). 30 D. A. Rabson, B. N. Narozhny, and A. J. Millis, Phys. Rev. B 69, 054403 (2004). 31 T. Hanisch, B. Kleine, A. Ritzl, and E. Muller-Hartmann, Annalen der Physik 507, 303 (1995), cond-mat/9501116. 32 G. Yu, Y. Yong-Hong, W. Yong-Gang, and Z. Qun, Com- mun. Theor. Phys. 43, 743 (2005). 

1711.00079

2

1711

Enhanced Superconductivity and Suppression of Charge-density Wave Order in 2H-TaS$_2$ in the Two-dimensional Limit

As superconductors are thinned down to the 2D limit, their critical temperature $T_c$ typically decreases. Here we report the opposite behavior, a substantial enhancement of $T_c$ with decreasing thickness, in 2D crystalline superconductor 2H-TaS$_2$. Remarkably, in the monolayer limit, $T_c$ increases to 3.4 K compared to 0.8 K in the bulk. Accompanying this trend in superconductivity, we observe suppression of the charge-density wave (CDW) transition with decreasing thickness. To explain these trends, we perform electronic structure calculations showing that a reduction of the CDW amplitude results in a substantial increase of the density of states at the Fermi energy, which contributes to the enhancement of $T_c$. Our results establish ultra-thin 2H-TaS$_2$ as an ideal platform to study the competition between CDW order and superconductivity.

cond-mat.mes-hall

cond-mat

a Enhanced Superconductivity and Suppression of Charge-density Wave Order in 2H-TaS2 in the Two-dimensional Limit Yafang Yang,1 Shiang Fang,2 Valla Fatemi,1 Jonathan Ruhman,1 Efr´en Navarro-Moratalla,3 Kenji Watanabe,4 Takashi Taniguchi,4 Efthimios Kaxiras,2, 5 and Pablo Jarillo-Herrero1, ∗ 1Department of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139 USA 2Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 3Instituto de Ciencia Molecular, Universidad de Valencia, c/Catedr´atico Jos´e Beltr´an 2, 46980 Paterna, Spain 4National Institute for Materials Science, Namiki 1-1, Tsukuba, Ibaraki 305-0044, Japan 5John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, USA (Dated: March 1, 2018) As superconductors are thinned down to the 2D limit, their critical temperature Tc typically decreases. Here we report the opposite behavior, a substantial enhancement of Tc with decreasing thickness, in 2D crystalline superconductor 2H-TaS2. Remarkably, in the monolayer limit, Tc in- creases to 3.4 K compared to 0.8 K in the bulk. Accompanying this trend in superconductivity, we observe suppression of the charge-density wave (CDW) transition with decreasing thickness. To explain these trends, we perform electronic structure calculations showing that a reduction of the CDW amplitude results in a substantial increase of the density of states at the Fermi energy, which contributes to the enhancement of Tc. Our results establish ultra-thin 2H-TaS2 as an ideal platform to study the competition between CDW order and superconductivity. Transition metal dichalcogenides (TMDs) 2H-MX2 (where M = Nb, Ta and X = S, Se) have attracted con- siderable attention as intriguing 2D crystalline supercon- ductors [1]. In these materials, superconductivity (SC) forms in an environment of pre-existing charge-density wave (CDW) order [2, 3], making it an ideal platform to study many-body ground states and competing phases in the 2D limit. In bulk crystals, the reported critical tem- perature of the CDW transition decreases from 120 K in 2H-TaSe2 down to 30 K in 2H-NbSe2. Superconductivity weakens in approximately reverse order, with Tc increas- ing from around 0.2 K in 2H-TaSe2 to 7.2 K in 2H-NbSe2. The relationship between CDW and superconductivity in such systems is still under debate [4, 5]. It is generally believed that such mutual interaction is competitive, but evidence to the contrary, indicating a cooperative inter- action, has also been reported in angular-resolved pho- toemission spectroscopy studies [3]. In TMDs, superconductivity and CDW instability can be investigated by adjusting the interlayer interactions through pressure [6, 7] or molecule intercalation [8, 9]. Recently, mechanical exfoliation has emerged as a robust method for producing ultra-clean, highly crystalline sam- ples with atomic thickness [10]. This offers a useful way to assess the effect of dimensionality and interlayer in- teractions on superconductivity and CDW. A material whose behavior as a function of layer thickness has been recently studied is 2H-NbSe2 [11–13], in which the super- conducting state is progressively weaker in thinner sam- ples, with Tc reduced from 7 K in bulk crystals to 3 K in the monolayer. The thickness dependence of CDW order is still under debate, considering discrepancies between Raman and scanning tunneling microscopy/spectroscopy (STM/STS) studies [14, 15]. Bulk 2H-TaS2, another member of the 2H-MX2 family, exhibits a CDW transition at 70 K and a SC transition at 0.8 K [2, 16–18]. Compared to 2H-NbSe2, 2H-TaS2 mani- fests a stronger signature of CDW transition in transport in the form of a sharp decrease of resistivity [9], and thus serves as a desirable platform to study the thickness de- pendence of CDW instability. STM/STS measurements on monolayer TaS2 epitaxially grown on Au(111) sub- strates show suppression of the CDW instability [19]. Recently, a study observed an enhanced Tc down to a thickness of 3.5 nm, utilizing TaS2 flakes directly exfoli- ated on a Si/SiO2 substrate [18]. Unfortunately, it was found that samples thinner than that become insulating, indicative of its particular susceptibility to degradation in ambient atmosphere. Therefore, exfoliation and encap- sulation in an inert atmosphere become crucial to attain high quality samples. Here we report that superconductivity persists in 2H- TaS2 down to the monolayer limit, with a pronounced increase in Tc from 0.8 K in bulk crystals to 3.4 K in the monolayer. Two transport observations, that in the bulk signal the CDW transition, are found to vanish in ultra-thin samples: (1) a kink in the temperature depen- dence of the resistivity, and (2) a change of sign in the Hall coefficient versus temperature. In search of an ori- gin for such trends, we perform electronic structure and phonon spectrum calculations. We show that suppres- sion of the CDW order leads to a substantial increase in the density of states at the Fermi level, N (EF ), which ultimately enhances Tc. Our observations also motivates consideration of quantum fluctuations of the CDW or- der as another mechanism that boosts Tc. This provides 2 utilizing a polymer pick-up technique [24] as illustrated in Fig. 1 (a), taking advantage of the van der Waals ad- hesion between 2D layers. With this method, we are able to obtain high quality few-layer TaS2 devices. As seen in Fig. 1 (b), when temperature is sufficiently low, a clear superconducting transition is observed for 1, 2, 3, 5, 7- layer, and bulk samples. By fitting the resistance to the Aslamazov-Larkin expression [25], we are able to deter- mine the mean-field superconducting transition tempera- ture Tc. When sample thickness is decreased from bulk to monolayer, the corresponding Tc monotonically increases from 0.9 K to 3.4 K [26]. The trend observed in our ex- periment is strictly opposite that of a previous finding on 2H-NbSe2 [12], where Tc decreases monotonically with thickness reduction, despite the fact that the two mate- rials are isostructural and isovalent. In 2H-NbSe2, the decreased Tc is attributed to a weaker interlayer Cooper pairing given by λinter ∼ cos(π/(N + 1)) as layer number N is reduced. In our case, however, it is surprising to see that even with a reduced interlayer Cooper pairing, the Tc is dramatically enhanced. To verify that the thick- ness dependence of Tc is independent of extrinsic factors (such as level of disorder, substrate, source of crystal, sample quality), we measure an additional set of bilayer and trilayer samples and plot our results alongside previ- ously reported Tc values [18] in Fig. 1 (c). Regardless of different sample preparation procedures and substrates, the trend of Tc versus thickness is consistent between the two sets of experimental results, indicative of an intrinsic origin underlying the enhancement of Tc. To further characterize the superconducting proper- ties of thin TaS2, we characterize the superconducting transition under both out-of-plane and in-plane magnetic fields. Fig. 1 (d) shows the H⊥ c2 as a function of tempera- ture. Close to Tc, the dependence of H⊥ c2 is fitted well by the phenomenological 2D Ginzburg-Landau (GL) model, which yields ξ(0) = 19, 20, 26 nm for 2, 3, 5-layer re- spectively, where ξ(0) denotes the GL coherence length at zero temperature. The reported value for 3D ranges from 22 nm [27] to 32.6 nm [28]. For in-plane field, we (cid:107) c2 = 32 T at 300 mK for a bilayer observe a much larger H (Tc = 2.8 K), which is more than six times the Pauli para- magnetic limit Hp, obeying a square root rather than a linear temperature dependence (inset of Fig. 1 (d)). This (cid:107) c2, often referred to dramatic enhancement of in-plane H as Ising superconductivity, has been observed in other 2D crystalline superconductors [12, 13, 29, 30]. The above observations verify that thin TaS2 behaves as a 2D super- conductor. We also show that the superconducting tran- sition also exhibits the Berezinskii-Kosterlitz-Thouless (BKT) transition as expected in 2D in the Supplemental Material [23]. Additionally, the critical current density increases by orders of magnitude as the devices become thinner (bulk Jc ≈ 700 A/cm2, trilayer Jc ≈ 7 × 105 A/cm2, bilayer Jc ≈ 1.2 × 106 A/cm2). The trend of FIG. 1. (a) Schematic of device fabrication and crystal structure of 2H-TaS2. (b) Resistance normalized by the nor- mal state (R/RN ) as a function of temperature for 1, 2, 3, 5, 7-layer and bulk (d = 40 nm) samples near the SC transition. The superconducting Tc is 3.4, 3.0, 2.5, 2.05, 1.6 and 0.9 K respectively, determined by fitting the tran- sition curve to the Aslamazov-Larkin formula (black solid lines). (c) Tc reported in this work (circles) and in a prior study (crosses) [18]. The dashed line guides the eye to the general trend. (d) Out-of-plane critical field Hc2 for 2, 3, 5-layer samples. The dashed lines are linear fits to c2 = φ0/(2πξ(0)2)(1 − T /Tc), where φ0, ξ(0) denote the H⊥ flux quantum and in-plane GL coherence length at zero tem- (cid:107) perature respectively. Inset: In-plane critical field H c2 nor- malized by Pauli limit (Hp ≈ 1.86Tc) for bilayer and bulk samples. The dashed line for bilayer is a fit to the Tinkham formula for 2D samples H [20]. The purple background indicates the Pauli limit regime. (e) Normalized critical current as a function of T /Tc. Dashed and dotted lines denote the models proposed by Bardeen [21] and Ambegaokar-Baratoff [22] respectively. 12φ0/(2πξ(0)d)(cid:112)(1 − T /Tc) √ (cid:107) c2 = new insights into the impact of reduced dimensions on many-body ground states and their interactions. In this work, we exfoliate and fabricate samples with a transfer set-up built inside a glove box filled with Argon gas, and encapsulate the TaS2 flake between two sheets of hexagonal boron nitride [23]. We build our devices 00.20.40.60.81T/Tc00.20.40.60.811.2Jc(T)/Jc(0)Bardeen2-layer3-layerAmbegaokar-Baratoffbulk123400.51T/Tc02462-layerbulkHc2ǁ/ Hp00.20.40.60.8T (K)Hc2 (T)(d)(e)2L3L5L = 19 nm = 20 nm = 26 nmT in this work prior work (0.5 R )cN(a)(b)(c)TaSSPDMSPC (cid:31)lmhBN2H-TaSCr/PdAr atmosphere2Si/SiO2012345T (K)00.20.40.60.81R/RN1-layer2-layer3-layer7-layerbulk01235701234Tc (K)10203040Thickness (nm)Layer number, N5-layercritical current density versus temperature for represen- tative thicknesses is shown in Fig. 1 (e), with a more detailed discussion in Supplemental Material [23]. In addition to SC transition, bulk 2H-TaS2 is known to manifest CDW order below 70 K. Fig. 2 (a) illustrates the atomic displacements in the CDW state for mono- layer. Two well established indicators of the CDW phase in bulk TaS2 are a kink in the resistivity occurring at TCDW = 70 K and a change of sign in the Hall coeffi- cient as the temperature is reduced below TCDW . We find that both disappear as the sample thickness is re- duced towards the 2D limit. Shown in Fig. 2 (b) is a plot of normalized resistance versus temperature on a linear scale. All samples manifest a linear decrease of resistance at high temperatures, consistent with phonon limited re- sistivity in a normal metal [31]. Below 70 K, the 7-layer and bulk devices undergo a CDW phase transition, pro- ducing a sudden drop in resistance. This is confirmed by calculating the temperature derivative of the resistivity, as shown in the inset of Fig. 2 (b). A peak in dρ/dT de- velops close to the transition. In 2, 3, 5-layer, however, such a sudden drop is not noticeable. The Hall effect has been used to verify the existence of a phase transition. The Hall coefficient is found to demonstrate a broad transition between 70 and 20 K and a change in sign at 56 K in bulk crystals [8]. This indicates that the CDW transition not only induces a structural change, but also alters the electronic proper- ties of the material. It has been shown that a two-carrier model with light holes and heavy electrons is necessary to explain the opposite signs for the Seebeck and Hall coeffi- cients measured above the CDW transition temperature; a single-carrier model describes the low-temperature be- havior [8]. In Fig. 2 (c), we plot the Hall coefficient of three representative thicknesses below 100 K. In the 5- layer device, a significant deviation from the bulk behav- ior is already apparent: the overall magnitude of RH and temperature where it switches sign, are both diminished. However, the most striking fact is the weak temperature dependence and absence of a sign change in the 2-layer sample. This provides unambiguous evidence that the CDW transition occurring in bulk samples is absent in the ultra-thin limit. Another interesting observation is the emergence of R ∼ T 2 behavior in ultra-thin samples. In Fig. 2 (d), we plot the subtracted resistance R − RN on a log-log scale. For bulk, the linear temperature dependence is disrupted by a sudden switch to T 5 near TCDW . R ∼ T 5 has been well known as a consequence of electron-phonon scattering at temperatures lower than the Debye temper- ature ΘD. In contrast, a gradual transition to R ∼ T 2 is observed in 2, 3-layer; it persist from 55 K down to the onset temperature of superconductivity. Naively, such T 2 behavior results from electron-electron scatter- ing within Fermi liquid theory. However, we find that the observed T 2 coefficient is too large to be explained 3 FIG. 2. (a) Illustration of the atom position of Ta and S atoms in the normal phase and the CDW phase. The pe- riodicity of the CDW order is 3 × 3. (b) Normalized resis- tance R(T )/R(250K) for 2, 3, 5, 7-layer and bulk samples, measured while cooling down. Inset: derivative of the re- sistivity ρ = d · R close to the CDW ordering temperature. An arrow is used to mark the TCDW for bulk and 7-layer, which both show a peak in dρ/dT at 70 K. (c) Hall coeffi- cient RH = d· VH /(I · B) measured while cooling down. Data for the bulk crystal is from [8]. (d) Resistance R − RN as a function of temperature plotted in a log-log scale, where RN is the residual resistance just above the onset temperature of superconductivity. For clarity, data for 5-layer is scaled by a factor of 0.8. within such a framework. Moreover, assuming the e-e scattering strength is not greatly altered from 3D to 2D [23], it is implausible that the T 2 behavior is completely absent in thick samples in the same temperature range. Here we propose an alternative mechanism that leads to a T 2 resistivity: scattering of electrons by soft phonons, i.e. critical CDW fluctuations, which can happen close to a finite momentum ordering transition [32] (see Sup- plemental Material [23]). This picture is motivated by the longitudinal and Hall resistivity data indicating the disappearance of the CDW order in thin samples. It as- sumes that although the long range CDW order has been destroyed, strong CDW fluctuations remain, which can scatter electrons. It is worth noting that a similar anti-correlation of trends of SC and CDW transition has also been observed in 2H-TaS2 crystals under pressure [6, 28] and single crys- tal alloys [33–35]. To better understand the connection between CDW and superconductivity, we recall that in McMillan's theory, the critical temperature is expressed (b)(a)(c)(d)1050100200T (K)10-1100101102R-R (Ω)T5T2T1T1T2N2-layer3-layer5-layer×0.87-layerbulkT CDWT (K)RH (10 cm3C-1)5-layer-4bulk, ref [8]2-layer20406080100-5-4-3-2-10123050100150200250T (K)R(T)/R(250 K)2-layer3-layer5-layer7-layerbulk00.20.40.60.811.250100T (K)020dρ/dT (Ω‧nm/K)70 K 40TaTa (CDW)SS (CDW)as [36] Tc = ΘD 1.45 exp[− 1.04(1 + λ) λ − µ∗(1 + 0.62λ) ], (1) where µ∗ is the Coulomb pseudopotential of Morel and Anderson, and λ is the electron-phonon coupling con- stant. Assuming µ∗ = 0.15 as suggested by McMillan [36], one can evaluate the inverted form of Eq. (1) and obtain λ = 0.482 for TaS2 with ΘD = 250 K [37] and Tc = 0.8 K [18], indicating that TaS2 lies in the interme- diate coupling regime. The CDW instability allows electronic systems to lower their energy by inducing energy gaps in the spectrum. Since N (EF ) can affect both µ∗ and λ, it plays an impor- tant role in determining Tc. We investigate the electronic and vibrational properties of 2H-TaS2 based on density functional theory (DFT). First, we implemented in VASP code [38, 39] in order to obtain the DOS in the normal and the CDW phases for monolayer, bilayer and bulk. A comparison of Fig. 3 (b) and (c) reveals an apprecia- ble reduction of DOS near the Fermi level induced by CDW order for all three thicknesses. This is consistent with previous magnetic susceptibility and heat capacity experiments showing a sharp drop of density of states below the CDW transition [40, 41]. Further, to visualize the effects of CDW on the pristine band structure, we derive the unfolded band structure in the CDW phase, shown in Fig. 3 (a), based on the Fourier decomposi- tions of the Bloch wavefunctions from the tight-binding Hamiltonians [42, 43] (see Supplemental Material [23]). It is clearly seen that a band gap, ∆CDW , emerges in the inner pocket around K along Γ-K and K-M. In ad- dition, the saddle point located along the Γ-K, is shifted to energies above the Fermi level. Next, we compute the phonon dispersion for the bulk and monolayer (see Sup- plemental Material [23]). In both the bulk and mono- layer, an acoustic mode that involves in-plane motion of Ta atoms softens and becomes unstable as the electronic temperature is lowered. We found that in both cases the instability occurs at approximately the same wave vector that corresponds to the CDW ordering QCDW ≈ 2/3 ΓM [44]. We then investigate the impact of progressive weaken- ing of the CDW with decreasing thickness by varying the magnitude of atomic distortion. A scaling factor, from 1 to 0, is used to define the fraction by which the magni- tude of the atomic displacement is reduced with respect to the stable distorted configuration. The correspond- ing DOS as a function of atom displacement amplitude is calculated and plotted in Fig. 3 (d). Using the bulk as a starting point, we take into account the change in N (EF ) and a small phonon energy shift calculated for monolayer, and plot the predicted Tc within the McMil- lan formalism in Fig. 3 (e). An enhancement of Tc up to 3.75 K is achieved when the amplitude of CDW goes to zero. This gives a rough estimate of the impact of 4 FIG. 3. (a) Band structure for monolayer 2H-TaS2. The grey lines show the unfolded band structure compared with origi- nal band structure in the normal phase (red lines). (b) Den- sity of states for monolayer/bilayer/bulk in the normal phase. Monolayer in the CDW phase is plotted as a grey shade for reference. (c) Density of states for monolayer/bilayer/bulk in the CDW phase. Monolayer in the normal phase is plotted as a grey shade for reference. (d) A comparison of density of states close to Fermi level for monolayer with various CDW amplitudes ranging from 1 (full amplitude) to 0 (total sup- pression). (e) Left axis: density of states at Fermi level N (EF ) as a function of CDW amplitude for monolayer. Right axis: Tc from Eq. (1) using calculated N (EF ). the suppression of the CDW order on the superconduct- ing Tc, which leads to a reasonable prediction that aligns with the experimental value. We note that it is not en- tirely clear that the enhancement of Tc is solely due to the enhanced DOS. There are several factors impacting Tc that have been discussed previously but not addressed in our calculation, such as substrate effects, the presence of a van Hove singularity near the Fermi level and enhanced electron-phonon coupling due to reduced screening in two dimensions [18], and a weaker interlayer Cooper pairing [12]. This is not the first experiment indicating that a CDW phase transition vanishes in reduced dimensions [8, 19, 45–47]. Its origin is still under debate [5, 45]. Recently, a study showed that lattice fluctuations aris- ing from the strong electron-phonon coupling act to sup- press the onset temperature of CDW order, leading to a pseudogap phase characterized by local order and strong phase fluctuations [48]. This is consistent with our model of presence of soft phonons, or CDW fluctuations [49] as primary contributor to the T 2 behavior of resistivity observed above TCDW . More interestingly, theory pre- dicts that quantum fluctuations caused by proximity to a CDW transition can boost superconducting pairing by providing sources of bosonic excitations [50]. Although there is no direct evidence that CDW fluctuations facili- tate superconductivity in 2H-TaS2, this scenario reveals a potentially rich relationship between CDW and SC. In conclusion, we observe enhanced superconductivity in atomically thin 2H-TaS2 accompanied with suppres- sion of the CDW order. Our electronic band structure calculation shows that suppression of the CDW phase leads to a substantial increase in N (EF ), which acts to boost the superconducting Tc. We further suggest that the emergence of R ∼ T 2 behavior in ultra-thin samples is attributable to the scattering of electrons with soft phonon modes, indicative of critical CDW fluctuations. Future studies of the layer dependence of the CDW order, for example, STM/STS and ultrafast spectroscopy studies, will be essential to understanding both the origin of the CDW and the relationship between CDW and superconductivity. We thank Yuan Cao, Jason Luo and Jiarui Li for ex- perimental help. We also thank Patrick A. Lee, Dennis Huang, Miguel A. Cazalilla and Bertrand I. Halperin for fruitful discussions. This work has been primarily sup- ported by the US DOE, BES Office, Division of Materi- als Sciences and Engineering under Award de-sc0001819 (YY and PJH) and by the Gordon and Betty Moore Foundations EPiQS Initiative through Grant GBMF4541 to PJH Fabrication work (ENM) and theory analysis were partly supported by the NSF-STC Center for In- tegrated Quantum Materials under award No. DMR- 1231319 (VF, SF) and ARO MURI Award W911NF-14- 0247 (EK). This work made use of the MRSEC Shared Experimental Facilities supported by NSF under award No. DMR-0819762 and of Harvards CNS, supported by NSF under Grant ECS-0335765. SF used Odyssey clus- ter of the FAS by the Research Computing Group at Harvard University, and the Extreme Science and Engi- neering Discovery Environment, which is supported by NSF Grant No. ACI-1053575. JR acknowledges the Gordon and Betty Moore Foundation under the EPiQS initiative under Grant No. GBMF4303. A portion of this work was performed at the National High Mag- netic Field Laboratory, which is supported by NSF Co- operative Agreement No. DMR-1157490 and the State of Florida. Growth of hexagonal boron nitride crystals was supported by the Elemental Strategy Initiative con- ducted by the MEXT, Japan and JSPS KAKENHI Grant Numbers JP15K21722 and JP25106006. 5 ∗ [email protected] [1] Y. Saito, T. Nojima, and Y. Iwasa, Nat. Rev. Mater. 2 (2016). [2] A. C. Neto, Phys. Rev. Lett. 86, 4382 (2001). [3] T. Kiss, T. Yokoya, A. Chainani, S. Shin, T. Hanaguri, M. Nohara, and H. Takagi, Nat. Phys. 3, 720 (2007). [4] M. Calandra, I. Mazin, and F. Mauri, Phys. Rev. B 80, 241108 (2009). [5] Y. Ge and A. Y. Liu, Phys. Rev. B 86, 104101 (2012). [6] D. Freitas, P. Rodiere, M. Osorio, E. Navarro-Moratalla, N. Nemes, V. Tissen, L. Cario, E. Coronado, M. Garc´ıa- Hern´andez, S. Vieira, et al., Phys. Rev. B 93, 184512 (2016). [7] C. Chu, V. Diatschenko, C. Huang, and F. DiSalvo, Phys. Rev. B 15, 1340 (1977). [8] A. Thompson, F. Gamble, and R. Koehler Jr, Phys. Rev. B 5, 2811 (1972). [9] J. A. Wilson, F. Di Salvo, and S. Mahajan, Adv. in Phys. 24, 117 (1975). [10] A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). [11] Y. Cao, A. Mishchenko, G. Yu, E. Khestanova, A. Rooney, E. Prestat, A. Kretinin, P. Blake, M. Shalom, C. Woods, et al., Nano Lett. 15, 4914 (2015). [12] X. Xi, Z. Wang, W. Zhao, J.-H. Park, K. T. Law, H. Berger, L. Forr´o, J. Shan, and K. F. Mak, Nat. Phys. 12, 139 (2016). [13] A. W. Tsen, B. Hunt, Y. D. Kim, Z. J. Yuan, S. Jia, R. J. Cava, J. Hone, P. Kim, C. R. Dean, and A. N. Pasupathy, Nat. Phys. 12, 208 (2016). [14] X. Xi, L. Zhao, Z. Wang, H. Berger, L. Forr´o, J. Shan, and K. F. Mak, Nat. Nanotechnol. 10, 765 (2015). [15] M. M. Ugeda, A. J. Bradley, Y. Zhang, S. Onishi, Y. Chen, W. Ruan, C. Ojeda-Aristizabal, H. Ryu, M. T. Edmonds, H.-Z. Tsai, et al., Nat. Phys. 12, 92 (2016). [16] S. Nagata, T. Aochi, T. Abe, S. Ebisu, T. Hagino, Y. Seki, and K. Tsutsumi, J. Phys. Chem. Solids 53, 1259 (1992). [17] I. Guillam´on, H. Suderow, J. G. Rodrigo, S. Vieira, P. Rodi`ere, L. Cario, E. Navarro-Moratalla, C. Mart´ı- Gastaldo, and E. Coronado, New J. Phys. 13, 103020 (2011). [18] E. Navarro-Moratalla, J. O. Island, S. Manas- Valero, E. Pinilla-Cienfuegos, A. Castellanos-Gomez, J. Quereda, G. Rubio-Bollinger, L. Chirolli, J. A. Silva- Guill´en, N. Agraıt, G. A. Steele, F. Guinea, H. S. J. van der Zant, and E. Coronado, Nat. Commun. 7, 11043 (2016). [19] C. E. Sanders, M. Dendzik, A. S. Ngankeu, A. Eich, A. Bruix, M. Bianchi, J. A. Miwa, B. Hammer, A. A. Khajetoorians, and P. Hofmann, Phys. Rev. B 94, 081404 (2016). [20] M. Tinkham, Introduction to superconductivity (Dover, New York, 2004). [21] J. Bardeen, Rev. Mod. Phys. 34, 667 (1962). [22] V. Ambegaokar and A. Baratoff, Phys. Rev. Lett. 10, 486 (1963). [23] See Supplemental Material for further details. [24] J. I.-J. Wang, Y. Yang, Y.-A. Chen, K. Watanabe, T. Taniguchi, H. O. Churchill, and P. Jarillo-Herrero, Nano Lett. 15, 1898 (2015). 6 [25] L. Aslamasov and A. Larkin, Phys. Lett. A 26, 238 Quant. Mater. 2, 11 (2017). (1968). [26] For the monolayer flake, determination of Tc becomes tricky due to electrical shortage to adjacent flakes with different thickness. Detailed analysis and determination of Tc by critical current mapping can be found in Sup- plemental Material [23]. [36] W. McMillan, Phys. Rev. 167, 331 (1968). [37] A. Schlicht, M. Schwenker, W. Biberacher, and A. Lerf, J. Phys. Chem. B 105, 4867 (2001). [38] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 (1996). [39] G. Kresse and J. Furthmuller, Comput. Mater. Sci. 6, 15 [27] Y. Kashihara, A. Nishida, and H. Yoshioka, J. Phys. (1996). Soc. Jpn. 46, 1112 (1979). [28] M. Abdel-Hafiez, X.-M. Zhao, A. Kordyuk, Y.-W. Fang, B. Pan, Z. He, C.-G. Duan, J. Zhao, and X.-J. Chen, Sci. Rep. 6 (2016). [29] J. Lu, O. Zheliuk, I. Leermakers, N. F. Yuan, U. Zeitler, [40] F. Di Salvo, R. Schwall, T. Geballe, F. Gamble, and J. Osiecki, Phys. Rev. Lett. 27, 310 (1971). [41] L. Mattheiss, Phys. Rev. B 8, 3719 (1973). [42] M. Farjam, arXiv:1504.04937 (2015). [43] V. Popescu and A. Zunger, Phys. Rev. B 85, 085201 K. T. Law, and J. Ye, Science 350, 1353 (2015). (2012). [30] Y. Saito, Y. Nakamura, M. S. Bahramy, Y. Kohama, J. Ye, Y. Kasahara, Y. Nakagawa, M. Onga, M. Toku- naga, T. Nojima, et al., Nat. Phys. 12, 144 (2016). [31] M. S. El-Bana, D. Wolverson, S. Russo, G. Balakrishnan, D. M. Paul, and S. J. Bending, Supercond. Sci. Technol. 26, 125020 (2013). [32] R. Hlubina and T. Rice, Phys. Rev. B 51, 9253 (1995). [33] K. E. Wagner, E. Morosan, Y. S. Hor, J. Tao, Y. Zhu, T. Sanders, T. M. McQueen, H. W. Zandbergen, A. J. Williams, D. V. West, and R. J. Cava, Phys. Rev. B 78, 104520 (2008). [34] L. Fang, P. Y. Zou, Y. Wang, L. Tang, Z. Xu, H. Chen, C. Dong, L. Shan, and H. H. Wen, Sci. Technol. Adv. Mater. 6, 736 (2016). [44] O. R. Albertini, A. Y. Liu, and M. Calandra, Phys. Rev. B 95, 235121 (2017). [45] A. Ayari, E. Cobas, O. Ogundadegbe, and M. S. Fuhrer, J. Appl. Phys. 101, 014507 (2007). [46] J. Pan, C. Guo, C. Song, X. Lai, H. Li, W. Zhao, H. Zhang, G. Mu, K. Bu, T. Lin, X. Xie, M. Chen, and F. Huang, J. Am. Chem. Soc. 139, 4623 (2017). [47] M. Yoshida, R. Suzuki, Y. Zhang, M. Nakano, and Y. Iwasa, Sci. Adv. 1, e1500606 (2015). [48] F. Flicker and J. van Wezel, Phys. Rev. B 94, 235135 (2016). [49] M. Naito and S. Tanaka, J. Phys. Soc. Jpn. 51, 219 (1982). [50] Y. Wang and A. V. Chubukov, Phys. Rev. B 92, 125108 [35] L. Li, X. Deng, Z. Wang, Y. Liu, and M. Abeykoon, npj (2015). 

1212.3401

1

1212

1.5 GHz Pulse Generation From a Monolithic Waveguide Laser With a Graphene-Film Saturable Output Coupler

We fabricate a saturable absorber mirror by coating a graphene film on an output coupler mirror. This is then used to obtain Q-switched mode-locking from a diode pumped linear cavity waveguide laser inscribed in Ytterbium-doped Bismuthate Glass, with high slope and optical conversion efficiencies. The laser produces mode-locked pulses at 1039nm, with 1.5GHz repetition rate at an average 202mW output power. This performance is due to the combination of the graphene saturable absorber with the high quality laser glass.

cond-mat.mes-hall

cond-mat

1.5 GHz Pulse Generation From a Monolithic Waveguide Laser With a Graphene-Film Saturable Output Coupler R. Mary1, S.J. Beecher1, G. Brown1, F. Torrisi2, S. Milana2, D.Popa2, T. Hasan2, Z. Sun2, E. Lidorikis3, S. Ohara4, A.C. Ferrari2 and A.K. Kar1 1SUPA, Institute of Photonics and Quantum Sciences, School of Engineering and Physical Sciences, Heriot-Watt University, EH14 4AS, UK 2Department of Engineering, University of Cambridge, Cambridge, CB3 0FA, UK 3Department of Materials Science and Engineering, University of Ioannina, Ioannina, Greece 4Asahi Glass Co., Ltd. Research Center, 1150, Hazawa-cho, Kanagawa-ku, Yokohama, Kanagawa 221-8755, Japan We fabricate a saturable absorber mirror by coating a graphene film on an output coupler mirror. This is then used to obtain Q-switched mode-locking from a diode pumped linear cavity waveg- uide laser inscribed in Ytterbium-doped Bismuthate Glass, with high slope and optical conversion efficiencies. The laser produces mode-locked pulses at∼1039nm, with 1.5GHz repetition rate at an average 202mW output power. This performance is due to the combination of the graphene saturable absorber with the high quality laser glass. Carbon nanotubes (CNTs) and graphene have emerged as promising saturable absorbers (SA) for a variety of applications1–3, opening a new phase in the development of passively Q-switched4 and mode-locked lasers5–15. While the predominantly used semiconductor saturable absorber mirrors (SESAMs) are limited by their narrow wavelength range17, and complex fabrication18, CNTs and graphene have simple, cost-effective production and integration1–16. Broadband operation is achieved with CNTs by combining tubes of different diameters12. How- ever, for a particular wavelength, only CNTs in reso- nance are used, the rest contributing insertion losses6. Graphene has an inherent ultra-wide spectral range due to the linear dispersion of the Dirac electrons1–4,6,7. This, along with the ultrafast recovery time19 and low satura- tion fluence6,11, makes it an excellent SA1,6–8,10,11. Saturable absorption can also lead to a regime of Q- switched mode-locking (QML)20, where the laser out- put consists of mode-locked pulses within a Q-switched envelope20. This arises due to Q-switching instabilities in the cavity20, typically due to long (i.e.>1µs) upper state lifetimes of the gain media in solid state lasers20. These lasers are useful for applications where the pulse energy stored within the Q-switched envelope20 is valu- able, such as nonlinear frequency conversion21, medical applications22, and micromachining23. With the emerg- ing trend in miniaturization of optical devices based on on-chip integration, the development of ultrafast lasers requires a complementary balance between device com- pactness and performance13,14. Ultra-compact high repe- tition rate (>1GHz) lasers are very useful for applications such as nonlinear microscopy24, frequency combs25 and spectroscopy26. The ease of SA integration into a com- pact cavity plays an important role13,14. Lasers employ- ing a waveguide cavity allow device compactness, mean- while emulating the advantages of fiber lasers, such as high beam quality17 and efficient heat dissipation13,14,17. Of the numerous methods for waveguide fabrication, a simple yet reliable technology is ultrafast laser inscription (ULI)27, which utilizes∼100fs focused pulses to induce permanent modifications within a substrate27. Mode- locked ULI waveguide lasers have been demonstrated us- ing CNT-SAs13,14. However, the fiber ring cavity13,14 did not allow its miniaturization, thus high pulse rates. Here we report pulse generation in a compact, ULI in- scribed waveguide laser in Ytterbium-doped Bismuthate Glass (Yb:BG), by using a graphene film (GF) trans- ferred to an output coupler (OC) mirror as a SA. We get mode-locked pulses with 1.5GHz repetition rate and 202mW average output power, with a 48% slope effi- ciency (i.e. rate of output to pump power in excess of the lasing threshold28) and 38% optical-to-optical con- version efficiency (i.e. rate of output to pump power28). The slope efficiency is high compared with that typical of monolithic pulsed waveguide lasers (e.g. 27%)29. A variety of approaches have been used to make graphene-based SAs. For example, graphene-polymer composites, fabricated from dispersions produced by liq- uid phase exfoliation (LPE) of graphite16, have been used to mode-lock fiber lasers at 1.51,7,11 and 2µm8. Films grown by chemical vapour deposition (CVD) with 1 layer15, 1-2 layers9, and non-uniform multi-layers10, have been used to mode-lock solid-state lasers at 2µm9,15 and fiber lasers at 1µm10. Ref.30 used flakes produced by micromechanical cleavage of graphite, with 4-40 lay- ers, for mode-locking of fiber lasers at 1.5µm. LPE graphene-polymer composites4, and flakes (10-40 layers) grown by carbon segregation on SiC31, have been used for Q-switching of fiber lasers at 1.5µm and solid-state lasers at 1µm, respectively. Graphene oxide (GO) was D 100 ) . u . a ( y t i s n e t n I (a) (b) G 2D D' LPE graphene dispersion D+D' 2D' LPE graphene film ) % ( e c n a t t i m s n a r T 80 60 40 20 (a) (b) (c) (d) 2 Quartz Graphene dispersion Graphene film Graphene film on quartz 1500 2500 2000 Raman shift (cm-1) 3000 0 200 400 600 Wavelength (nm) 800 1000 1200 FIG. 1. Raman spectra measured at 514nm of (a) graphene dispersion in SDC-Water and (b) graphene film. FIG. 2. Transmittance of (a) quartz, (b) graphene dispersion, (c) graphene-film; (d) graphene-film on quartz. also used as SA, either as a film in solid-state lasers at 2µm32, or as composite in fiber lasers at 1.5µm2. How- ever, GO is an insulating material with many defects and gap states33, and may not offer the wideband tunability of graphene. Carbon segregation and CVD require high substrate temperatures9,15,16,32, followed by transfer to the target substrate9,10,15. Micromechanically cleaved graphene has high structural and electronic quality2, but is limited in terms of yield, thus impractical for large- scale applications16. LPE has the advantage of scala- bility, room temperature processing and high yield, and does not require any growth substrate16. Dispersions produced by LPE can easily be embedded into polymers composites and integrated into various systems2,16. Here we adopt a novel approach and use LPE graphene in a polymer-free film. This makes it suitable for high- power applications and device miniaturization. The GF-SA is prepared as follows. Graphite flakes (Sigma Aldrich) undergo LPE34 and are dispersed in deionised water with sodium deoxycholate, as for Refs.6 and 8. High Resolution Transmission Electron Microscopy (HRTEM), optical and Raman Spectroscopy are then used to characterize the dispersions. HRTEM shows that the sample consists of∼26% single-,∼22% bi- and∼18% tri-layers8,35, with∼1µm average size. The dispersion then undergoes vacuum filtration via 25nm pore-size fil- ters. This blocks the flakes, while allowing water to pass through, resulting in a GF. This is then placed on an OC mirror, to be used in the laser, and on a quartz plate, for optical characterization, by applying pressure and heat (∼80◦C, to improve adhesion) for two hours, followed by dissolution of the filter in acetone. The film is∼45nm thick, as determined by profilometry. The GF density is∼0.72g/cm3, derived by measuring with a microbalance the filter weight before and after the GF deposition. This is∼3 times smaller than the density of bulk graphite. Raman spectra are acquired at 457, 514, 633nm using a Renishaw InVia micro-Raman spectrometer. Fig.1(a) plots a typical Raman spectrum of graphene flakes in the dispersion. Besides the G and 2D peaks, signifi- cant D and D' bands are also present36,37. We assign the D and D' peaks to the sub-micrometer edges of our flakes38, rather than to a large amount of disorder within the flakes. This is supported by the G peak dispersion, Disp(G)=0.02cm−1/nm, much lower than in disordered carbons39. Fig.1(b) plots the GF Raman spectrum at 514nm. Similar to the individual flakes discussed above, Disp(G) is 0.02 cm−1/nm39. The 2D peak is still single Lorentzian, but∼24cm−1 larger than for the individual flakes. Thus, even if the flakes are multi-layers, they are electronically decoupled and, to a first approximation, behave as a collection of single layers35,40. The ratio of the 2D and G integrated areas, A(2D)/A(G), is at most∼2, thus we estimate a doping∼1.3x1013cm−2 [42], i.e. a Fermi level shift∼4-500meV41,42. Fig.2(b) plots the transmittance of the graphene dis- persion (diluted to 10% to avoid scattering losses at higher concentrations). Using T = e−αlc where l[m] is the light path length, c[gL−1] is the concentration of dispersed graphitic material, and α[Lg−1m−1] is the ab- sorption coefficient, with α ∼1390Lg−1m−1 at 660nm43, we derive c∼0.18gL−1. The peak∼266nm is a signa- ture of the van Hove singularity in the graphene den- sity of states44. Fig.2(a,c,d) plot the transmittance of quartz, pure GF and GF on quartz. The transmit- tance and reflectance at 1039nm (the laser wavelength) are∼59% and∼11% respectively. To estimate the num- ber of graphene layers from these measurements we use the recurrent matrix method, including the correction to the graphene optical conductivity induced by doping45. While pristine graphene absorbs 2.3% per layer, dop- ing, and consequent Pauli blocking, can significantly de- crease this6,46. By comparing our calculations with the data at 1039nm we estimate that our GF consists of∼40 3 FIG. 5. Mode-locked pulse train. flections. Two identical lenses with a 6.2nm focal length couple the pump light efficiently into the waveguide, and a half-wave plate varies the pump polarization. A dichroic mirror with 99% reflection from 1010-1200nm and <2% at the pump is the pump mirror. The mirrors are butt-coupled to either waveguide end using an index matching gel, which also reduces the parasitic Fresnel re- flections at the interfaces49. A dichroic mirror separates the QML output from the residual pump light. The laser operation initiates abruptly at a threshold pump power of 100mW in a self-starting QML regime. The cavity is optimized by adjusting the pump coupling efficiency, pump beam polarization, and GF-SA posi- tion. The mode area on the GF-SA is dictated by that of the waveguide. Using a fast photodiode and a wide- bandwidth oscilloscope, the initial QML repetition rate is measured as 200kHz, with 17mW average output power. Mode-locked pulses at a fundamental repetition rate of 1.514GHz, corresponding to the free spectral range of the cavity, are measured within the Q-switched envelope. Fig.4 shows the QML pulse repetition rate and energy evolution within a single Q-switched envelope. As the launch pump power is increased, the period between the Q-switched pulses reduces, indicating a tendency towards CW mode-locking. At the highest available pump power of 530mW, the Q-switching modulation has a frequency of 0.95MHz, and an average output power of 202mW, corresponding to a pulse energy of 220nJ. This is dis- tributed along the mode-locked pulses existing within the Q-switch envelope. Fig.5 shows a constant mode- locked pulse train behaviour measured on a timescale of 500ps/div. Mode-locking at the fundamental repeti- tion rate is also verified by measuring the rf spectra with a Rigol 1030 spectrum analyzer (Fig.6). The∼4.2MHz spectral width indicates no pure CW mode-locking50, as further shown in the inset of Fig.5. The optical spectrum is given in Fig.7. The spectral bandwidth, corresponding to a pump power of 530mW, FIG. 3. Laser schematic. L1 and L2: coupling lenses; PM: polarization maintaining fiber; GSA: Graphene saturable ab- sorber; OC: output coupler; DM: Dichroic mirror. QML Repetition Rate Pulse Energy 1.0 0.8 0.6 0.4 0.2 240 200 160 120 80 ) z H M ( e t a R n o i t i t e p e R L M Q l P u s e E n e r g y ( n J ) 190 285 380 475 570 Input Power (mW) FIG. 4. Repetition rate, and pulse energy within a single Q-switched envelope as a function of input pump power. layers. Taking into account that its density is∼1/3 of graphite, this number of layers correspond to an overall thickness∼40nm, in good agreement with that measured by profilometry. We note that a 40nm thick undoped and compact GF would absorb 100% of the incident light and be near impossible to saturate, thus the low density and doping of our film are essential for the SA to work. Fig.3 is the schematic of our cavity. We use a 50mm Yb:BG substrate with 1.6x1026m−3Yb3+ dopants and 2.03 refractive index as gain medium. The waveguide is inscribed by focusing the pulses, through a 0.4NA lens, 200µm below the substrate surface, by a master oscillator power amplifier fiber laser (IMRA FCPA µ- Jewel D400) delivering 350fs pulses at 1047nm and 1MHz repetition rate. An automated x-y-z stage translates the sample, thus extending the positive index change at the laser focus to form a waveguide. Low inser- tion loss waveguides with symmetric cross-sections are realised using a multi-scan technique47, inscribed with pulse energies∼50nJ. Previously, highly efficient contin- uous wave lasing was demonstrated from these, with top slope efficiency∼79%48. The pump source is a polarisation-maintaining fiber- coupled diode laser at 976nm, with 530mW maximum pump power, and an angle cleaved fiber to avoid back re- ) m B d ( y t i s n e t n I -20 -30 -40 -50 -60 -70 -80 1.5138 GHz ) m B d ( y t i s n e t n I -40 -50 -60 -70 -80 -90 1.510 Frequency (GHz) 1.515 1.520 1.48 1.50 1.52 1.54 1.56 Freguency (GHz) . ) . u a ( y t i s n e n t I FIG. 6. RF Spectrum measured at the maximum pump. 1.0 0.8 0.6 0.4 0.2 0.0 1038.0 1038.5 1039.0 1039.5 Wavelength (nm) 1040.0 1040.5 FIG. 7. Optical Spectrum. 200 ) W m 150 ( r e w o P t u p 100 t u O e g a r e v A 50 0 0 100 300 200 Input Power (mW) 400 4 500 FIG. 8. Output power. Slope efficiency∼48% (38% optical- to-optical efficiency). The highest output power is 202mW. is 1.1nm. With increasing pump, the spectral peak mi- grates slightly to longer wavelengths. For the maximum input pump power of 530mW, we have an average output power of 202mW. The average output power dependence on the pump is given in Fig.8. The QML waveguide laser has a high slope efficiency of 48%, and a 38% over- all optical-to-optical conversion efficiency. Stable QML pulses are observed over∼24 hours, establishing the good quality of the GF-SA. In conclusion, a monolithic waveguide laser with stable and efficient Q-switched mode-locking was demonstrated using a transferred graphene film to an output coupler. This is a robust, reliable, practical passive mode-locking element, with easy integration in a waveguide cavity. We funding acknowledge from ERC Grant NANOPOTS, EPSRC Grant EP/G030480/1, a Royal Society Wolfson Research Merit Award, The Royal Academy of Engineering, Emmanuel College and King's College, Cambridge. 1 T. Hasan, Z. Sun, F. Wang, F. Bonaccorso, P. H. Tan, A. G. Rozhin, and A. C. Ferrari, Adv. Mater. 21, 3874 (2009). 2 F. Bonaccorso, Z. Sun, T. Hasan, A. C. Ferrari, Nat. Phot. 4, 611 (2010). 3 Z. Sun, T. Hasan, A. C. Ferrari, Physica E 44, 1082 (2012). 4 D. Popa, Z. Sun, T. Hasan, F. Torrisi, F. Wang, A. C. Ferrari, Appl. Phys. Lett. 98, 073106 (2011). 5 D. Popa, Z. Sun, T. Hasan, W. B. Cho, F. Wang, F. Torrisi, A. C. Ferrari, Appl. Phys. Lett. 101, 153107 (2012). 6 Z. Sun, T. Hasan, F. Torrisi, D. Popa, G. Privitera, F. Wang, F. Bonaccorso, D. M. Basko, and A. C. Ferrari, ACS Nano 4, 803 (2010). 7 Z. Sun, D. Popa, T. Hasan, F. Torrisi, F. Wang, E. Kelle- her, J. Travers, V. Nicolosi, and A. Ferrari, Nano Res. 3, 653 (2010). 8 M. Zhang, E. J. R. Kelleher, F. Torrisi, Z. Sun, T. Hasan, D. Popa, F. Wang, A. C. Ferrari, S. V. Popov, J. R. Taylor, Opt. Express 20, 25077 (2012). 9 J. Ma, G. Q. Xie, P. Lv, W. L. Gao, P. Yuan, L. J. Qian, H. H. Yu, H. J. Zhang, J. Y. Wang, D. Y.Tang, Opt. Lett. 37, 2085 (2012). 10 L. M. Zhao, D. Y. Tang, H. Zhang, X. Wu, Q. Bao, K. P. Loh, Opt. Lett. 35, 3622 (2010). 11 D. Popa, Z. Sun, F. Torrisi, T. Hasan, F. Wang, and A. C. Ferrari, Appl. Phys. Lett. 97, 203106 (2010). 12 F. Wang, A. G. Rozhin, V. Scardaci, Z. Sun, F. Hennrich, I. H. White, W. I. Milne, A. C. Ferrari, Nat. Nanotechnol. 3, 738 (2008). 13 G. Della Valle, R. Osellame, G. Galzerano, N. Chiodo, G. Cerullo, P. Laporta, and O. Svelto, Appl. Phys. Lett. 89, 231115 (2006). 14 S. J. Beecher, R. R. Thomson, N. D. Psaila, Z. Sun, T. Hasan, A. G. Rozhin, A. C. Ferrari, and A. K. Kar, Appl. Phys. Lett. 97, 111114 (2010). 15 A. A. Lagatsky, Z. Sun, T. S. Kulmala, R. S. Sundaram, S. Milana, F. Torrisi, O. L. Antipov, Y. Lee, J. H. Ahn, C. T. A. Brown, W. Sibbett, A. C. Ferrari, arXiv:1210.7042 (2012). 16 F. Bonaccorso, A. Lombardo, T. Hasan, Z. Sun, L. Colombo, A. C. Ferrari, arXiv:1212.3319 (2012) 17 O. Okhotnikov, A. Grudinin, and M. Pessa, New J. Phys. 6, 177 (2004). 18 U. Keller, and In E. Wolf, Progress in Optics (Elsevier, Amsterdam, 2004). 19 D. Brida, A. Tomadin, C. Manzoni, Y. J. Kim, A. Lom- bardo, S. Milana, R. R. Nair, K. S. Novoselov, A. C. Fer- rari, G. Cerullo, and M. Polini, arXiv:1209.5729 (2012). 20 C. Hnninger, R. Paschotta, F. Morier-Genoud, M. Moser, U. Keller, J. Opt. Soc. Am. B 16, 46-56 (1999). 21 R. W. Boyd, Nonlinear Optics (Ac. Press, 2008). 22 S. H. Chung, E. Mazur, J. Biophot. 2, 557 (2009). 23 S. Nolte, C. Momma, H. Jacobs, A. Tnnermann,B. N. Chichkov, B. Wellegehausen, H. Welling, J. Opt. Soc. Am. B 14, 2716 (1997). 24 J. Mertz, Curr. Opin. Neurob. 14, 610 (2004). 25 S. A. Diddams, J. Opt. Soc. Am. B 27, B51-B62 (2010). 26 M. P. Moreno, S. S. Vianna, J. Opt. Soc. Am. B 28, 2066 (2011). 27 G. DellaValle, R. Osellame, P. Laporta, J. Opt. A 11, 013001 (2009). 28 C. Grivas, Progr. Quant. Electr. 35, 159 (2011). 29 A. Choudhary, A. A. Lagatsky, P. Kannan, W. Sibbett, C. T. A. Brown, and D. P. Shepherd, Opt. Lett. 37, 4416 (2012). 30 A. Martinez, K. Fuse, S. Yamashita, Appl. Phys. Lett. 99, 121107 (2011). 31 H. Yu, X. Chen, H. Zhang, X. Xu, X. Hu, Z. Wang, J. Wang, S. Zhuang, and M. Jiang, ACS Nano 4, 7582 (2010). 32 J. Liu, Y. G. Wang, Z. S. Qu, L. H. Zheng, L. B. Su, and J. Xu, Laser Phys. Lett. 9, 15 (2012). 33 C. Mattevi, G. Eda, S. Agnoli, S. Miller, K. A. Mkhoyan, O. Celik, D. Mastrogiovanni, G. Granozzi, E. Garfunkel, and M. Chhowalla, Adv. Funct. Mater. 19, 2577 (2009). 34 Y. Hernandez, V. Nicolosi, M. Lotya, F. M. Blighe, Z. Y. Sun, S. De, I. T. McGovern, B. Holland, M. Byrne, Y. K. Gun'Ko, J. J. Boland, P. Niraj, G. Duesberg, S. Krishnamurthy, R. Goodhue, J. Hutchison, V. Scardaci, A. C. Ferrari, J. N. Coleman, Nat. Nanotechnol. 3, 563 5 (2008). 35 T. Hasan, F. Torrisi, Z. Sun, D. Popa, V. Nicolosi, G. Privitera, F. Bonaccorso, and A. C. Ferrari, Phys. Status Solidi B 247, 2953 (2010). 36 A. C. Ferrari, J. C. Meyer, V. Scardaci, C. Casiraghi, M. Lazzeri, F. Mauri, S. Piscanec, D. Jiang, K. S. Novoselov, S. Roth, and A. K. Geim, Phys. Rev. Lett. 97, 187401 (2006). 37 A. C. Ferrari, J. Robertson, Phys. Rev. B 61, 14095 (2000). 38 C. Casiraghi, A. Hartschuh, H. Qian, S. Piscanec, C. Georgi, A. Fasoli, K. S. Novoselov, D. M. Basko,A. C. Fer- rari, Nano Lett. 9, 1433 (2009). 39 A.C.Ferrari, J. Robertson, Phys. Rev. B 64, 075414 (2001). 40 S. Latil, V. Meunier, L. Henrard, Phys. Rev. B 76, 201402 (2007). 41 A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. K. Saha, U. V. Waghmare, K. S. Novoselov, H. R. Krishnamurthy, A. K. Geim, A. C. Ferrari, and A. K. Sood, Nat. Nano. 3, 210 (2008). 42 D. M. Basko, S. Piscanec, and A. C. Ferrari, Phys. Rev. B 80, 165413 (2009). 43 M. Lotya, Y. Hernandez, P. J. King, R. J. Smith, V. Ni- colosi, L. S. Karlsson, F. M. Blighe, S. De, Z. Wang, I. T. McGovern, G. S. Duesberg, and J. N. Coleman, J. Am. Chem. Soc. 131, 3611 (2009). 44 V. G. Kravets, A. N. Grigorenko, R. R. Nair, P. Blake, S. Anissimova, K. S. Novoselov, and A. K. Geim, Phys. Rev. B 81, 155413 (2010). 45 K. F. Mak, M. Y. Sfeir, Y. Wu, C. H. Lui, J. A. Misewich, and T. F. Heinz, Phys. Rev. Lett. 101, 196405 (2008). 46 Z. Q. Li, E. A. Henriksen, Z. Jiang, Z. Hao, M. C. Martin, P. Kim, H. L. Stormer, and D. N. Basov, Nature Phys. 4, 532 (2008). 47 Y. Nasu, M. Kohtoku, Y. Hibino, Opt. Lett. 30, 723 (2005). 48 R. Mary, S. J. Beecher, G. Brown, R. R. Thomson, D. Jaque, S. Ohara, and A. K. Kar, Opt. Lett. 37, 1691 (2012). 49 M. Bass, E. W. Van Stryland, Fiber optics handbook: fiber, devices, and systems for optical communications (McGraw-Hill, 2002). 50 D. v. d. Linde, Appl. Phys. B 39, 201 (1986). 

1907.10058

1

1907

Transport in two-dimensional topological materials: recent developments in experiment and theory

We review theoretical and experimental highlights in transport in two-dimensional materials focussing on key developments over the last five years. Topological insulators are finding applications in magnetic devices, while Hall transport in doped samples and the general issue of topological protection remain controversial. In transition metal dichalcogenides valley-dependent electrical and optical phenomena continue to stimulate state-of-the-art experiments. In Weyl semimetals the properties of Fermi arcs are being actively investigated. A new field, expected to grow in the near future, focuses on the non-linear electrical and optical responses of topological materials, where fundamental questions are once more being asked about the intertwining roles of the Berry curvature and disorder scattering. In topological superconductors the quest for chiral superconductivity, Majorana fermions and topological quantum computing is continuing apace.

cond-mat.mes-hall

cond-mat

Topical Review Transport in two-dimensional topological materials: recent developments in experiment and theory Dimitrie Culcer1, Aydin Cem Keser1, Yongqing Li2, Grigory Tkachov3 1 School of Physics and Australian Research Council Centre of Excellence in Low-Energy Electronics Technologies, UNSW Node, The University of New South Wales, Sydney 2052, Australia E-mail: [email protected] E-mail: [email protected] 2 Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China E-mail: [email protected] 3 Institute of Physics, Augsburg University, 86135 Augsburg, Germany E-mail: [email protected] July 2019 Abstract. We review theoretical and experimental highlights in transport in two-dimensional materials focussing on key developments over the last five years. Topological insulators are finding applications in magnetic devices, while Hall transport in doped samples and the general issue of topological protection remain controversial. In transition metal dichalcogenides valley-dependent electrical and optical phenomena continue to stimulate state-of-the-art experiments. In Weyl semimetals the properties of Fermi arcs are being actively investigated. A new field, expected to grow in the near future, focuses on the non-linear electrical and optical responses of topological materials, where fundamental questions are once more being asked about the intertwining roles of the Berry curvature and disorder scattering. In topological superconductors the quest for chiral superconductivity, Majorana fermions and topological quantum computing is continuing apace. 9 1 0 2 l u J 3 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 5 0 0 1 . 7 0 9 1 : v i X r a CONTENTS Contents 1 Introduction 2 Background 3 Hamiltonians and Kinetics 4 2D Topological Insulators 4.1 CdTe/HgTe/CdTe quantum wells . . . . 4.2 InAs/GaSb . . . . . . . . . . . . . . . . 4.3 Other 2D TI candidates . . . . . . . . . 2 3 7 8 8 8 9 5 3D Topological Insulators 10 5.1 Early experiments on 3D TIs . . . . . . 10 5.2 Electronic properties of 3D TIs . . . . . 11 5.2.1 Aharonov-Bohm effect . . . . . . 11 5.2.2 Weak antilocalisation . . . . . . 12 5.2.3 Quantum Hall effect . . . . . . . 14 5.2.4 Planar Hall effect . . . . . . . . . 15 5.2.5 Topological magnetoelectric effect 16 16 5.3 Novel magnetism with 3D TIs . . . . . . 5.3.1 Breaking time-reversal symme- try in 3D TIs . . . . . . . . . . . 5.3.2 Magnetically doped 3D TIs . . . 5.3.3 Magnetic heterostructures . . . . 5.3.4 Magnetic ordering in TIs: theory 5.3.5 Spin-orbit torque . . . . . . . . . 5.3.6 Anomalous Hall effects . . . . . . 5.3.7 Topological protection . . . . . . 10 Topological weak superconductivity and the fractional Josephson effect. 10.1 Majorana zero modes in a 1D chiral TS 10.2 Fractional Josephson effect. Phe- nomenology . . . . . . . . . . . . . . . . 10.3 Recent theories of the fractional JE. Non-equilibrium dynamics . . . . . . . . 10.4 Equilibrium tests of the fractional JE . . 10.5 Beyond the 4π periodicity . . . . . . . . 11 Unconventional superconductivity 11.1 Mixed-parity superconducting order pa- rameter. Phenomenology . . . . . . . . 11.2 Theory of the mixed-parity proximity effect . . . . . . . . . . . . . . . . . . . . 11.3 Odd-frequency triplet superconductivity 11.4 Nonunitary triplet pairing and charge- spin conversion . . . . . . . . . . . . . . 12 Outlook 1. Introduction 2 35 36 37 38 38 40 40 40 41 43 43 45 The new millennium has witnessed the seemingly inexorable rise of topological phenomena, culminating with the 2016 Nobel Prize in Physics. The term topological materials encompasses a broad range of structures that exhibit topological phases [1, 2], which can be characterised by a topological invariant that remains unchanged by deformations in the system Hamiltonian, e.g. the Z2 invariant, the Chern number, or more generically the Berry curvature, whose integral over the Brillouin zone yields the Chern number. Topological materials include topological insulators (TI) [3, 4], Weyl and Dirac semimetals (WSM, DSM) [5], transition metal dichalcogenides (TMD) [6], which are often termed 2D materials by themselves, graphene subject to a proximity effect [7, 8], and topological superconductors [3]. The distinction is not always clear cut, since some categories overlap: for example, certain dichalcogenides can become topological insulators under appropriate circumstances [9, 10], while others, such as WTe2, can be Weyl semimetals. 2D topological materials offer opportunities that did not exist previously. The electron gas is on the surface and is directly accessible, unlike semiconductor heterostructures, where it is buried. This facilitates excellent electrostatic control over the conduction in the 2D channel, which enables transistor applications, 16 17 18 19 20 22 24 25 6 Valley-dependent phenomena 7 Probing Fermi arcs in Weyl semimetals 26 8 Non-linear electrical response 9 Chiral superconductivity, Majorana edge modes and related phenomena 9.1 Chiral superconductivity. A primer . . . 9.2 Search for intrinsic chiral superconduc- tivity and related phenomena . . . . . . 9.3 TI materials as platform for topological 28 29 30 31 superconductivity and Majorana fermions 32 9.4 Chiral TS with a single Majorana edge mode in QAHI/supercondictor structures 33 CONTENTS 3 as well as accessibility to light for optical applications. The major advantages of TMDs is that they can be made atomically thin and have a direct band gap, making them ideal for optical emitters and detectors. They exhibit strong excitonic effects and piezoelectricity [11]. WSMs and TMDs can have very high mobilities. Many topological materials exhibit very strong spin-orbit coupling, which, under appropriate conditions, yields dissipationless edge state conduction without large magnetic fields, leading to potential transistor applications through the quantum spin and anomalous Hall effects. An example is provided by WSe2, which has an extraordinarily large spin-orbit coupling, a wide direct band gap, and especially a strong anisotropic lifting of the valley degeneracy in a magnetic field, which makes it ideal for accessing the valley degree of freedom. Likewise, in TI spin-orbit torques have taken off spectacularly, especially since the low mobilities of TIs are not a concern as long as large currents can be achieved by increasing the electron density, as was done in Bi2Te3 [12]. Topological superconductivity has seen spectacular growth, with exciting developments in achieving chiral superconductivity, Majorana edge modes the fractional Josephson effect and unconventional Cooper pairing. [13], In direct analogy with the rise of graphene, topological materials have matured into a topological zoo with broad applications across different fields. In this review we emphasise this breadth of interest while bringing out conceptual unifying features such as spin- and pseudospin-charge coupling, the Berry curvature and inter-band effects and their interplay with disorder, and Cooper pairing in topological materials. We concentrate on the most actively researched two- dimensional transport phenomena in TI, TMD, WSM Fermi arcs and TSC, and include highlights from optical studies, since transport and optical responses are intertwined, and are broadly described by linear response theory. They must frequently be considered on the same footing, for example in determining the non-linear optical response, in which a DC shift or rectification current is also generated. The bulk of the review focuses on surface states. We first present a conceptual overview of the subject, followed by a review of transport in non- superconducting topological materials. The theoretical component of this section centres on insights obtained from linear-response theory and extensions thereof, while the experimental component reviews progress in the laboratory. After introducing the model Hamiltonians and outlining the concepts behind linear response and inter-band coherence, we discuss transport in topological insulators, with particular emphasis on weak localisation and anti-localisation, spin-orbit the quantum Hall effect, torques and their relation to the current-induced spin polarisation and spin-Hall transport, magnetoresistance and the anomalous Hall effect. We pay special attention to the continuing controversy surrounding topological protection and transport by the edge states of topological insulators. The anomalous Hall effect has an extension in transition metal dichalcogenides, which have multiple valleys, and exhibit a valley-Hall effect, which is reviewed next. In Weyl semimetals we focus on Fermi arcs and their manifestation in transport. We discuss the non-linear electrical response, a burgeoning field with important unanswered theoretical questions and potential applications in photovoltaics and solar energy. The latter half of the review is devoted to the latest theoretical and experimental developments in chiral superconductivity, Majorana edge modes and related phenomena in topological superconductivity. 2. Background In this section we attempt a conceptual summary of topological materials and Weyl-Dirac physics in a condensed matter context, and transport phenomena in these materials. Our aim is to provide basic explanations for commonly encountered terms in the literature. (1) Degeneracy points. In resonators and crystals, the spectrum can be drawn as a function of wave/crystal momenta. If the energy levels are degenerate at a fixed momentum, a perturbation would lift the degeneracy, a phenomenon called avoided crossings. However in a two band system, the perturbation moves the degeneracy point in momentum space rather than rendering the spectrum gapped. This is because, a 2×2 Hermitian matrix can be decomposed in terms of the identity and the Pauli matrices, H2×2 = a0(k)σ0 + ai(k)σi and for the eigenvalues to coincide viz. E1 = E2 = a0, three functions ai must vanish. If this accidentally happens at the point k∗, then a small perturbation would simply move this point in the Brillouin zone (BZ). The situation was well understood since the early days of quantum mechanics [14]. It was later understood for example that the dispersion looks conic around the accidental degeneracy, dubbed diabolical point [15]. This fact can easily be seen by linearizing Eq. 1 around k∗. For simplicity, if we assume that this cone is isotropic, the Hamiltonian around the degeneracy point behaves either like H = vF σσσ · k, or −vF σσσ· k up to an additive constant. Save for the value of the Fermi velocity vF , these are the chiral and anti- chiral Weyl partners that make up a Dirac fermion, such as a relativistic electron, albeit in the limit of zero rest mass. CONTENTS 4 Dirac/Weyl equation, chirality. Let us briefly explore this analogy with relativistic motion of high energy particles. Dirac equation describes posi- tive/negative energy states describing particles/anti- particles with spin. A particle can possess a spin vector that is parallel/anti-parallel to its orbital motion. This is captured in the helicity operator σσσ · k. Since, mass- less particles travel at the speed of light, this property is Lorentz invariant and equivalent to 'chirality'. Our linearized Hamiltonian is proportional to the chirality operator, hence the sign of Fermi velocity gives the chi- rality. If around the degeneracy point, the cone is not isotropic, we can simply contract or stretch this cone to make it so, and use this straight forward definition. i (kj − k∗ In other words, if we expand ai(k)σi ≈ aj j )σi, the sign of the determinant of aj i doesn't change if we stretch or rotate the cone, hence sgn(det(ai j)) defines chirality. Location of Weyl points in BZ. We should not take this analogy too literal, because crystals can have a lot of different properties while the space-time vacuum is constrained by many symmetries. For example, the excitations in a crystal do not have to obey a Lorentz symmetry, hence the cone around the degeneracy point can be anisotropic and arbitrarily tilted. Moreover, we can have a degeneracy at k∗ +, with positive chirality − (cid:54)= k∗ and a negative chirality counter-part at k∗ +. So unlike the ordinary (relativistic) massless Dirac fermion where the chiral partners exist at the same point, we can have degeneracy points with either chirality at various locations in the Brillouin zone. Total chirality, Nielsen-Ninomiya Theorem. In- deed, every positive chirality degeneracy point must come with its negative chirality counter-part as long as the bands are defined on a periodic structure, that is the BZ. This statement is called the Nielsen-Ninomiya theorem. Instead of a rigorous proof, we will give an in- tuitive explanation. If we cut a 1D dispersion relation curve f (kz), with a constant energy line, we intersect the curve various times. We can convince ourselves that if the band is a smooth curve, which it is, we in- tersect it at even number of points, moreover, at half of these points, the curve has positive slope and at the other half, negative. In three dimensions, suppose that the degeneracy points lie on the kz axis. Linearizing the Hamiltonian in the kx − ky direction looks for ex- ample like, H = f (kz)σz + kxσx + kyσy. Since f (kz) is periodic in BZ, the linearized Hamiltonian around the degeneracy points are kxσx + kyσy ± kzσz, hence have opposite chirality. Weyl points under discrete symmetries. As long as we have center of inversion in the crystal, that is inversion symmetry I, we know where to find the chiral If we have a Weyl point at k∗, say with partners. cone +σσσ· (k − k∗), we must have −σσσ· (k + k∗) at −k∗. These are nothing but the +/− chiral Weyl points. Similar analysis applies if the crystal has mirror planes. Another important discrete symmetry is time reversal T . Intuitively, when the direction of time is reversed, so does that of momentum and spin. Since chirality is spin component in the direction of momentum, it In the condensed matter case, is invariant under T . instead of spin we are talking about the band index or eigenvector of σ, usually called pseudo-spin. In this case, T switches the sign of k and σy which leaves chirality invariant. Therefore if there is T -symmetry, the number of Weyl points must be a multiple of four: for every pair that are T -partners with positive chirality another pair with negative chirality must exist so that the total chirality is zero. We must be more careful, when both inversion and time-reversal symmetries present in the system like, as we discuss below. Dirac points and mass gap instability. In an actual crystal we have many bands. We can argue that the other bands are further away in energy so we can apply the analysis to those two bands that cross each other and treat the rest as conduction and valence bands. But what if, the bands come in completely degenerate pairs? Indeed this is the case, if we have both T (there is no magnetization in any form) and I (there is a center of inversion). For example, if we have an electronic spin up state at k, then the time reversed partner state sits at −k with spin down. Moreover, by inversion symmetry, another state with spin down must sit right on top of the one with spin up at k. There is no surprise here, electrons come with spin partners and if there is no magnetic couplings the spin is treated as dummy, hence there are at least two electrons at any momentum. A perturbation can not open a gap as long as it does not contain spin dependent forces, hence all the analysis so far goes as is, save for a doubling of every Weyl point. However, spin-orbit coupling exists and sometimes very strong in these materials. Therefore spin degeneracy is already lifted. Nevertheless, the existence of IT symmetry still guarantees two-fold degeneracy. This fact is known as Kramer's theorem, that we can informally illustrate as follows. We have already mentioned that, the existence of T , requires that a Weyl point at k∗ in BZ is accompanied by another with the same chirality at −k∗. On the other point at k∗ again! So, after all, in case there is IT , we must have two opposite chirality Weyl fermion sitting on top of each other. This means isolating the 2 × 2 subspace like we did in Eq. 1 is not possible, because a generic perturbation applies on a 4 × 4 subspace that has IT . It turns out that, we can decompose such a matrix in terms of 5 generators instead of the 3 Pauli hand I requires that we have opposite chirality Weyl CONTENTS 5 matrices. Therefore, we need 5 functions, each taking 3 momenta as parameters, to vanish. This is not possible unless enforced by additional symmetry or fine tuning. Once it is achieved, the resulting system is called a Dirac semi-metal, owing to the fact that two chiral Weyl partners at the same momentum point make up a Dirac particle with mass zero. Even when a gap opens, the resulting massive Dirac fermion might not be a trivial insulator. Topological insulator as a Dirac system, gapless boundary modes. We can see what happens to the linearized band structure when Weyl fermions of opposite chirality coincide at the same momentum point. A perturbation can open a mass gap m of either sign that renders the linearized Hamiltonian H = −vF σσσ · (k − k∗) m m vF σσσ · (k − k∗) (cid:18) (cid:19) (2) that is the four component Dirac equation with mass m. The real surprise comes, when we impose a boundary to this material where m, switches sign. If the boundary is defined by z = 0, with m = msgn(z), it turns out that there must be a localized state −(cid:82) z (cid:48) (cid:48) (ψ1(cid:105),ψ2(cid:105))T where ψ1 and ψ2 are 2-spinors that satisfy ψ(cid:105) = e 0 m(z )dz ψ1(cid:105) = −iσzψ2(cid:105) (3) ∗ y) ∗ x) + vF σy(ky − k To solve the Dirac equation, the two-spinor ψ2 must be an eigenstate of Hboundary = vF σx(kx − k This means we have a gapless mode localized at the interface. In general, there is a gapless Hamiltonian, proportional to σσσ(cid:107)·k(cid:107) parallel to the surface. Ordinary insulators, including the vacuum obey the Dirac equation, with a positive mass. Therefore, if a gapless surface mode does not exist on the boundary of an insulator, we call it ordinary, if it does, the crystal is called a topological insulator (TI). This fact has a microscopic interpretation. The bands are nothing but coalesced atomic orbitals. If strong spin orbit coupling reverses the energy order of atomic orbitals, the band gap is inverted, so does the sign of the Dirac mass m. At a boundary with an ordinary insulator, the atomic orbitals should go back to their natural order at which point they have to meet at the same energy, so the gap must close at the boundary. Number of gapless surface mode as a topological invariant. If the number of boundary modes were to exceed 1, we would be able to gap them out in pairs if we have two without breaking T . For example, boundary modes σxkx + σyky ± mσz are both gapped and time reversal partners. Therefore the number of gapless boundary modes is either 1 or 0, which is called the Z2 invariant. Chern insulator. Breaking T at the boundary, say by depositing magnetic impurities on the surface, can gap out the only gapless mode on the surface of a topological insulator. Now the boundary is described by the 2D massive Dirac Hamiltonian (cid:48) )dx (4) (cid:48) 0 m(x ψ(cid:105) = e−(cid:82) x H2D = σxkx + σyky + mσz called the Chern insulator. Just as in the 3D case, there will be a gapless mode at the boundary where the mass switches sign, say at x = 0 from −m to m. Since ky is a good quantum number, the ansatz eikyσy = 1(cid:105) solves the eigenvalue equation. This mode has the dispersion ky, hence is chiral, meaning it only propagates in +y direction and can not backscatter. Indeed it is the boundary mode of the quantum Hall insulator. The above discussion implies that we can obtain quantum Hall effect if we take a spherical topological insulator and pierce it with a magnetic field. The gapless surface modes on the upper and lower hemispheres will acquire gaps with opposite signs due to the magnetic field. A chiral mode will develop on the equator. If we put terminals on two antipodal points on the equator and pass current between them, being chiral, only half of the equator will carry the current. The excess electrons compared to the other half, leads to a transverse voltage. We can compute the ensuing Hall conductivity by counting the excess charge in a 1D channel. If the current is I, then N = I/(evF L) additional electrons populate one side of the equator with length L. The difference in transverse chemical potential due to this many excess electrons is found by using the dispersion relation vF δk = hvF N L = hI/e = e∆V . This means the Hall conductivity is σxy = e2/h. This is the quantum anomalous Hall effect (QAHE). The time reversal breaking is usually achieved by magnetic impurities deposited on the surface of real topological insulators. Topological state of graphene, Haldane model. We also have a strictly 2D realization of the 2D Dirac Hamiltonian, that is graphene. Graphene comes with two copies of Dirac fermions due to having two high symmetry points in the BZ, excluding the spin degeneracy. These can acquire equal and opposite mass gaps when T is intact but I is broken, say due to sublattices being at different energy. The resulting system is an ordinary insulator. If on the other hand, T is broken, as in Haldane model [16], the two fermions acquire the same mass, the resulting becomes a Hall insulator, just like the surface of a TI with broken T . However, in actual graphene, there is an additional spin degeneracy. If a spin-orbit coupling exists, and T is intact, we have two copies of the Haldane model that are time-reversals of each other. The chiral edge modes then counter propagate and therefore the Hall conductivity vanishes. However, CONTENTS 6 since these modes have opposite momentum, and spin is locked to momentum, in the presence of electric field, they create a net spin current and produce the quantum spin Hall effect (QSHE) [17] Fermi arcs. The relation between the bulk band structure and the boundary modes is one of the key points in the study of topological systems. These modes due to their protected structure has unusual transport signatures and lead to precise quantization of transport coefficients such as Hall conductivity. In addition to the gapless surface mode of a topological insulator and the chiral edge modes of a gapped surface mode, we also have chiral modes due to Weyl fermions called Fermi arcs. If the Weyl fermion comes incident to a boundary, say x = 0 from the left, it reflects with momentum in −x-direction. However, since chirality of a Weyl fermion is fixed, the spin in x-direction must flip as well. This means the Weyl fermion at the boundary can not have spin in x, hence is an eigenstate of cos(α)σy + sin(α)σz. If we assume α = 0, and that we have a positive chirality Weyl fermion at k = 0, we get the same chiral solution as in the Chern insulator, ψ+ = exp(−kzx+iky)σy = +(cid:105) but this time the decay constant into the bulk is kz < 0. If we do the same analysis for the Weyl fermion with opposite chirality, we find that it is ψ− = exp(kzx + iky)σy = +(cid:105), so this time, kz > 0 so that the wave decays into the bulk in x < 0. As kz → 0, the decay constant approaches zero and the surface state becomes a bulk mode. Since Weyl points come in chiral partners located at k∗ ±, the surface mode has kz < k∗ −, hence a 'Fermi arc' in the BZ. Just as in the Chern insulator, these chiral modes contribute to Hall transport. Since now there are az(k∗ −)/2π modes available, if the sample has size az in z-direction, the 3D quantum anomalous Hall conductivity is σxy = ∆ke2/(2πh), where ∆k is the separation of nodes in BZ. + and kz > k∗ +−k∗ All this looks quite curious, but why exactly is it that the boundary and bulk properties of certain modes are related and in any way relevant to transport properties of actual materials with so much other complications such as disorder and interactions? Berry Curvature. To answer this question, we simply write down the equation of motion for a wave packet of electrons moving in the lattice. The Lorentz force due to external electromagnetic field steers this packet around the crystal, more generally in the phase space. However, the Hamiltonian of the electron depends on the position of the packet in the phase space. This means we are dealing with a situation where the Hamiltonian is adiabatically changed through a parameter, in this case momentum, as the wavefunction evolves, which was considered by Berry. The wavefunction acquires a phase at every point in parameter space to account for the changes in the Hamiltonian, called the Berry phase. This immediately produces a gauge field defined on the parameter space, which is BZ in a crystal. After all, what is electromagnetic field? It is a manifestation of wavefunctions acquiring different phases at every point of space-time. Now it also acquires local phases in momentum space, hence there must be an analogous term in the equation of motion (5) (6) x = ∇kε − k × ΩΩΩ, k = − e(E + x × B) The electric-like component is due to the phase that is local in time, that is the dynamical phase e−iεt a wave- function acquires during its evolution. The spatial part of phase is Ai = −i(cid:104)u∂kiu(cid:105) where u is the periodic part of the Bloch wave. Therefore the arising magnetic-like field is ΩΩΩ = ∇k × A (7) (8) (9) (cid:73) Ω · dS = ±1 Berry Curvature due to Weyl points. Now we see that the BZ structure directly influences the motion of electron, we can calculate the Berry curvature Ω due to a Weyl node. We find that it satisfies 1 2π when the surface surrounds a +/− chirality Weyl point respectively. The fact that Weyl points must come in chiral partners imply that the if we take a surface that contains all Weyl points, or the BZ itself, the result is zero. This means that +/− chirality Weyl nodes are sources/sinks of Berry curvature. The integral of Berry curvature over the BZ is called the Chern number and turns out to be an integer topological it measures how many times the pseudo-spin vector wraps around the unit sphere as we traverse across the whole BZ. At an interface between two crystals with different Chern numbers, the net chirality of boundary modes must match the difference in Chern numbers, a fact commonly referred to as the 'bulk boundary correspondence' [18, 19]. Berry Curvature and Chern invariant. In the two band case, invariant. Topologically protected transport properties. More- over, transport properties can be expressed in terms of the Chern number, because computation involves a summation over occupied states. In an insulator, this means integrating over the completely filled band over the BZ. Specifically in the Chern insulator, this inte- gral turns out to be the Chern number. Disorder and interactions can not destroy this property as it is due to filled states [18, 19]. CONTENTS 3. Hamiltonians and Kinetics (10) topological x − 3kxk2 y), The surface states of insulators are described by the following Dirac-Rashba Hamiltonian: HT I = A (sxky − sykx) + λsz(k3 where A and λ are material-specific parameters, s is the vector of Pauli spin matrices, and k is the two- dimensional wave vector. The hexagonal warping term is particularly strong in Bi2Te3. These states reside on opposite surfaces of a three-dimensional slab, yet usually all surfaces have topological states, which in Hall transport in particular mean that current can flow around the edges. The limitations of effective k · p methods for thin films are discussed in [20], where it is pointed out that open boundary conditions may yield different conclusions regarding the edge states than ab- initio calculations. Dirac and Weyl semimetals are described as different cases of the following Hamiltonian [5], HDW SM = vτxσ · k + m τz + b σz + b (cid:48) τzσx, (11) where σ represents a pseudospin degree of freedom, τ the valleys or nodes, m is a mass parameter, and b and b0 are effective internal Zeeman fields. Dirac semimetals have a single node and correspond to the case m = b = b(cid:48) = 0, while the simplest model of a Weyl semimetal has two nodes that arise in the case b > m and b(cid:48) = 0, that act as source and sink of Berry curvature. In known materials, there are typically many more pairs of nodes. If the Weyl points arise due to broken inversion symmetry with respect to Mx, My mirror planes, there are 4 pairs of Weyl nodes, (±kx,±ky,±kz) including time reversal partners. There are as many as eight pairs in TaAs family due to the additional C4 rotational symmetry (±ky,±kx,±kz) [21]). separation between a Weyl node and its partner with opposite chirality determines the electromagnetic response of the material that contains the so called 'axion term'. Such a term is the source of chiral magnetic effect (CME) and the quantum anomalous Hall effect (QAHE) in Weyl semimetals [22, 23, 24]. The Transition metal dichalcogenides are described by the following generalised Hamiltonian [6]: ∆ 2 σz − I 2 σz − λ τ sz, (12) HT M D = A (τ σxkx + σyky) + where s represents spin, τ = ± is the valley, and σ is an orbital pseudospin index, analogous to the sublattice pseudospin encountered in graphene. The spin splitting at the valence band top caused by the spin-orbit coupling is denoted by 2λ. Additional terms encapsulate the spin splitting of the conduction band, the electron-hole asymmetry and the trigonal warping of the spectrum [25, 26]. The trigonal warping term has the form: 7 κ 2 (σ+k2 + + σ−k2−), (13) HT W = where σ± = σx ± i σy. There is always a gap between the valence and conduction bands, which is manifested in the mass appearing in each of the copies of the Dirac Hamiltonian. Consequently, to satisfy time reversal invariance, the materials have two valleys, which are related by time reversal. Interaction terms coupling TMDs to external electromagnetic fields are covered in [27]. Linear response calculations traditionally employ standard, well-established transport theory techniques such as the Kubo and Keldysh formalisms or the semiclassical wave-packet approach combined with the Boltzmann equation. Yet the presence of spin- and pseudospin-charge coupling in topological material Hamiltonians causes electromagnetic fields to induce inter-band dynamics, which have a subtle interplay with disorder. Such effects are most clearly seen in the density-matrix theory [28]. The single-particle density matrix ρ obeys the quantum Liouville equation dρ dt where H is the total Hamiltonian. The density matrix is decomposed into two parts: one part, denoted by (cid:104)ρ(cid:105), is averaged over impurity configurations, while the remainder, which is eventually integrated over, is denoted by g: ρ = (cid:104)ρ(cid:105) + g, with (cid:104)g(cid:105) = 0. In linear response to an electric field E the density matrix comprises equilibrium and non-equilibrium components (cid:104)ρ(cid:105) = (cid:104)ρ0(cid:105) + (cid:104)ρE(cid:105), where (cid:104)ρE(cid:105) is the correction to the equilibrium density matrix (cid:104)ρ0(cid:105) to first order in E. The kinetic equation is linearised with respect to E: i  [H, ρ] = 0, (14) + (cid:48) (cid:26) (cid:27) , (cid:48) (cid:48) (cid:48) (cid:48) = k + ∂k (15) k )] δmm k ) − f0(εm + J((cid:104)ρE(cid:105))mm [f0(εm i  [H0,(cid:104)ρE(cid:105)]mm (cid:48) ∂f0(εm k ) + iRmm d(cid:104)ρE(cid:105)mm dt eE  · where H0 is the band Hamiltonian, J((cid:104)ρ(cid:105)) is a generalised scattering term, f0(εm k ) the Fermi-Dirac distribution, and Rmm ∂k (cid:105) is the Berry connection. Its appearance in the driving term gives rise to the Berry curvature intrinsic contribution to the Hall conductivity of systems with broken time reversal symmetry, and to other response properties. The Fermi occupation number difference factor makes it evident that this term drives off-diagonal response and therefore inter-band coherence contributions to the electrical response. k i ∂um(cid:48) = (cid:104)um k k (cid:48) This formulation is exactly equivalent to the quantum Boltzmann equation and has shed light on CONTENTS 8 the physical origin of the chiral anomaly of Weyl semimetals [29]. It shares a similar philosophy with the Keldysh method but without requiring an Ansatz for the Keldysh component and integration over an additional energy variable, which become opaque in complex, multi-band systems. In this approach one can immediately separate intrinsic effects, extrinsic effects, and effects that combine interband coherence and disorder. A similar, Boltzmann equation-based theory gives a full account of phonon physics [30], while an related approach being developed at present [31] relies on the semiclassical transport framework. It will be interesting to see what insight can be gained from the density matrix theory in describing localisation physics, which received tremendous attention in the first half of this decade [32, 33, 34, 35, 36, 37, 38, 39] and interaction physics in topological materials [40]. The theory is being generalised to second order in the electric field [41] where, in addition to the Berry curvature dipole term to be discussed below, additional disorder-mediated corrections to the non-linear Hall tensor were identified that have the same scaling in the impurity density. quantization of edge conductance is much less precise than the quantum Hall effect [47]. This was attributed to spin dephasing [48] or inelastic backscattering processes related to charge puddles in the bulk [49, 50]. For a sufficiently long QSH edge channel, the transport is no longer in the ballistic regime, and the longitudinal resistance increases linearly with channel length L, namely R ∼ h , where Lϕ is the spin dephasing/inelastic backscattering length, usually on the order of micrometer. L Lϕ 2e2 The edge channels in the inverted HgTe quan- tum wells were imaged with a scanning microwave impedance probe technique, and the edge conduction was however found to vary very little for magnetic fields up to 9 T [51]. The robustness of edge transport was also observed in a nonlocal electron transport exper- iment. Using samples with voltage probes separated up to 1 mm, Gusev et al. found that the nonlocal re- sistance is of the order 100 kΩ and insensitive to the strength of the strength of in-plane magnetic field for B < 5T [52]. Such low resistances are quite surprising, given that the inelastic scattering length is on the order of micrometer in HgTe/CdTe edge channels [45, 46]. 4. 2D Topological Insulators 4.2. InAs/GaSb In this section, we review the experimental work on two-dimensional (2D) topological insulators (TIs) that are protected by the time reversal symmetry and can be characterized by topological invariant Z2. First identified by Kane and Mele, this type of 2D TIs can be manifested as the quantum spin Hall effect (QSHE) in transport[42, 43], and thus often referred to as quantum spin Hall (QSH) insulators. 4.1. CdTe/HgTe/CdTe quantum wells Following the prediction of Bernevig et al. [44] that the HgTe/CdTe quantum well with inverted band structure could be a 2D TI, Konig et al. reported the evidence of QSHE in this system in 2007 [45]. As shown in Fig. 1, the four-terminal resistances of micrometer-scale Hall-bar shaped HgTe/CdTe samples in the inverted regime is approximately 2h/e2, a value expected for the helical edge transport in the ballistic regime. In contrast, for a sample with the thickness of the HgTe layer less than 6.3 nm, the critical thickness for band inversion, the resistance increases to the order of mega-ohm, consistent with the insulating behavior expected for the topologically trivial regime. Further evidence for the QSHE in HgTe quantum wells was obtained with non-local transport in micrometer-sized devices of various geometries [46]. The experimental results are in agreement with the Landauer-Butikker theory adapted for the edge transport in 2D TIs. However, even for micrometer-sized QSH samples, the InAs/GaSb quantum wells were also predicted to be a 2D TI candidate [53]. In 2011, Knez et al. reported the evidence for edge transport in this system, despite that the parallel bulk conduction had to be subtracted from the total conductance [54]. Du et al. [55] subsequently managed to use Si-doping to suppress the bulk conductivity and observed quantized edge channel conductance with a precision level of ∼ 1%. The edge channel conductance in this sample was found to remain quantized for in-plane magnetic fields up to 12 T and in a wide range of temperatures (20 mK to 4 K). The insensitivity of the edge channel conductance to strong magnetic fields is puzzling at the first sight, since the Zeeman energy is expected to open a gap, leading to a destruction of the helical edge states. This phenomenon was explained recently by two groups, who pointed out that the small energy gap generated by the Zeeman energy is buried in the bulk valence band, and hence the helical edge transport is barely influenced by the magnetic field [56, 57]. Similar argument was also used to explain the robustness of edge transport in HgTe quantum wells [57]. It should be noted, however, if the Dirac point is indeed buried in the bulk bands, this would be are detrimental to pursue the Majorana physics (see Secs. 9 and 10 for a review of topological superconductivity), despite that pronounced superconducting proximity effect has been demonstrated in a HgTe/CdTe QSH system [58]. Nichele et al. reported that the transport CONTENTS 9 Figure 1. Observation of the quantum spin Hall effect in HgTe/CdTe quantum wells. (a) Band gap Eg of HgTe/CdTe quantum well as a function of d, the thickness of the HgTe layer. Band inversion (negative Eg) takes place when d > dc = 6.3 nm. The inset shows the evolution of several subbands of the quantum well. (b) Four-terminal resistances of four Hall-bar shaped HgTe/CdTe quantum well samples (I-IV). Sample I is in the normal regime (d = 5.5 nm), whereas samples II-IV are in the inverted regime (d = 7.3 nm). The sizes of the samples II, III and IV, given in the formats of L × W , are 20 × 13.3 µm2, 1 × 1 µm2, and 1 × 0.5 µm2, respectively. The definitions of L and W are given in the upper-right inset. Adapted from Konig et al., Science 318, 766 (2007) [45]. properties of micrometer-sized InAs/GaSb wells in the non-inverted regime are phenomenologically similar to those observed in the inverted regime [59]. The downward band bending of the InAs conduction near the sample edge was proposed as a possible origin of the topologically trivial edge states [59]. The existence of such trivial edge states was further manifested in counterflowing edge transport in the quantum Hall regime [60]. Shojaei et al. recently measured the temperature and magnetic field dependences of a dual- gated InAs/GaSb quantum well, and concluded that the small hybridization gap (a few meV) in the inverted regime is overwhelmed by the disorder effect, and the transport is thus similar to a disordered two- dimensional metal of symplectic class [61]. 4.3. Other 2D TI candidates In addition to aforementioned semiconductor het- erostructures, many 2D materials have been identi- fied theoretically to be 2D TIs, such as monolayers of Si [62], Ge [62], Sn [63], Bi [64], ZrTe5 [65] and WTe2 [66], Bi bilayers [67], as well as many hybrid structures based on graphene [68]. A lot of experimen- tal efforts were devoted to single-element TI candidates (e.g. silicene, germanene, stanine [69]), which were prepared mostly with molecular beam epitaxy (MBE) and characterized in situ with scanning tunneling mi- croscopy (STM) or angle resolved photoemission spec- troscopy (ARPES). Evidence for the existence of edge channels has been obtained with STM measurements of Bi bilayers grown on Bi2Te3 thin films [67] and on Bi crystals [70]. In a subsequent ARPES experiment on Bi bilayers, however, very large Rashba spin-splitting was observed in the edge states, suggesting a topologically- trivial origin [71]. In spite of a lot of progress, most of these single- element films have so far been grown on conducting substrates, and thus are unsuitable for elucidating their topological nature with transport studies. Recently, Reis et al. [64] reported that honeycomb-structured Bi monolayers can be epitaxially grown on insulating SiC substrates. By using scanning tunneling spectroscopy (STS) measurements, they observed an energy gap of 0.8 eV, and also obtained evidence for conducting edge states. With help of the first principles calculations, Reis et al. claimed that the large gap arises from covalent bonding between the Bi orbitals and the substrate and it is topologically nontrivial [64]. If the QSHE is confirmed in the future, this system will be exceptionally attractive for exploiting the helical edge transport at high temperatures. An alternative approach to obtain 2D TIs is to exfoliate layered compounds with strong spin- orbit interaction, such as WTe2 [66, 72, 73, 74] and ZrTe5 [65, 75, 76]. Following a theoretical prediction of the 1T' phase of transition metal dichalcogenide (TMDC) monolayers being 2D TIs [66], much work has been done on WTe2 monolayers, the only member of the MX2 family (M=W, Mo, and X=Te, Se, S) in which the 1T' phase is energetically favored [72]. An ARPES experiment performed by Tang et al. showed that WTe2 monolayer has a bulk energy gap of 55 meV, much larger than those in HgTe/CdTe and InAs/GaSb quantum wells. The existence of edge channels in WTe2 monolayers has been confirmed CONTENTS 10 for lower with the STM [72, 73], transport [74], and scanning microwave impedance probe measurements [77]. More recently, Wu et al. [78] reported the edge conductance values consistent with the QSHE. In this experiment, however, the observation of quantized edge transport requires a spacing of 100 nm or the neighboring voltage probes in the devices based on WTe2 monolayers. On the other hand, the edge conductance varies very little for temperatures below 100 K. These features are quite different from those of the HgTe/CdTe [46] and InAs/GaSb [55] quantum wells, in which the quantized edge conductance is limited to liquid helium temperatures, but can survive for the channel lengths of the micrometer scale. It remains to be understood why the edge conductance in WTe2 monolayers is so robust against thermal agitation, but very demanding on the channel In addition to WTe2, the layered compound length. ZrTe5 has also attracted a lot of interest. This can be attributed to its controversial topological classification, very low carrier density, and high carrier mobility [65, 75, 76, 79]. The latter features make ZrTe5 particularly suitable for seeking exotic quantum transport properties. Recently, Tang et al. observed a 3D quantum Hall effect in ZrTe5 bulk single crystals [79]. Although ZrTe5 can be easily exfoliated down to nanometer thicknesses, transport measurement of ZrTe5 monolayer, a 2D TI candidate with a band gap of 0.1 eV [65], has not been reported so far, probably due to the chemical instability of this compound. 5. 3D Topological Insulators The 3D counterpart of the 2D TIs can be characterized by four Z2 indexes, {v0, (v1, v2, v3)}. This type of 3D TIs are also protected by the time-reversal symmetry, and can be divided into strong and weak 3D TIs, corresponding to v0 = 1 and 0, respectively [80]. The latter can be regarded as a stack of 2D TIs and therefore have gapless states on the side surfaces, while the former does not have a 2D counterpart. The strong 3D TIs have so far received much more attention than the weak 3D TIs. In this review, we therefor solely focus on the strong 3D TIs, and refer to them as 3D TIs or TIs for brevity. 5.1. Early experiments on 3D TIs Early experimental efforts on 3D TIs were mainly focused on confirming the existence of the helical surface states. ARPES measurements played a decisive role in identifying 3D TIs [80, 81, 82]. Among them, Bi1−xSbx is the first material confirmed to be a 3D TI [83]. The electronic structure of Bi1−xSbx is, however, very complicated due to the coexistence of multiple bands in the surface states. This makes Bi1−xSbx not suitable to be studied as a model 3D TI system. Experimental attention was quickly shifted to the compounds in tetradymite family (Bi2Se3, Bi2Te3, Sb2Te3 and their derivatives), in which the surface states are featured by a single Dirac cone, accompanied by a large bulk band gap [83, 84, 85, 86] (See Fig. 2). Among them, the binary compounds (e.g. Bi2Se3) can be prepared in the form of bulk single crystals, thin films, nanobelts or nanoplates, but they tend to have a high density of defects, and consequently the bulk remains conducting even at liquid helium temperatures [82, 87]. Synthesis of ternary or quaternary compounds turned out be an effective approach to suppress the bulk conductivity. One example is (Bi1−xSbx)2 Te3, in which the chemical potential can be continuously tuned from the bulk conduction band to the valence band by increase the Sb concentration [88, 89]. This doping scheme is based on the fact that undoped Bi2Te3 is often n-type and Sb2Te3 is p-type. The other example exploiting the compensation effect is (Bi1−xSbx)2 (Te1−ySey)3, in which the composition control can increase the bulk resistivity to the order of ohm·cm for single crystal samples [90]. In addition to ARPES, STM was also used Quasiparticle extensively in the 3D TI studies. interference experiments provided the evidence for the suppression of backscattering [93, 94, 95], a property arising from the spin-momentum locking (and π Berry phase, see Section I, in the surface states. The Landau level spectra obtained with STS measurements are consistent with the linear dispersion of surface Dirac fermions [96]. STM/STS measurements also produced atomic scale information of various defects in 3D TIs [97]. Such information is valuable for gaining in- depth understanding of the physics of defect formation, and thus offers new opportunities for further improving the sample quality. Additional evidence for the π Berry phase in the surface states was also gained from the Shubnikov- de Haas The longitudinal conductivity in high magnetic fields satisfies (SdH) oscillation measurements. + 1 2 − γ 2π B ) 2π (16) = SF 4π2 ∆σxx ∝ cos h where BF = 1 e is the frequency of SdH ∆( 1 oscillations. Here, SF is the extremal area of the Fermi spheroid perpendicular to the magnetic field B for a 3D system (and in 2D, it is reduced to the area inside the Fermi circle), and γ is the Berry's phase, which takes a value of π for a TI surface and 0 for an ordinary 2D electron system. The γ value can be extracted from the plot of Landau level index n as a function 2π , where Bn is the magnetic of , i.e. n = F 1 Bn Bn − γ (cid:20) (cid:18) BF B (cid:19)(cid:21) , CONTENTS 11 Figure 2. Crystalline and electronic structures of the Bi2Se3 family of 3D TIs. (a) Crystal structure of Bi2Se3, which has a layered structure of Van der Waals type, which can be separated in to quintuple layers (QLs) of Se-Bi-Se-Bi-Se. Each QL is about 1 nm thick. (b,c) Band diagrams of Bi2Se3 (b) andBi2Te3 (c) given by first principles calculations. The gapless surface states are also shown. (d,e) Band structures of Bi2Se3 (d) and Bi2Te3 (e) obtained with ARPES measurements. Adapted from Zhang et al., Nat. Phys. 5, 438 (2009) [84]; Xia et al., Nat. Phys. 5, 398 (2009) [91]; Chen et al., 325, 178 (2009) [92]. field corresponding to the nth conductivity minimum. SdH measurements performed on Bi2Se3, Bi2Te3 and Bi2Te2Se single crystals yielded nontrivial values of Berry's phase for the surface states [98, 99, 85, 86, 90]. However, caution has to be taken in order to avoid inaccurate assignment of longitudinal conductivity minima [100], and to account for nonlinearity in the energy-momentum relation due to broken particle-hole symmetry, as well as the Zeeman effect. Nevertheless, SdH measurements can provide valuable electronic parameters relevant to the device applications, such as the carrier concentration, effective mass, and quantum lifetime. 5.2. Electronic properties of 3D TIs 5.2.1. Aharonov-Bohm effect The interplay between the Aharonov-Bohm (AB) phase, the Berry's phase and quantum confinement effect makes TI nanostruc- tures very appealing for exploring novel mesoscopic physics. In 2009, Peng et al. observed that when a magnetic field is applied parallel to the long axis of Bi2Se3 nanowires, the conductance exhibits oscillations with a period of h/e, with the maxima appearing at in- teger multiples of the flux quantum (Φ0 = h/e) [101]. These features are in contrast to the Altshuler-Aronov- Spivak oscillations [102] observed in hollow metallic cylinders [103] and carbon nanotubes [104], in which the interference between various time-reversed paths leads to h/2e-periodicity oscillations for the diffusive transport. In the diffusive regime, the h/e-period Aharonov-Bohm oscillations, which would be most pro- nounced in a single ring-like structure, are expected to be suppressed in a nanowire due to the ensemble av- erage of various electron trajectories, which carry ran- dom phase factors [105]. According to the theory reported in Refs. [106, 107], the h/e oscillations observed in the TI nanowires [101] can be regarded as a transport hall- mark of the helical surface states in 3D TIs. The π Berry's phase modifies the 1D quantization condi- tion around the perimeter of nanowires, leading to a gap opening at the Dirac point when the magnetic flux through nanowire is even multiples of half-flux- (cid:1), and a gapless 1D mode for odd quantum (cid:0) Φ0 2 = h 2e numbers of half-flux-quantum [106, 107]. In the ballis- tic regime, this peculiar electronic structure results in the h/e oscillations with conductance maxima (min- ima) located at odd (even) multiples of Φ0/2. The experimental data reported in Ref. [101], however, ex- hibit the opposite behavior. This was explained by Bardarson et al. [107], who considered the joint effects of doping and disorder, and showed that for certain disorder strengths, the conductance maximum can os- cillate between the zero flux and half flux quantum as the chemical potential varies. This phenomenon was subsequently observed in Bi2Se3 nanowires by Jau- regui et al. [108]. It was also predicted that both h/e and h/2e oscillations can appear in TI nanowires and their relative strength varies with increasing disorder CONTENTS 12 Figure 3. AB/AAS oscillations in TI nanowires. (a) Schematic sketch of the measurement geometry, in which the current and magnetic field are applied parallel to the long axis of the wire. (b) Scanning electron image of a Bi2Se3 nanowire with multiple electrical contacts. (d) Band diagrams of a TI nanowire with zero (left) and half (right) flux quantum applied (i.e. φ = 0 and φ0). (e) Conductance fluctuations in a Bi1.33Sb0.67Se3 nanowire in which the chemical potential is lowed by surface coating of F4-TCNQ molecules (strong acceptors) and electrical back-gating. Panels (a-c) are taken from Peng et al., Nat. Mater. (2009) [101]; panel (d) from Bardarson et al., PRL (2010) [107], and panel (e) from Cho et al., Nat. Commun. (2015) [110]. (c) Resistance fluctuations with h/e-period oscillations when a parallel magnetic field is applied. strength [107]. This was confirmed by the transport measurement of Se-encapsulated Bi2Se3 nanowires, in which the disorder strength was controlled by inten- tional aging of the samples [109]. tuned close to the Dirac point. A direct observation of the 1D helical mode in a TI nanowire, however, requires the chemical potential In Ref. [110], the Fermi level in (Bi, Sb)2Se3 nanowires was controlled by electrical gating and the surface coating of acceptor-type molecules. The conductance maxima at half flux-quantum (and minima at zero- flux) appear to be consistent with the theory. Similar conductance oscillations were also observed in a back- gated Bi2Se3 nanowire sample [109], but the amplitude of conductance oscillations was, however, one order of magnitude lower than that observed in Ref. [110]. Therefore, it is desirable to further explore the helical 1D mode at half flux quantum in the samples with weak disorder and low chemical potential. In this regime, the variation in the longitudinal conductance is expected to be close to e2/h as the magnetic flux is increased from zero to Φ0/2 [107]. 5.2.2. Weak antilocalisation Another method for probing the surface transport is to utilize the weak It can be understood antilocalisation (WAL) effect. in the following semiclassical picture. low temperatures, electrons can remain phase coherent after many scattering events, namely lϕ (cid:29) le, where lϕ the dephasing length, and le is the mean free path. Since a pair of time-reversed paths for a closed electron At trajectory has exactly the same length, constructive interference takes place for a particle with negligible spin-orbit coupling. This leads to an increase in the returning probability of the particle and hence decrease in conductivity. For a Dirac fermion on the TI surface, the spin-momentum locking introduces an additional phase π between the pair of counterpropagating loops. The consequence is the destructive interference and the suppression of backscattering. This is known as weak antilocalisation, an effect opposite to the weak localisation [111, 112, 113]. The WAL effect has been observed frequently in 3D TIs [82]. It was, however, not easy to attribute it to the helical surface states exclusively, due to coexistence of parallel conduction channels, such as the bulk layer [114, 115] and the surface accumulation layer arising from downward band bending [116]. The spin-orbit coupling strengths are also very strong in these unwanted 2D channels, and they may belong to the same symmetry class (symplectic metal) as the helical surface states [111, 117]. Therefore, the WAL correction to longitudinal conductivity in these conducting channels can be described by the same equation, In addition to the parallel conductivity, the coupling between these channels the surface-bulk scatterings and inter-surface hybridization) further complicates the analysis of the WAL effect in many 3D TI samples [118, 119]. Nevertheless, valuable information on the surface states can still be obtained by systematically tuning the chemical potential and if they can be treated independently. (e.g. CONTENTS 13 Figure 4. Weak antilocalization effect in TI thin films. (a) Schematic sketch of helical 3D TI surface states, in which the spin- momentum locking leads to a Berry's phase of π. (b) Destructive interference between a pair of time-reversed path due to the π Berry's phase. (c) Sketch of the Hall-bar shaped Bi2Se3 thin film samples, in which the chemical potential is tuned with a back-gate via the STO substrate as a high-kappa dielectric. (d) Sheet carrier density dependence of the prefactor alpha extracted from the fits of magnetoconductivity data to the HLN equation. The α value remains close to 1/2 for a wide range of carrier densities (0.8 - 8.6×1013 cm−2), which correspond to Fermi levels located in the bulk conduction band (upper inset). (f) Gate-voltage dependence of a Bi2Se3 sample, in which the α value can be tuned from about 1/2 to nearly 1, when the bulk carriers are depleted by applying negative gate voltage. The upper and lower insets show the band diagrams for α = 1/2 and 1, respectively. Adapted from Chenet al., Phys. Rev. Lett. 105, 176602 (2010); et al., Phys. Rev. B 83, 241304 (2011) [120]; Zhang et al., Adv. Func. Mater. 21, 2351 (2011) [121]. quantitatively analyzing of the transport data. ln − Ψ B (17) (cid:20) + 2 Bϕ B (cid:19)(cid:21) (cid:18) Bϕ (cid:19) (cid:18) 1 A convenient method to unveil the WAL effect in TIs is to measure the longitudinal resistivity in per- pendicular magnetic fields. The magnetoconductiv- ity can be described by the Hikami-Larkin-Nagaoka (HLN) equation [111] simplified for the strong spin- orbit coupling limit: αe2 πh σ(B) − σ(0) = − in which Ψ is the digamma function, B is the magnetic field, Bϕ is the dephasing field, lϕ is the dephasing length. The value of prefactor α is dependent on the number of conduction channels and the coupling strengths between them. For instance, α = 1/2 if only one channel contributes, and α = n/2 for n independent and equivalent channels. In case of two channels with non-negligible inter-channel coupling or asymmetry in the dephasing lengths (i.e. lϕ,1 (cid:54)= lϕ,2), α would take a value between 1/2 and 1 and the Bϕ value would be determined by a complicated function of lϕ,1, lϕ,2, and other transport parameters [118]. It is worth noting that the HLN formula needs to be modified to account for strong spin-orbit scattering expected in topological materials [37, 38, 39]. Measurements of TI thin films or microflakes with conducting bulk often yield α values close to 1/2. This can be attributed to strong coupling between the surface and bulk states, which makes the sample behaving like a single channel system [118]. When the bulk carriers are depleted by electrical gating or reduced by appropriate doping, crossover from α ∼ 1/2 to α ∼ 1 can be observed [120, 122, 123, 124, 125] (See Fig. 4 for an example). The α value close to 1 is a signature of decoupling between two conduction channels. If the film is sufficiently thin or the bulk carrier density is low enough, a full depletion of bulk carriers can be achieved with electrical gating. In this case, α = 1 would mean a pair of equivalent surfaces decoupled to each other. Such a transport regime is suitable for pursuing intrinsic quantum transport properties of the helical surface states. For instance, Liao et al. found that the electron dephasing rate has an anomalous sublinear power-law temperature dependence in the (Bi, Sb)2Te3 thin films in the decoupled surface transport regime. In contrast, when the same film is tuned to the bulk conducting regime, the dephasing rate returns to linear temperature dependence commonly seen in conventional 2D electron systems. The coupling of the surface states to charge puddles in the bulk was proposed as a possible origin of the sublinear power law [126]. CONTENTS 14 Figure 5. Observation of the quantum Hall effect in a 3D TI. (a) Gate-voltage dependence of longitudinal resistance of (Bi,Sb)2(Te,Se)3 (BSTS) microflake, in which the chemical potential is tuned with a back gate via a 300 nm thick SiO2 layer. (b) Longitudinal conductivity σxx and σxy plotted as functions of back-gate voltage in a magnetic field of B = 31 T applied perpendicularly to the BSTS flake. Here the total Hall conductivity contains contributions from the top and bottom surfaces. The top surface component remains quantized as 1 2 e2/h, while the bottom surface contribution can be tuned with the back-gate voltage. Adapted from Xu et al., Nat. Phys. 10, 958 (2014) [127] 5.2.3. Quantum Hall effect The Dirac electrons on the TI surface states can in principle exhibit half- integer quantum Hall effect (QHE) due to the existence of a zero-energy Landau level [80, 81]. Despite the absence of spin or valley degeneracies in many 3D TIs, direct observation of the half-integer quantum Hall plateau has not been possible because both the top and bottom surfaces contribute to transport. The quantised Hall conductivity in a slab geometry can be written as σxy = σt +σb = (νt +νb) e2 h = (Nt + 1 2 +Nb + 1 2 ) e2 h , (18) where σH is the Hall conductivity, νi the filling factor, and Ni the Landau level index of surface i, with i = t, b. Here t and b denote the top and bottom surfaces, respectively. In a magnetic field of the order of 10 T, the carrier mobility of several thousand cm2/V·s is usually sufficient for observing the QHE. Bismuth chalcogenides with such mobilities were not difficult to obtain even in early days of TI research [98]. The failure in observing the QHE in TIs can be attributed to the conducting bulk, which provides backscattering paths for the chiral edge channels in the quantum Hall regime. The first observation of QHE in TIs was realized in 2011 with a strained 70 nm thick HgTe epilayer with electron mobility 3.4×104 cm2/V·s. Multiple quantum Hall plateaus in σxy were observed but they were accompanied substantial longitudinal conductivities. In 2014, two groups managed to observe the QHE in bismuth chalcogenides. Xu et al. observed well- defined quantum Hall plateau of the zeroth Landau level at B > 15T and T = 0.35K using bottom-gated microflakes exfoliated from (Bi, Sb)2 (Te, Se)3 single crystals [127]. Yoshimi et al. utilized high quality (Bi, Sb)2Te3 thin films grown in InP(111) substrates and realized the σH = ±1e2/h plateaus in a magnetic field of 14 T and at T = 40 mK [128]. Using a sample with quality higher than those in Ref. [127] and dual gating, Xu et al. studied the zero Hall state (with (νt, νb) = (−1/2, 1/2) or(cid:0) 1 (cid:1)) in detail. 2 ,− 1 2 Their joint local and nonlocal measurements suggest the existence of a quasi-1D dissipative edge channel with nearly temperature independent conductance for T < 50 K [129]. The QHE has also been observed in ungated Bi2Se3 thin films of both n-type and p-type. Such observations represent a remarkable technical achievement in reducing the density of defects in Bi2Se3 by optimizing the MBE growth processes [130, 131]. It is also noteworthy that Zhang et al. recently observed a well-defined quantum Hall νtot = 1 plateau in an exfoliated Sn-doped (Bi, Sb)2 (Te, S)3 microflake in magnetic fields less than 4 T at T = 6 K [132]. Further improvement in the quality of TIs may enable observations of the fractional QHE and other interaction-induced quantum phenomena. Theoretical developments on exotic quantum Hall physics in topological materials merit a brief mention. In TIs, the half-quantised quantum Hall effect on a single surface was addressed in [266], while the Landau levels of thin films were recently discussed in [267]. The Hall plateaux fall into two patterns, one pattern the filling number covers all integers while only odd integers in the other. The quantum spin Hall effect, normally protected by time reversal symmetry, is destroyed by the magnetic field together with structure inversion asymmetry. Li et al [268] showed that in a MoS2 trilayer the Landau level energies grow linearly with B, rather than with √B, and display an unconventional Hall plateau sequence consisting of odd negative integers and even positive integers, with the crossover at n = 0. A spin- resolved p − n junction also exhibits Landau levels at CONTENTS 15 Figure 6. Planar Hall effect (PHE) and anisotropic magnetoresistance (AMR) in a dual-gated (Bi,Sb)2Te3 thin film. (a) Hall resistance Ryx and longitudinal resistance Rxx versus ϕ, the angle between the current and the in-plane magnetic field. (b) The amplitude of PHE as a function of the strength of magnetic field. (c) PHE amplitude as a function of gate voltages, which are symmetrically tuned (i.e. the top and bottom gate voltages remain equal). The upper right inset shows the Hall resistance curves under various gate voltages. The change in chemical potential can be estimated from the gate-voltage dependence of the Hall resistance (bottom curve and scale bar). Adapted from Taskin et al., Nat. Commun. 8, 1340 (2017) [143]. fractional filling factors. Later, it was shown that, at low doping, TMDs under shear strain develop spin- polarized Landau levels residing in different valleys, with Landau level gaps between 10 and 100 K [269]. A superlattice arising from a Moire pattern can lead to topologically nontrivial subbands, with edge transport quantised. focused on the (111) surface of SnTe in a quantising magnetic field, predicting a nematic phase with spontaneously broken C3 symmetry due to the competition between intravalley Coulomb interactions that favor a valley- polarised state and weaker intervalley scattering that increases in strength as the magnetic field is ramped up. [270] Ref. 5.2.4. Planar Hall effect When a magnetic field is applied parallel to the surface of a 3D TI, the transport properties were once considered to be unaffected because the Zeeman term can be gauged way by shifting the Dirac cone in the momentum space. Taskin et al., however, observed that in dual-gated devices of bulk-insulating (Bi, Sb)2Te3 thin films with dominating surface transport the in-plane magnetic field can induce an anisotropic magnetoresistance (AMR) and the planar Hall effect (PHE) [133]. In contrast to the anomalous Hall effect, the planar Hall effect consists of a resistivity anisotropy induced by an in-plane magnetic field, which anisotropically lifts the protection of surface Dirac fermions from backscattering, in a configuration in which the conventional Hall effect vanishes. Both AMR and PHE are phenomenologically similar those encountered in ferromagnetic materials. The former exhibits an angular dependence of cos2 ϕ and the latter is proportional to sin ϕ cos ϕ, where ϕ is the angle between the magnetic field and the applied current. The AMR and PHE are two aspects of the same anisotropic scattering process in conventional magnetic materials, and consequently they have the same amplitude. Taskin et al. proposed a theoretical model to account for these effects in 3D TIs. It is based on the assumption that the in-plane magnetic field breaks the time-reversal symmetry, and removes the protection from backscattering for spins perpendicular to the magnetic field, but at the same time maintains the protection for spins parallel to the magnetic field. This model can qualitatively reproduce the double- dip structure of the AMR/PHE amplitude as function of gate voltage. They estimated that about 10% of impurity scatterings are related to spin flips at B = 9 T [133]. The planar Hall effect can appear in the presence of magnetic disorder in a TI/ferromagnet structure when the external magnetic field is aligned with the magnetisation orientation [134], as well as in tunneling across a single ferromagnetic barrier on a TI surface when the magnetisation has a component along the bias direction [135]. Recently, the PHE and AMR effects were observed in Sr0.06Bi2Se3 [136] and Sn-doped (Bi, Sb)2 (Te, S)3 samples [137]. Nandy et al. developed a semiclassical Boltzmann theory of the transport in Bi2Se3 and showed that the nontrivial Berry phase of the bulk states can also lead to the PHE, as well as negative longitudinal magnetoresistance [138]. Certain contributions to the planar Hall effect can be traced to the Berry phase and orbital magnetic moment familiar from semiclassical dynamics [139]. Similar phenomena have been observed topological semimetals [140] and discussed in the context of chiral anomaly [141, 138]. However, according to a recent theoretical study, the negative longitudinal magnetoresistance is not necessarily associated with the chiral anomaly [142]. CONTENTS 16 Figure 7. Topological magnetoelectric effect in Bi2Se3 thin films. (a) Sketch of the Faraday rotation measurement setup in the THz regime. (b) Faraday rotation angle as a function of the frequency for severalBi2Se3 thin films with thicknesses of 6 to 12 quintuple layers (QL). The dashed horizontal lines denote theoretical values expected from the magnetoelectric effect. The upper-right corner shows the quantum Hall effect of a 8 QL thin film prepared in a similar condition. Adapted from Wu et al., Science 354, 1124 (2016) [146]. 5.2.5. Topological magnetoelectric effect In the previous subsections we have been mainly concerned with dc transport properties. The difference between a 3D TI and an ordinary insulator can also be manifested in their response to the electromagnetic field. According to the topological field theory [144], the Maxwell Lagrangian has an axion term α 4π2 θE·B, in which α = e2 c is the fine structure constant, and θ = π for a 3D TI. In contrast, θ takes a value of 0 for an ordinary insulator. When the time-reversal symmetry (TRS) is broken (by applying external magnetic field, utilizing proximity effect of a magnetic insulator, or magnetic doping, see Section 5.3.2), the axion term leads to some interesting modifications in the Maxwell equations, and hence the topological magnetoelectric effect [144, 145]. This can be detected by optically techniques, such as Faraday and Kerr rotations. In 3D TIs, the Faraday and Kerr rotation angles are quantized in units of the fine structure constant, if the Fermi level can be placed between the Landau levels or inside the magnetic gap induced by exchange interaction. In the former case the Faraday rotation angle follows tan θF = α(cid:0)Nt + 1 (cid:1) in a free- 2 + Nb + 1 2 standing TI film, while the latter is corresponding to a quantum anomalous Hall insulator in which tan θF = α. The quantized Faraday rotation has been observed with time-resolved terahertz (THz) spectroscopy by a few groups [146, 147, 148]. Wu et al. used high quality Bi2Se3 thin films grown on sapphire substrates and capped with MoO3, in which the chemical potential is as low as 30 meV even without gating. Fig. 7 shows the Faraday rotation angles are close to the quantized values expected for total Landau level filling factor Nt + Nb = 1, 3, 4, and 6. It is noteworthy that the quantization of θF can be observed in magnetic fields down to 5 T, much lower than B = 20 T required for observation the QHE in similar films [146]. The quantized Faraday rotation was demonstrated in a strained HgTe thin film tuned with a Ru top gate in high magnetic fields [147]. The THz spectroscopy measurements of Cr-doped (Bi, Sb)Te3 thin films showed that a material-independent scaling function of θF and Kerr rotation angle θK approaches to the fine structure constant as the dc Hall conductance increases toward 1 e2/h [148]. 5.3. Novel magnetism with 3D TIs 5.3.1. Breaking time-reversal symmetry in 3D TIs Breaking TRS in 3D TIs are predicted to result in a plethora of novel quantum phenomena, and many of them have been realized in experiment. The surface states with broken TRS can be described with the following Hamiltonian H = vF (kxσy − kyσx) + ∆σz, (19) in which the mass term ∆σz is introduced by the exchange interaction. The mass term turns the gapless surface states into a system with a gap of 2∆. The broken TRS also modifies the spin texture of the surface states. The electron spins are no longer locked perpendicularly to the momentum. They rather form a hedgehog-like spin texture in which the spin direction is perpendicular to the TI surface at µ = ±∆ and slowly evolves into the in-plane helical structure for µ (cid:29) ∆. As a consequence, the Berry phase becomes ϕ = π(1 − ∆/µ) if the Fermi level lies outside the mass gap (µ > ∆). When the Fermi level is located CONTENTS 17 (cid:0) 1 inside the gap, a topological invariant, known as the Chern number C, can be defined by integration over the Brillouin zone. For a slab-shaped sample, C = ± surfaces contribute 1/2 to the total value, and the sign of C is identical to that of ∆. (cid:1) = ±1, in which both the top and bottom 2 + 1 2 The topologically nontrivial electronic structure described above can be manifested in electron transport. If the Fermi level lies in the mass gap, the quantum anomalous Hall effect (QAHE) occurs. The Hall conductivity follows σxy = C e2 h , accompanied by vanishing longitudinal conductivity (σxx = 0). Both properties are the hallmarks of a Chern insulator, in which either QAHE or QHE can be observed. In the latter, the Chern number is equal to the integer filling factor of the Landau level, also known as the TKNN number [149]. For a Chern insulator with boundary, electron transport is carried by 1D ballistic chiral edge states surrounding an insulating bulk. Such transport is often called dissipationless because the edge transport can be free from backscattering for macroscopic distances. It should be noted, however, that energy dissipation cannot be completely avoided in a Chern insulator. It takes place at the so-called hot spots at the corners of source and drain contacts [150], and the contact resistance is of the order 1 C h e2 . level (cid:17)3 (cid:16) ∆ When the Fermi lies outside the mass gap, the bulk becomes conducting, and the Hall conductivity is no longer quantized. Ado et al. [257] considered all three sources of the anomalous Hall (AH) conductivity, namely the intrinsic contribution, skew scattering and side jump, and predicted that for µ (cid:29) ∆. Lu et al. predicted that the σxy ∝ variation of the Berry phase from ϕ = π at µ (cid:29) ∆ to ϕ = 0 at the edge of the mass gap (i.e. µ = ±∆) can induce a crossover from the WAL to the weak localisation [152]. µ Two approaches have been widely explored to open the mass gap in 3D TIs [153, 154]. One is to intro- duce magnetic order by doping with transition metal elements. The other is to utilize interfacial exchange interactions in TI/magnetic insulator heterostructures. In the following two subsections, we shall review the ex- perimental progresses in these two directions, while the subsequent subsection covers theoretical developments in this field. 5.3.2. Magnetically doped 3D TIs Many transition metal and rare earth elements have been used as magnetic dopants to introduce ferromagnetic order in 3D TIs. Among them, Cr, V, Mn and Fe have received most attention [155, 92, 156, 157, 158]. In particular, Cr-doped (Bi, Sb)2Te3 (Cr-BST) thin films allowed for the first experimental observation of the QAHE [157], as shown in Fig. 8. Enormous experimental efforts have Figure 8. Observation of the quantum anomalous Hall effect. (a) Sketch of the backed-gated Cr-doped (Bi,Sb)2Te3 thin film. The perpendicular magnetization induced by the Cr doping turns the massless surface states into a massive, gapped 2D system. (b) Qualitative chemical potential of longitudinal conductivity (σxx) and Hall conductivity (σxy). Quantization of σxy can be realized when the Fermi level is tuned into the exchange gap. (c) Experimentally observed longitudinal resistivity (ρxx) and Hall resistivity (ρxy) plotted as a function of the back-gate voltage. (d) Hall resistivity as a function of applied magnetic field at several gate voltages. Taken from Chang et al., Science 340, 167 (2013) [157]. since been made to optimize the MBE growth Cr- BST thin films. However, observation of the QAHE in Cr-BST thin films is still limited to temperatures of a few hundred mK or lower, despite that the Curie temperature is on the order of 10 K [153, 154]. This can be attributed partially to the spatial fluctuations in the Dirac mass gap. According to an STM study, the local exchange gap in a Cr-BST thin film varies from 9 meV to 51 meV due to fluctuations in the local density of magnetic dopants [159]. In the first observation of QAHE [160], the zero-field longitudinal conductivity was about 0.1 e2 h even at T =30 mK and the backscattering through the bulk states restricted the accuracy of quantum anomalous Hall (QAH) resistance to about 1% [160]. More precise QAHE was later realized in V-doped Sb2Te3 thin films, in which the QAH plateau reached a precision level of 0.02% at T =25 mK [158]. Recently, two groups have managed to observe QAH resistances with precision levels of 1 part in 106 and 0.1 part in 106 in Cr-BST [161] and V-doped (Bi, Sb)2Te3 (V-BST) [162] thin films, respectively. A lot of efforts have also been devoted to increasing the observation temperature of QAHE. One method is to utilize the modulation doping technique developed in semiconductor heterostructures [163]. Two ultrathin (Bi, Sb)2Te3 layers heavily doped with Cr are placed 1 nm from the top and bottom surfaces while the majority of the film is not magnetically doped. The Hall resistance in this multilayer structure reaches 0.97 h e2 at T = 2 K. The other method is to CONTENTS 18 use a co-doping scheme. The QAHE can be realized at T =1.5 K in a Cr and V co-doped (Bi, Sb)2Te3 sample. It is believed that the increased homogeneity is responsible for the realization of QAHE at higher temperatures [164]. It was also shown that there is a large difference in the coercive fields of Cr- and V-doped BST thin films. This property was employed to realize the axion insulator state [165, 166, 167], in which the magnetization directions of Cr-BST and V-BST layers are opposite to each other, and the longitudinal conductivity vanishes like an QAH insulator, but the Hall conductivity displays a zero plateau. These features are nearly perfectly borne out in the transport data of a V-BST/BST/Cr-BST trilayer structure [165, 166]. The axion insulator phase can exist in a wide range of magnetic fields (µ0H = 0.2 − 0.8T) at low temperatures. It is remarkable that the longitudinal resistance can be as high as 1GΩ in this insulating phase. Similar results are also reported in Ref. [167], which shows the zero-plateau of the Hall resistance, in addition to those of the longitudinal and Hall conductivities. In contrast to the Cr and V doping, experiments with other magnetic dopants have been futile in producing a well-defined QAH phase to date [92, 156, 168, 169, 170]. This was somewhat surprising considering that sizable gap opening in the surface states was observed in Fe-doped and Mn-doped Bi2Se3 thin films with ARPES measurements [92, 156]. In a subsequent experiment, however, S´anchez-Barriga et al. showed that in Mn-doped Bi2Se3 thin films the gap in the surface states, which can be as large as 200 meV, survives at room temperature, in stark contrast with the low Curie temperatures in the sample (less than 10 K for both bulk and surface magnetism) [171]. They suggested that this kind of energy gap originates from nonmagnetic resonant scatterings related to the impurities in the bulk, instead of the magnetic exchange interaction [171]. Such nonmagnetic scatterings complicate the electronic structure and are detrimental to realizing the QHAE. Recently, Liu et al. further showed that the AH resistance in (Bi1−xMnx)2 Se3 thin film can have too components with different signs, and the surface component has a sign opposite to that of the QAH phases in Cr- or V-doped BST films [172]. Similar sign in the AH resistance was also observed in Mn-doped Bi2 (Te2Se)3 [168], Mn-doped Bi2Te3 [173], and Cr- doped Bi2Te3 [174, 175]. It remains to be investigated whether such 'anomalous' sign in σxy is related to the non-magnetic scatterings [176]. Nevertheless, the results described above suggest that the influence of magnetic dopants is far more complicated than the mean field exchange interaction considered in the massive Dirac fermion model. A deep understanding of the impurity effect, both magnetic and non-magnetic, are crucial to finding magnetic TIs with improved quality. is surface states 5.3.3. Magnetic heterostructures Another approach to open a gap in the TI to make use of the proximity effect in TI/magnetic insulator (MI) heterostructures. Both ferromagnetic and antiferromagnetic insulators with spin aligned perpendicular to the surface are capable of generating a mass gap via the exchange interaction across the interface. One advantage of the heterostructure approach is that it may overcome the difficulty of magnetic inhomogeneity encountered in magnetically doped TIs [159]. This not only allows for the observation of QAHE at high temperatures, but also facilitates the study of exotic quasiparticles, such as chiral Majorana zero modes [177, 178, 179] and magnetic monopoles [144]. [183], BaFe12O19 [184], Cr2Ge2Te6 Many ferromagnetic or ferrimagnetic insulators, including EuS [180, 181], GdN [182], Y3Fe5O12 (YIG) [185], Tm3Fe5O12 (TIG) [186], Fe3O4 [187], and CoFe2O4 [188], have been used to fabricate the MI/TI heterostruc- tures. Evidence for interfacial magnetic interactions has been obtained by the measurements of the WAL effect in perpendicular or parallel magnetic fields [180, 184], Kerr spectroscopy [183], and polarized neutron reflectometry [189, 190]. However, the gap opening ef- fect are very weak in most of these heterostructures, as evidenced by the fact that the reported AH resis- tances did not exceed several Ohms. Nevertheless, AH resistances up to 120 Ω have been observed recently in a (Bi, Sb)2(Te, Se)3/(Ga, Mn)As heterostructure, in which the magnetic semiconductor layer is in an insu- lating regime and has a perpendicular easy axis [191]. More recently, in a Cr2Ge2Te6/TI/Cr2Ge2T6 sandwich structure AH resistances of the order kΩ has been achieved [192]. It is noteworthy that in these two stud- ies, both the TI and MI layers were fabricated with MBE in ultrahigh vacuum. This suggests that high in- terface quality is crucial in obtaining strong magnetic proximity effect in the TI surface states. The exchange interaction between antiferromag- netic insulators and TIs may also open a sizable gap in the surface states. Based on the first prin- ciples calculations, the magnetic proximity effect in MnSe/Bi2Se3 heterostructure can induce a gap of about 54 meV [193]. Another theoretical study of the same heterostructure, however, suggested that the gap associated with the Dirac surface states is very small (8.5 meV), and also coexists with trivial metallic states [194]. An ARPES study of MnSe ultrathin lay- ers grown on Bi2Se3 however, revealed an energy gap of CONTENTS 19 100 meV, which is free of any other electronic states. This surprising result is attributed to direct interac- tion of the Dirac surface states with a Bi2MnSe4 septu- ple (SL) layer (1 SL=Se-Bi-Se-Mn-Se-Bi-Se), which is formed from the intercalation of a MnSe layer into the quintuple layers of Bi2Se3. The calculation reported in Ref. [195] also produces a Chern number C = −1, indicating high temperature QAHE can in principle be observed in this system. Very recently, exciting results have been reported on a antiferromagnetic compound in the same family, Bi2MnTe4. The first principles cal- culations suggested that when Bi2MnTe4, which has a layered structure, is exfoliated into ultrathin layers, the ground state can oscillate between a QAH insulator and an axion insulator phase, depending on whether the number of Bi2MnTe4 septuple layers is odd or even [196, 197]. It is remarkable that a quantized Hall resistance was observed lately in a 5 SL microflake of Bi2MnTe4 at T =4 K, although a high magnetic field had to be applied to align the Mn2+ spins [198]. Fur- ther work along this line may lead to the observation of both QAHE and the axion insulator state in zero magnetic field in stoichiometric materials. This might pave a way for discovering novel quantum phenomena as well as manipulate exotic quasiparticles for potential applications. 5.3.4. Magnetic ordering in TIs: theory Magnetic order has a profound effect on charge transport through a material. When a current passes through a magnetised material topological mechanisms and scattering processes predominantly deflect electrons in one direction. This results in an additional current perpendicular to the driving current, which depends on the magnetisation, and vanishes if the material is non-magnetic. Whereas the Hall effect of classical physics requires an external magnetic field, this effect, termed the anomalous Hall effect, requires only a magnetisation. It is often the smoking gun for detecting magnetic order and remains one of the biggest focus areas for research on TIs. In ultra- thin films of magnetic topological insulator a quantised anomalous Hall effect (QAHE) exists, carried by edge states believed to be topologically protected, with a transverse conductivity of e2/h. The quantum anomalous Hall effect is believed to be dissipationless and is being investigated for electronics applications [235]. Prior to discussing anomalous Hall effects, we examine the critical aspect of magnetic ordering in TI, which has been the subject of a number of papers. In undoped TI the ordering was originally believed to be due to the van Vleck mechanism [236] and to point out of the plane, in contrast to thin film ferromagnets, in which the magnetisation typically points in the plane. This interpretation is being challenged, as we shall see below. Interestingly, Rosenberg and Franz [237] showed that in a magnetically doped metallic TI the surface magnetic ordering can persist up to a higher critical temperature than the bulk, so that a regime exists where the surface is magnetically ordered but the bulk is not. This is believed to be because the metallic surface state is more susceptible to magnetic ordering than the insulating bulk. Magnetic impurities were also shown to give rise to a Kondo effect [238] and non-Fermi liquid behavior [239]. Even in the absence of magnetism distinctive features appear in the static spin response in 2D TIs for the topologically nontrivial band-inverted structure [240]. Electronic correlations such as the Hubbard U affect the ordering in ferromagnetically doped TI thin films, as was seen in V-doped Sb2Te3 [241]. The on-site Coulomb interaction can turn the TI thin film into a Mott insulator and facilitate it entering the quantum anomalous Hall phase, discussed below. Ferromagnetic order is determined by p-orbital-assisted long-range superexchange as well as short-range double-exchange between the partially filled d-bands, which enhances it relative to Cr doping. In determining the highest possible observation temperature of the QAHE the two important energy scales are the band gap of the magnetic TI film, given by the size of the magnetisation, and the ferromagnetic Curie temperature. The smaller of the two defines the observation temperature of the QAHE. Two interesting publications have shed light on co-doping mechanisms that can be used to enhance the temperature at which the QAHE is observed, which are currently a few tens of mK. Qi et al [242] determined that the QAHE can occur at high temperatures in n-p co- doped TIs, taking as an example vanadium-iodine co- doped Sb2Te3. The chosen dopants have a preference for forming n-p pairs due to mutual electrostatic attraction, thereby enhancing their solubility. While doping with V alone would shrink the bulk gap, co- doping with I restores it to its original value. Even at 2% V and 1% I, the QAHE persists until 50K. The authors ascribe this enhancement to the fact that compensated n − p codoping preserves the intrinsic band gap of the host material. In a similar vein, Kim et al showed that the QAHE temperature can be enhanced significantly by Mo-Cr co-doping Sb2Te3 [243]. At the same time, these authors discovered that the ferromagnetic order in Cr-doped Sb2Te3 survives when the spin-orbit interaction is turned off, implying that the magnetic order is not governed by the van Vleck-type mechanism, which relies on nontrivial band topology. Since the system is an insulator for Cr doping ≤ 10%, the RKKY mechanism can also be discounted, while the surface states were found to CONTENTS 20 spin polarisation, which can be detected in dc transport by utilising ferromagnetic electrodes as the probe [201]. In a typical experiment a TI is placed on top of a ferromagnet, a current is driven through the TI, and the effect of the current on the magnetisation of the ferromagnet is observed. For a large enough current density the magnetisation may switch, and for technological applications it is desirable to make this critical current density as low as possible, with operation ideally possible at room temperature. The efficiency in the generation of SOT can be characterized by a dimensionless parameter, η = 2e JS , where Js is the spin current density and Jc is Jc the charge current density. ST-FMR measurements have been carried out on many TI/ferromagnet including Bi2Se3/NiFe [202, 203], heterostructures, Bi2Se3/CoFeB [204], (Bi, Sb)2Te3/NiFe [205], and Bi2Se3/YIG [206]. The extracted maximum value of the charge-spin conversion efficiency η spreads from about 0.4 to 3.5 for each of these systems. Even though the η values are in general larger than those of heavy metal/ferromagnet heterostructures [202], they are two orders of magnitude smaller than that the values (η = 140 − 425) obtained by the second harmonic measurements of the Hall voltages in (Bi, Sb)2Te3/(Cr, Bi, Sb)2Te3 heterostructures [207]. Such a large discrepancy was resolved in Ref. [208], in which the second harmonic Hall voltage is identified to mainly originate from asymmetric magnon scattering, instead of the contribution from the damping-like SOT due to the current induced spin polarization. ferromagnet/TI heterostructures, Recently, SOT-induced magnetisation switching room temperature in has been demonstrated at several such as CoTb/Bi2Se3 [210], a TI-ferrimagnet heterostructure with perpendicular magnetic anisotropy, accompanied by a large spin-Hall effect, NiFe/Bi2Se3 [211] due to the spin polarisation induced in the topological insulator, Bi1−xSbx/MnGa [212] with a sizable spin-Hall effect, and BixSe1−x/CoFeB [209]. The current density required for magnetic switching is often on the order of 105A/cm2, which is considerably lower than those for heavy metals (e.g. 5.5 × 106A/cm2 for β − Ta [213]). These experiments demonstrated the potential of the TI-based magnetic random-access memory. However, it should be noted that the ferromagnetic layer used in these experiments are metallic and often have a conductivity much larger (or least comparable to) that of the TI layer. Substantial amount of energy is wasted due to the current shunting by the ferromagnetic layer. It would be very appealing to achieve efficient current induced magnetization reversal in a magnetic insulator. The 2D materials will offer excellent opportunities for fabricating TI/MI heterostructures with higher efficiency in generating SOT, since a Figure 9. Room temperature current-induced magneti- zation switching in ferromagnetic metal/TI heterostructure. (a) Schematic of a BixSe1−x (4 nm)/Ta (0.5 nm)/CoFeB (0.6 nm)/Gd (1.2 nm)/CoFeB (1.1 nm) stack used in the mea- (b) Anomalous Hall resistance RAH as a function surements. of applied magnetic field. (c,d) Current-induced switching of the magnetization due to the spin-orbit torque generated by the in-plane current in the BixSe1−x underlayer in the presence of 80 Oe (c) and −80 Oe (d) in-plane bias H-fields. From Mahendra et al., Nat. Mater. 17, 800 (2018)[209]. be of secondary importance in magnetic ordering. DFT calculations reveal that magnetic ordering arises from long-range exchange interactions within quintuple layers, mediated by directional bonds with Te by certain sets of orbitals on the Cr and Sb atoms, in a similar way to hydrogen local moments in graphene. A Zeeman field can also be induced in a TI via a magnetic proximity effect. Two studies from the same group shed light on this process for a TI on a ferromagnetic insulator [244] as well as for a TI on an antiferromagnetic insulator [245], the latter using analytics as well as density functional theory in Bi2Se3/MnSe(111). A unified qualitative picture emerges. Charge redistribution and mixing of orbitals of the two materials cause drastic modifications of the electronic structure near the interface. In addition to the topological bound state an ordinary bound state is present, which is gapped and spin polarised due to hybridisation with the magnet. The two overlap in space in such a way that the ordinary state mediates indirect exchange coupling between the magnet and the topological state, and the latter acquires a gap at the Dirac point. insulators 5.3.5. Spin-orbit torque Much of the recent interest in topological centred around the phenomenon of spin-orbit torque [199, 200], which is the Onsager reciprocal of charge pumping. The spin- momentum locking in the TI surface states offers a convenient and efficient means to electrically generate a is CONTENTS 21 number of 2D materials have been identified to be magnetic insulators(semiconductors) [214], and atomically sharped interface can be obtained handily with van der Waals epitaxy. surface states Two fundamental effects underpin spin-orbit torques. One is the generation of a spin polarisation of the surface states by an electrical current, or magneto-electric effect [215, 216]. In a TI/ferromagnet heterostructure the current-induced spin polarisation in the exerts a torque on the ferromagnetic layer, which can be regarded as a TI counterpart of the Rashba spin-orbit torque (SOT) in heavy metal/ferromagnet heterostructures [217]. This effect requires, at the very least, spatial inversion symmetry breaking, hence cannot originate from the bulk of the TI. The second effect is the spin-Hall effect, which, conversely, stems from the bulk of the TI and is zero for the surface states. This refers to a non-equilibrium spin current driven by an electric field. The exact origin of the strong torque observed in any one experiment, in particular whether it stems from the spin-momentum locking of the surface states or from spin-Hall currents in the bulk, In this context Ref. [218] considered a TI/ferromagnet heterostructure in which the Fermi energy was varied so that at one extreme transport is entirely surface-dominated while at the other it is entirely bulk-dominated, and showed that the spin Hall torque remains small even in the bulk-dominated regime. A current-induced spin polarisation was observed in Bi2Se3 by performing spin torque ferromagnetic magnetic resonance (ST-FMR) experiments [202], where the spin polarisation is in the plane. In a series of more recent experiments, the current-induced spin polarisation in WTe2 was shown to be out of the plane [219, 220]. tends to be unclear. The current-induced spin polarisation has a simple explanation. A steady-state current corresponds to a net momentum, and since the spin of the surface states is locked to the momentum, this automatically ensures there is a net spin polarisation. This is quite general in a topological insulator, as long as no warping terms are present, and is valid for currents both longitudinal and transverse to the applied electric field. Yet research is beginning to emerge demonstrating that the dynamics in the vicinity of the TI/ferromagnet interface are non- trivial, and may be vital in understanding what is measured experimentally. A numerical study of the spin-orbit magneto-electric effect [221] showed that the stead-state surface spin polarisation extends ≈ 2nm into the bulk of the TI as a result of wave function penetration into the bulk, as in Fig. 10. When the hexagonal warping term ∝ λ is taken into account, in addition to modifications to the conductivity [222], an out of plane spin polarisation also emerges. A Figure 10. Current-induced spin polarisation and spin texture, adapted from [221]. (a) The arrangement of Bi and Se atoms in a supercell of a Bi2Se3 thin film with a thickness of 5 quintuple layers. The inset in panel (a) shows the Brillouin zone in the kx − ky plane at kz = 0. (b) The vector field of the non- equilibrium spin polarisation S(r) within selected planes shown in (a), generated by injection of an unpolarised charge current along the x-axis. The planes 1 and 3 correspond to the top and bottom metallic surfaces of the film, while plane 2 resides in the bulk at a distance d ≈ 0.164 nm away from plane 1. (c) The vector fields in (b) projected onto each of the selected planes in (a). The real space grid of r points in panels (b) and (c) has spacing ≈ 0.4A. recent computational work considered a Bi2Se3/Co bilayer [223] and demonstrated that the Co layer is substantially modified to acquire spin-orbit properties of Bi2Se3, so when current flows through the Co, a non- equilibrium electronic spin density will be generated that is noncollinear to the Co magnetisation. An additional contribution to the spin polarisa- tion, and therefore the spin-orbit torque, exists when a TI or a 2D spin-orbit coupled semiconductor is placed on a ferromagnet with magnetisation perpendicular to the 2D plane. It has been termed intrinsic, meaning it is due to the band structure, and it stems from the fact that the magnetisation opens a gap in the spec- trum of the TI/semiconductor. This contribution to the spin polarisation is associated with the same term in the charge conductivity that leads to the anoma- lous Hall effect. In a 2D semiconductor with spin-orbit coupling linear in momentum it is known that the in- trinsic anomalous Hall conductivity is cancelled out by scalar disorder when the vertex corrections are taken into account [224], in analogy with cancellations oc- curring in the spin-Hall effect [225]. Ado et al [226] revealed that, as expected, for spin-independent disor- der the same is true for the intrinsic spin-orbit torque, and the only remaining torque is due to the current- induced spin polarisation. We note that this cancella- tion requires both spin-orbit split sub-bands to cross the Fermi surface and does not apply to TI. Ref. [226] also introduced a more physical representation which allows the decomposition of the torques into a dissi- (a)yzxBiSexy(b)(c)123kxkyΓMKCONTENTS 22 pationless component (field-like) invariant under time reversal, and a dissipative component (damping-like) that changes sign under time reversal. In a subsequent work, Xiao and Niu [227] showed that the net result for the intrinsic torque depends on the structure of the disorder potential. For a non-equilibrium spin polarisation S acting In on a magnetisation M the torque T ∝ M × S. the magnetism literature spin torques are customarily broken up into two components, one referred to as field- like, and the other as damping-like or antidamping-like. This nomenclature stems from the Landau-Lifshitz- Gilbert equation, in which the magnetic field term M × H is responsible for precession and the Gilbert term M × (M × H) for damping of the precessional motion. In the context of spin transfer torques this nomenclature is justified, since the damping-like torque is dissipative, while the field-like contribution is dissipationless. This nomenclature has also permeated the literature on spin-orbit torques, where the field- like and anti-damping-like torques ∝ M × S and ∝ M × (M × S) respectively, where S is the current- induced spin polarisation. As explained in [226] the analogy is not exact and can be misleading. The current-induced spin polarisation, for example, has different directions for Rashba and Dresselhaus spin orbit coupling, yet is always dissipative, as it depends on the scattering time τ . For a realistic structure both the Rashba and Dresselhaus interactions tend to be present and the direction of the current-induced spin polarisation is unknown a-priori. Fischer et al [228] concentrated on heterostruc- tures comprising either a TI on a ferromagnet or a TI on a magnetically-doped TI. In both cases the mag- netisation was taken to be in the plane. In addition to the current-induced spin polarisation, an additional out-of-plane torque was found, arising from spin diffu- sion across the interface combined with spin precession of the current-induced spin polarisation around the in- plane magnetisation. The two have different transfor- mation properties under magnetisation reversal. For a TI on a ferromagnet both torques have comparable effi- ciencies, while for a TI on a magnetically doped TI the spin transfer-like torque is found to dominate. Within a similar setup, differences between the single Fermi surface Rashba-Dirac Hamiltonian and two Fermi sur- face Rashba Hamiltonian are considered [229]. A series of papers have reported drastic enhance- ments of spin-orbit related effects in hybrid graphene- TI structures. In Ref. [230] it is reported that epitax- ial graphene on the TI Sb2Te3 evolves into the quan- tum spin-Hall phase and develops a spin-orbit gap of 20 meV. Another paper [231] reports the possibility that the addition of graphene monolayers or bilayers to a TI-based magnetic structure greatly enhances the current-induced spin polarisation by a factor of up to 100, due to the high mobility of graphene and to the fact that graphene very effectively screens charge im- purities, which are the dominant source of disorder in topological insulators. Zhang et al computed the spin-transfer torque in graphene-based TI heterostruc- tures [232], induced by the helical spin-polarized cur- rent in the TI, which acts as a quantum spin Hall insu- lator. The torque was found to have a similar magni- tude to ferromagnetic/normal/ferromagnetic graphene junctions, and to be immune to changes in geometry. A more recent work studied the spin proximity effect in graphene/TI heterostructures [233], predicting a siz- able anisotropy in the spin lifetime in the graphene layer. Finally, when the spin-orbit coupling is siz- able, the current-induced spin polarisation and spin- Hall effect drastically alter the non-local resistance of graphene [234], which can become negative and os- cillate with distance, even in the absence of a mag- netic field. Even though the results were derived for adatom-functionalised graphene, they are expected to apply generally to 2D systems exhibiting both current- induced spin polarisations and spin-Hall transport. Anomalous Hall effects The 5.3.6. controversy surrounding the origins of the anomalous Hall effect go back nearly seventy years [246]. Three main contributions have been identified, one of which is intrinsic and is now known to be related to the Berry curvature of Bloch electrons. It is associated with a deflection of particle trajectories under the action of the spin-orbit interaction in the band structure of the material, and was shown to be important in TI and TI thin films [247, 248] as well as in longitudinal transport in the presence of a magnetic field [249]. The other two are termed skew scattering and side jump, and are extrinsic, meaning they depend on the disorder configuration. Skew scattering refers to asymmetric scattering of up and down spins, while side jump represents a transverse shift in the wave-packet centre of mass in the course of scattering, also asymmetric between spin up and spin down. Skew scattering and side-jump were originally introduced for an electron with a scalar dispersion scattering off a spin- dependent potential, and were much later generalised to spin-dependent dispersions in the presence of scalar potentials. The complex interplay of spin-dependent dispersions and spin-dependent scattering potentials has been addressed in a small number of papers [250, 251, 37, 39, 38], while the subtle debate surrounding the side-jump in particular is covered extensively in Ref. It should be noted that the intrinsic contribution also has a disorder correction, which in the Kubo formalism includes the ladder diagrams [224] and in the density matrix formalism involves an additional [246]. CONTENTS 23 driving term [28]. This is simply a reflection of the fact that (i) the non-equilibrium correction to the density matrix is an expansion in powers of the disorder strength ni and (ii) the leading term in the expansion , since it is linear in the transport scattering is ∝ n time that is needed to keep the Fermi surface near equilibrium. The next-to-leading term is thus of order n(0) i −1 i . Efforts to identify topological systems exhibiting a quantised anomalous Hall effect continue. Thin-film topological crystalline insulators with ferromagneti- cally ordered dopants can support quantum anoma- lous Hall phases with Chern numbers between -4 and 4 [252]. The QAHE can be induced by an in-plane mag- netisation in atomic crystal layers of group-V elements with a buckled honeycomb lattice according to [253]. For weak and strong spin-orbit couplings, the systems harbor QAHEs with Chern numbers of C=±1 and ±2, respectively, which could be observable at room tem- perature. Very recently, a new material entered the topological insulator stage, when a topological phase transition tuned by an electric field was demonstrated in ultrathin Na3Bi [10], with an accompanying quan- tum spin-Hall effect. Very recently a QAHE at a rel- atively high temperature was reported in a flat-band twisted bilayer graphene sample [254] in which strong correlations result in the system to choosing a single valley and well-defined spin orientation. the bottom of The conductivity of TIs doped with magnetic impurities has been the subject of several papers. In TI the situation is somewhat different depending on whether the surface states are metallic or not, due to the presence of Berry curvature monopoles of opposite polarities at the top of the valence band and the bottom of the conduction band. When the chemical potential is in the gap one expects one TI surface to contribute e2/h to the anomalous Hall conductivity. As soon as the chemical potential passes the conduction band this topological contribution is cancelled by the monopole in the conduction band. What remains is the Fermi surface contribution, which depends on the magnetisation and on the disorder profile. Ref. [255] showed that the surface conductivity of magnetic TIs is anisotropic, and strongly depends both on the direction of the spins of magnetic impurities and on the magnitude of the bulk magnetization. It reaches a minimum when the spin of surface impurities are aligned perpendicular to the surface of TI, because the backscattering probability is enhanced due to the magnetic torque exerted by impurities on the surface electrons. Moreover, Zarezad et al demonstrated numerically that randomly distributed magnetic clusters with temperature-dependent mean sizes are liable to form on the surface of the TI [256]. Figure 11. Anomalous conductivity calculated in the standard non-crossing approximation (σnc xy, dashed line), and including the contribution of the crossed diagrams (σxy, solid line), adapted from [257]. The anisotropic magnetoresistance depends strongly on the spin directions of the magnetic clusters, in a very different way from the case of non-interacting impurities. Undoubtedly one of the biggest revelations in this field has been the fact that diagrams with two intersecting disorder lines, an inherent part of skew scattering on pairs of closely located defects, influences the anomalous Hall effect substantially and reduces the Fermi surface contribution at high densities [257], as shown in Fig. 11. Going beyond the ladder approximation is therefore imperative, and subsequent works also showed that the disorder potential correlation length modifies the result [258], as well as spin-charge correlations in the disorder profile [176]. example, Hall transport has also been investigated under A dynamical Hall effect, optical driving fields. for can be driven by a strong a.c. electromagnetic field as seen in [259] for light of subgap frequency near the absorption edge in a magnetically doped TI. Although the light is off-resonance, in the strong-field regime there is always a finite electron population in the conduction band due to nonlinear effects. A similar analysis was performed on the quantum anomalous Hall effect in intense fields [260], where the Hall conductivity was shown to remain close to e2/(2h) at low fields and low frequencies. At strong fields, the half quantisation is destroyed and the dynamical Hall conductivity displays coherent oscillations as a function of field strength due to the formation of Floquet sub-bands and associated transitions between them. A topic gaining currency at the interface between topological materials and magnetism is the interplay of skyrmions with TI surface physics. Skyrmions 24681012140.00.20.40.60.8−σxy[e2/h]ǫ/mσxyσncxyCONTENTS 24 in which the spin at are topological magnetic excitations with particle- like properties, the core and the spin at the perimeter point in opposite directions. They result from the competition between the Dzyaloshinskii-Moriya interactions and exchange interactions, and give rise to a nonzero Berry curvature in real space. This is associated with an emergent magnetic field, which deflects conduction electrons and causes a topological Hall effect. Skyrmions on TI surfaces provide the opportunity to examine the interplay of real-space and momentum-space Berry In general skyrmion effects cannot be curvatures. features are treated perturbatively as topological missed in such approaches. Araki and Nomura combined a non-perturbative solution to the scattering of massless Dirac fermions with the Boltzmann equation, demonstrating analytically that skyrmions contribute to the anomalous Hall conductivity [261] because at the skyrmion boundary Dirac electrons acquire a phase factor that is absent in the dynamics of Schrodinger electrons. The essential ingredient is the sign change in the out-of-plane magnetic texture between the centre of the skyrmion and the boundary, since in the skyrmion model considered the Berry curvature is zero. Scattering of massive Dirac fermions by a skyrmion in a TI/ferromagnet structure was the subject of a concomitant numerical investigation [262], which concentrated on its effect on the longitudinal conductance. Under certain circumstances the electrical signal due to the skyrmion may be distinguishable from the uniform ferromagnetic background. A concerted effort is being directed towards the understanding of magnonic effects in magnetic TI heterostructures. These include the spin-Seebeck ef- fect, a voltage signal induced in a metallic system by thermally driven spin currents in adjacent mag- netic systems, which was examined in the vicinity of a TI/ferromagnetic insulator interface [263]. In this sys- tem the spin-Seebeck effect is induced by surface elec- trons scattering off the nonequilibrium magnon popu- lation at the surface of the thermally driven ferromag- netic insulator. Similarly, [264] identified magnon-drag thermoelectric effects stemming from the electromotive force induced by magnons and a thermoelectric ana- logue of the anisotropic magnetoresistance. Yasuda et al report a large unidirectional magnetoresistance in TI heterostructures [265], which is attributed to asymmet- ric scattering of electrons by magnons. Its large mag- nitude is due to spin-momentum locking and a small Fermi wave number at the TI surface, and is expected to be maximized around the Dirac point. 5.3.7. Topological protection Topological protection in TI has preoccupied researchers for over a decade. It is variously understood to mean: (i) that the topo- logical surface states are protected against Anderson localisation due to the fact that backscattering is for- bidden; (ii) that the topological surface states cannot be eliminated by any time-reversal preserving pertur- bation; (iii) that the edge states in TIs are topologically protected. Whereas (i) and (ii) are widely accepted we stress that (i) does not imply high mobilities can be achieved in TIs, as was believed for some time, since carriers can scatter through any angle other than π, and in fact TIs continue to have very poor mobilities. The controversy surrounding (iii) affects the QAHE and QSHE and is especially relevant in light of the experimental observation that quantised conductance occurs in very short channels, of a few hundred nm. To begin with, A series of studies have challenged the universality of topological protection. the topological surface states themselves are sensitive to the type of metallic contacts placed on the TI, as demonstrated by ab initio calculations of Bi2Se3 [271]. These reveal that Au and graphene leave the spin- momentum locking mostly unaltered, while Ni, Pd, and Pt strongly hybridise with the TI and relax spin- momentum locking. Size quantisation effects also influence the topological properties [272]. Interestingly, the edge itself may experience spontaneous time- reversal symmetry breaking due to edge reconstruction when a smooth potential is considered, rather than the infinitely sharp theoretical approximation [273], as to mimic the positive-charge distribution on the gate. If this falls smoothly to zero near the edge, the electron density will mimic this by separating the each giving rise to a decrease in edge modes, density. Since the edge modes have opposite chiralities, time reversal symmetry is spontaneously broken. In this case backscattering is enabled and the conductance quantisation of the quantum spin-Hall effect is consequently destroyed, while a spontaneous anomalous Hall effect appears at zero magnetic field. The electron density seeks in Fig. 12. Much of the discussion centres on the effect of impurities on edge state transport. Tanaka et al examined the effect on the conductance of a quantum spin-Hall device of a magnetic impurity, which can backscatter an electron from one edge state to the other [274]. If the Kondo exchange is taken to be isotropic, so that the total spin of the electrons and impurities is conserved, and all electrons moving in the same direction are taken to have the same spin, the correction to the conductance due to this impurity vanishes in the dc limit, in contradiction Anisotropic exchange to an earlier paper [275]. introduces corrections that can be sizable above the Kondo temperature, but are suppressed as T → 0. The treatment was generalised by Altshuler et al, CONTENTS 25 interactions, short-range nonmagnetic impurities act as noncollinear magnetic scatterers, which enable strong backscattering and cause deviations from quantisation even at zero temperature [279]. Due to fluctuations in the donor density, doping TIs tends to create a non-uniform potential landscape consisting of electron and hole puddles, especially It is natural to ask dangerous at small energies. what effect these have on edge transport. In [280] a puddle is modeled by a quantum dot, coupled to the helical edge states by tunnelling, and quantum dots with even numbers of electrons are considered. Elastic processes involving electron dwelling in the dot do not lead to any backscattering. Yet inelastic backscattering processes are enhanced by electron dwelling in the dot by increasing the time electrons interact with each other. At temperatures lower than the quantum dot energy spacings, the conductance correction depends strongly on the position of the Fermi level with respect to the dot energy levels. The enhancement of the resistance is much stronger for dots with an odd number of electrons, due to the Kondo effect [281], but the corresponding temperature dependence is relatively weak. Indeed, current experiments see no signature of Kondo transport [282]. 6. Valley-dependent phenomena Whereas TMDs have a massive Dirac spectrum, unlike TIs, this describes a lattice pseudospin degree of freedom rather than an angular momentum stemming from the real spin. Nevertheless, TMDs have two degenerate valleys, which are related by time reversal, so that the masses in the two valleys have opposite signs. Unlike graphene they typically have sizable spin- orbit interactions. Large spin splittings were identified in the conduction band of TMD monolayers in [283], while the MoS2-WS2 heterojunction was shown to have an optically active band gap [284] with the lowest energy electron-hole pairs spatially separated and living in different layers. The spin-orbit interaction has a noticeable effect on the optical conductivity [285, 286]. Given that the spin is essentially locked to the valleys it is not clear that the two can be manipulated independently, yet accessing the elusive valley degree of freedom is fascinating from a fundamental standpoint. The massive Dirac spectrum gives rise to the possibility of manipulating the valley degree of freedom. For example, the analogue of the anomalous Hall effect in doped TMD monolayers is the valley Hall effect, which has been detected in MoS2 [287] and has generated considerable excitement. Physically it corresponds to an anomalous Hall effect with different signs for different valleys. There is no net charge Figure 12. Edge reconstruction and topological protection, adapted from from [273]. Panels (a) -- (c) describe the schematics of the results for three different distributions of the confining positive charge (light orange), characterised by w, the length scale over which it decays to zero. The edge modes are marked by broken blue (spin up) and solid red (spin down) lines. Panels (d) -- (f) depict the occupations of the electronic states, using the Hartree-Fock approximation, demonstrating a single drop in density for a sharp edge ( w = 0) in (d), spin separation for smoother edge ( w = 5) in (e), and an even smoother edge ( w = 20) in (f). Y denotes the position of state, in units of the effective magnetic length; Y = 0 is the center of the density drop. Panels (g) and (h) depict the same distributions as in (d) and (f), respectively, using exact diagonalisation. Panel (i) depicts the edge spin magnetisation as a function of the slope of the positive- charge density, suggesting a continuous phase transition. who considered scattering by a disordered chain of Kondo impurities [276]. As the authors point out, in disordered systems with strong spin-orbit interactions it is unlikely that any component of the total spin is conserved. When this is taken into account, backscattering processes emerge that persist down to absolute zero. The edge electrons experience Anderson localisation for an arbitrarily weak anisotropy in the coupling to the spin impurities provided the sample is long enough. Interactions between electrons cannot destroy the localisation, unless they happen to be strongly attractive. Black-Schaffer and Nazarov focused on the coupling between surface and bulk states in the presence of strong non-magnetic potential impurities, which create localised resonances appearing at ever lower energies as the impurity strength is increased [277]. At large strengths the resonance goes through the Dirac point, causing two Dirac points to emerge on both sides of the resonance. Both the surface states and the resonances penetrate approximately 10 layers into the sample, enabling second-order bulk- assisted scattering processes, which act to destroy the topological protection. Coupling between the edges and the bulk was investigated from a different angle in a subsequent paper [278], where edge and bulk states were considered to be at the same Fermi energy. In this case backscattering between the two leads to Anderson localisation of both edge and bulk states. Finally, in the presence of even weak electron-electron Occupation, Hartree-Fockabc-2.5 -1.2 0 1.2 2.5defgh-10 -7.5 -5 -2.5 0 2.5 5 7.5 10iOccupation, Hartree-FockOccupation, Hartree-FockOccupation, Exact diagonalizationHartree Fock-15 -10 -5 0 50.61.00.20.40.81.0YY-15 -10 -5 0 50.61.00.20.40.81.0Y< S >101468122400 2 4 6 8 10 12 14 exact diagonalizationYYww = 5w = 20ZCONTENTS 26 current, because the anomalous Hall currents from the two valleys cancel each other out, but electrons from different valleys flow to different sides of the sample generating a valley polarisation, which can be detected using circularly polarized light. In bilayer MoS2 this was shown to be controllable electrically using a top gate voltage, which breaks the inversion symmetry of the bilayer [288]. The valley Hall effect generates a non-local resistance, which has a non-trivial dependence on the longitudinal resistivity, weakening at large valley Hall angles [289], where the valley Hall angle is determined by the ratio of the valley Hall conductivity and the longitudinal charge conductivity. Because the valleys are spin polarised the valley Hall effect is typically accompanied by a spin-Hall effect [290, 291], valley currents tend to be spin- polarised [292], and spin and valley polarisations are generated concomitantly, although they can be distinguished at high magnetic fields [293]. Likewise, spin and valley noise are coupled [294], with fluctuations in the Faraday rotation signal being connected to the valley degree of freedom as well as to the spin, a fact that is ascribed to intervalley scattering processes. Again, given that spin noise is sensitive to an applied magnetic field, spin and valley dynamics may be distinguished in certain regimes. On a similar note, TMDs also exhibit strong hyperfine interactions, which can create a feedback mechanism in which spin-valley currents generate significant dynamical nuclear polarization which in turn Zeeman shifts excitonic transitions out of resonance with an optical driving field, saturating the production of spin-valley polarisation [295]. Similar findings have been reported theoretically in Bi monolayers Bi2XY, where X,Y ∈ {H, F, Cl, Br, or I} [296], where a staggered exchange field is introduced by doping with transition-metal atoms or by magnetic substrates. Unsurprisingly these findings have stimulated a lot of activity in optics [297, 298], and various schemes have also been proposed for controlling and enhancing the spin and valley polarisations and currents by employing electric and magnetic fields [293, 299] and optical techniques [300]. A half-quantised valley Hall effect has been predicted when the chemical potential lies in the mass gap, driven by the same Berry curvature mechanism as the quantised anomalous Hall effect in TIs [301]. By mapping the system onto the Landau-Zener problem, the authors argue this response is dominated by bulk currents arising from states just beneath the gap rather than by edge modes, as the latter are not topologically protected, as in TI, and may be absent. The potential gradient due to the external electric field divides the system into three regions, of which the middle region is gapped and the surrounding regions have carriers in the conduction and valence bands respectively. Upon reflection from the central gapped region carriers are believed to experience side jumps leading to a Hall effect. A dissipationless response is found even when topologically protected edge modes are absent, and is independent of the gap size. To date this claim remains unverified experimentally. In TMD bilayers the situation is somewhat different [302]. The Berry curvature has sizable contributions from both the intralayer and the interlayer couplings, the latter leading to a dependence on the stacking configuration and enabling tunability in double gated devices. The valley and spin Hall conductivities are not quantised, but can change sign as a function of the gate electric field. Structures that may host topologically protected states supporting persistent spin or valley currents include interfaces where the Dirac mass changes sign in 2D Dirac materials with spin-orbit coupling [303]. In this case the topologically protected states are so- called Jackiw-Rebbi modes with a linear dispersion, supporting spin and valley currents parallel to the interface. Whereas the valley Hall effect has been responsible for the bulk of transport research on TMDs, magneto- transport has begun to attract attention. Seeing as the valley pseudospin and the magnetic field are both odd under time reversal one may expect magneto- transport phenomena without a counterpart in single- valley systems. Sekine and MacDonald demonstrated that the interplay between the Berry curvature, a perpendicular magnetic field, and disorder scattering in TMD monolayers gives rise to a longitudinal magnetoconductivity contribution that is odd in the valley pseudospin and odd in the magnetic field [304]. For this contribution to be visible a valley polarisation must exist in the system. The dependence of the magnetoresistance on the magnetic field changes from quadratic to linear when a finite valley polarisation is induced by optical pumping. 7. Probing Fermi arcs in Weyl semimetals Weyl semimetals harbor unusual surface states known as Fermi arcs, with each arc connecting the surface projections of two Weyl nodes of opposite chirality, as shown in Fig 13. For a basic understanding their dispersion can be approximated as linear and, as shown in [9], for a surface in the xy-plane may be written as εk = ±vky, where v is the magnitude of the velocity, and the line connecting the two Weyl nodes is (cid:107) kx. The Fermi arc states exist for −k0 ≤ kx ≤ k0, where ±k0 represent the locations of the Weyl nodes. Note that: (i) the two Fermi arcs are only degenerate at εk = 0; (ii) Fermi arcs on opposite surfaces have opposite CONTENTS 27 are gapless and the rare states are only quasilocalised. This coupling is non-perturbative and gives the arcs spectral weight in the bulk so that they are no longer bound to the surface and may no longer be said to be topologically protected even for weak disorder. In- terestingly, the surface chiral velocity persists even for strong disorder, making the Fermi arc visible in spec- troscopic measurements. Experimentally, quasiparticle scattering and interference was imaged on the surface of the WSM TaAs [307], providing spectroscopic evi- dence of Fermi arc states. Nevertheless, the scatter- ing wave vectors observed experimentally are consis- tent with theoretical predictions that assume particle propagation through the bulk of the sample in addi- tion to propagation on the surfaces. Indeed, aside from hybridisation with rare states, the connection between surface and bulk provided by the Fermi arcs has highly non-trivial consequences not only for scattering and in- terference, but also for transport. To begin with, scattering of electrons between the surface and bulk states caused by inhomogeneities introduces dissipation in Fermi arc transport [308]. In a 1D description that neglects surface-bulk coupling, quenched disorder effects result in a single phase factor that is odd under exchange of spatial variables. Consequently, disorder effects disappear from the current-current correlation function, which would imply a dissipationless conductivity. Disorder averaging, however, is still responsible for the finite width of the surface states, without which the longitudinal conductivity would diverge. The authors of [308] show rigorously, however, that an impurity scatters surface waves into the bulk, resulting in dephasing of the Fermi arc states and dissipation. The deeper conclusion is that generically an effective theory of surface states in the presence of disorder is not well-defined for gapless systems. longitudinal The geometry of the Fermi arc dispersion is also important [309]. A straight arc eliminates the impact of electron-phonon scattering on surface transport because scattering only occurs between states with the same velocity along the direction of the current. Scattering off surface disorder is also suppressed in this geometry, such that for strong disorder a straight arc yields a surface conductivity 1-2 orders of magnitude larger than a TI. This is traced to the different hybridisation strengths between surface and bulk in WSMs and TIs. In the absence of a magnetic field Fermi arcs may alter the anomalous Hall response of WSMs by introducing a residual contribution, with the result depending on the degree of tilting of the arcs on opposite surfaces [310]. In an external magnetic field it emerges [311] that closed orbits can be formed in which Fermi arc states on opposite surfaces are connected by Figure 13. Schematic of the Fermi arcs on opposite WSM surfaces, adapted from [9]. (a) Top surface (b) Bottom surface. The red (blue) points are the gapless points, which have positive (negative) monopole charges for the Berry curvature. velocities; and (iii) in contrast to TI surface states, Fermi arc states are open, effectively disjoint segments of a 2D Fermi surface, connecting the Fermi surfaces for carriers with opposite chiralities that otherwise appear to be disconnected. As the system evolves from a Weyl semimetal to a TI with decreasing thickness, the arcs on opposite surfaces merge into a surface Dirac cone [9]. In a similar manner to TI surface states, Fermi arc wave functions in real space extend into the material. If the sample is very thin, in analogy with a TI thin film, the wave functions corresponding to the Fermi arcs on the top and bottom surfaces overlap significantly, which enables back scattering between certain pockets of the Brillouin zone. This is expected to enhances Friedel oscillations, as seen in the local density of states [305]. Since Fermi arcs couple points in the Brillouin zone with opposite Berry curvatures, it has been be- lieved that their existence is topologically protected. Recent studies however cast doubt on this belief, and to date it remains an open question. For example, Wil- son et al [306] argue that Fermi arc states in WSMs hybridise with Lifshitz rare states in the bulk to a much greater degree than TI surface states, since the latter exist in the bulk gap and are localised whereas WSMs CONTENTS 28 trajectories that traverse the bulk of the sample. As a result quantum oscillations in e.g. the density of states can be observed in magnetic fields up to a critical value, given by the thickness of the sample. Following the prediction, Shubnikov-de Haas oscillations involving Fermi arc states were indeed detected in Cd3As2 [312]. In strong magnetic fields, Fermi arcs can also give rise to a 3D quantum Hall effect, with complete loops formed via wormhole tunneling assisted by the Weyl nodes [313]. A peculiar phenomenon present in WSMs in magnetic fields is the chiral magnetic effect, the generation of an electrical current parallel to a magnetic field when the Weyl nodes are offset in energy by a finite amount. Fermi arcs are expected to make a finite contribution to the chiral magnetic effect opposite in sign to the bulk contribution, which change the sign of the overall result [314]. This holds even in the infinite-system limit because, even though the number of surface modes decreases, the remaining modes are more sensitive to magnetic flux. Research is burgeoning on the unusual optical properties of Fermi arc states. They interact strongly with light and have a large optical conductivity for light polarised transversely to the arc [315]. They affect the plasmon dispersion, which is already unconventional in bulk WSMs [316, 317]. There the axion term lifts the degeneracy of the three gapped plasmon modes at q = 0 [316]. This because WSMs are gyrotropic, meaning their dielectric tensor is asymmetric, and the degree of asymmetry is proportional to the separation between the nodes, which acts as an effective applied magnetic field in momentum space. In magnetic systems with broken time reversal symmetry Fermi arc plasmons are chiral, with constant frequency contours that are open and hyperbolic [318]. Their dynamics is also strongly linked to quantum non-local effects, for example they decay by emitting electron-hole pairs in the bulk [319]. 8. Non-linear electrical response Undoubtedly the most exciting novel development in topological materials transport has been the take-off of non-linear electrical effects, which can be enabled by the lack of inversion symmetry, the application of a magnetic field, or the presence of a valley polari- sation. An early work illustrated the role of mirror planes in determining the charge and spin response [320], yet the community is still only beginning to ap- preciate the richness and variety of physical phenom- ena that emerge when higher-order responses in the applied fields are allowed. Geometrical phase effects in first order response, as well as their interplay with disorder, are relatively well understood. However in the second order response most of the ground work re- mains to be done. The quantity of interest is the next term in perturbation theory beyond linear response to an electric field, which can be formulated diagrammat- ically, semiclassically, or in terms of the density matrix. The non-linear optical response, particularly strong in TMDs, encompasses conceptually new phenomena in- cluding second harmonic generation and nonreciprocal, rectification and shift currents. Non-reciprocal refers to phenomena that have a built-in bias direction, such current flow in a p − n junction. These relatives of the photogalvanic and photovoltaic effects are frequently encoded by the Berry phase, the toroidal moment, and the magnetoelectric monopole, or may have extrinsic origins such as magnon scattering. The relevant con- cepts are beautifully summarised in Ref. [321]. Growth in this field has been spurred by spectacular experi- mental developments, motivated by the possibility of detecting the relevant response by scanning higher har- monics of the applied frequency. One of the grand aims is to find a Hall effect in time-reversal symmetric sys- tems: two papers have reported a nonlinear Hall effect in bilayer/few-layer WTe2 [322, 323]. In [322] the non- linear Hall effect results in a much larger transverse than longitudinal voltage, with a nonlinear Hall an- gle of nearly π/2, which may have topological origins [324]. The non-linear Hall effect is generally extrinsic in time-reversal invariant systems, but can be intrinsic if time-reversal symmetry is broken [325, 326]. The static non-linear Hall conductivity contains an intrinsic contribution proportional to the Berry curvature dipole in reciprocal space, that is, the term kΩk, where Ωk is the Berry curvature [327, 328, 329]. The nonlinear Hall effect arising from the Berry curvature dipole in TMDs with time-reversal symmetry was examined in Ref. [330], which showed that such a current is present when only one mirror line exists, while in certain TMD phases a finite Berry curvature dipole emerges when strain or electrical displacement fields are applied. WSMs are also expected to be excellent candidates for nonlinear effects because of their large Berry curvature concentrated near the Weyl points, and in a related study, the Berry curvature dipole was investigated in WSMs [331], concluding that type-II Weyl points, having a strong tilt, were preferable to type-I. In this vein, Nandy and Sodemann [41] calculated the non-linear Hall conductivity of two-dimensional tilted Dirac fermions using a multi-band quantum Boltzmann equation, identifying disorder contributions in addition to the intrinsic Berry curvature dipole. Recent work by the same group [332] reveals that the rectification current obeys a sum rule controlled by the Berry connection. Rectification at relatively high frequencies has also been studied in Ref. Non-linear Hall effects can also be induced by disorder [334]. In the context of tilted Dirac cones we note the [333]. CONTENTS 29 revealing renormalisation group work of Ref. [335] on the Coulomb interaction and quenched disorder in tilted TIs. Along the tilting direction a random scalar or vector potential dynamically generates a new type of disorder, dominant at low energies, which turns the system into a compressible diffusive metal, with the fermions acquiring a finite scattering rate. The band- touching point is replaced by a bulk Fermi arc in the Brillouin zone. The consequences of these features for transport remain to be determined. An in-plane magnetic field has a subtle effect on the non-linear electrical response of a hexagonally warped TI at frequencies very close to the DC limit [336]. Whereas an in-plane magnetic field merely shifts the origin of a Dirac cone with no physically measurable effect, when the Hamiltonian contains a warping term the effect on the spectrum is non- trivial, and a strong non-linear response results, termed bilinear electromagnetic response. Its sign and the magnitude depends sensitively on the orientation of the current with respect to the magnetic field as well as the crystallographic axes, so that the spin texture of the topological surface states could be mapped via a transport measurement. The bilinear magnetoelectric resistance was measured in hexagonally warped TIs in [337], On the other hand, [208] a second harmonic Hall voltage was in Ref. detected in the presence of in-plane magnetic field and magnetisation in TI heterostructures, believed to be due to asymmetric magnon scattering. In a similar manner to a magnetic field, a time-reversal breaking valley polarisation allows second harmonic generation even in centrosymmetric crystals, and this in turn can provide a direct measure of the valley polarisation [338]. Fig. 14. cf. effect in ferroelectric materials, Noncentrosymmetric crystals are anticipated to exhibit a dc photocurrent in the nonlinear optical response [339]. This so-called shift current has attracted intensive attention as part of the bulk photovoltaic in the quest for efficient solar cell paradigms. Its fundamentally origin is the fact that, as an electron is excited by light from the valence band into the conduction band, its centre of mass changes due to the difference in the value of the Berry connection in the two bands. The final expression can be formulated in terms of the Berry curvature dipole in momentum space discussed above and is gauge Since the centre of mass is shifted the invariant. effect has been termed a shift current. It is being pursued in topological materials as well. Kim et al [340] identified a sizable shift current generated in hexagonally warped TIs by linearly polarised light. A related nonlinear spin current also exists in TIs, which can be excited by THz light [341]. When inversion Figure 14. Bilinear magnetoelectric resistance, adapted from [337]. (a) Hexagonally warped energy dispersion for the surface states with Fermi surface lying in the conduction band. (b) Hexagonally warped spin texture at the Fermi contour of the surface states. (c) Variation of the electron distribution along the k-axis parallel to the applied electric field E: δf1 (blue curve) and δf2 (yellow curve) are the corrections to the equilibrium distribution of first and second order in the electric field, respectively. Solid arrows represent the excess of electrons with spins along the arrow direction, and hollow arrows represent depletion of the same. (d) When an electric field E is applied along a certain direction in k-space (dash-dotted line), a non- equilibrium spin current Js(E2) is generated at the second order of the electric field, due to spin-momentum locking. (e) and (f) When an external magnetic field is applied, the nonlinear spin current is partially converted into a charge current J(cid:48) e(E2): a high-resistance state is reached (e) when the magnetic field is antiparallel to the spin direction of the electronic states with k (cid:107) E, while a low-resistance state is reached (f) when the magnetic field is parallel to that spin direction. symmetry is broken but time-reversal symmetry is preserved a nonlinear anomalous Hall effect emerges in certain TMDs [342], while a dissipationless nonlinear anomalous node conductivity is also expected in WSMs [343]. 9. Chiral superconductivity, Majorana edge modes and related phenomena In the preceding sections, we have dealt with normal topological materials in which no pairing interaction among electrons However, pairing interactions bring about new topological phenomena. Some earlier theoretical examples pertain to superfluid helium 3 (3He) [344, 345], anyon superfluids [346], takes place. !"#0!"%#<0!"%#>0!"(#>0!"(#<0−#*#*+,-+(eHigh resistance,:;+(<+Low resistance,:;+(<+fcdab@*TSS#C#DKMGCONTENTS 30 d-wave superconductors [347, 348, 349], and the fractional quantum Hall effect [350]. The theory [350] has illustrated the concept of a topological superconductor as a system with a bulk pairing gap and a gapless Majorana mode at the boundary or on a topological defect. Such emergent Majorana modes obey the quantum statistics of non-Abelian anyons [351, 352, 353], offering a route to braiding-based topological quantum computation [354]. Research into topological superconductivity has intensified with the discovery of topological insulators and related topological materials, with a flurry of theoretical as well as experimental activity that followed. Distinct classes of superconductors have been identified (see, e.g., Refs. [355, 356, 357]). Their properties have been a subject of several review articles [358, 359, 360, 361, 362, 363, 364, 365]. topological Still, the field of topological superconductivity is growing fast, having in some areas advanced well beyond the existing review literature. This pertains, in particular, to the latest developments in chiral superconductivity, Majorana edge modes, the fractional Josephson effect as well as unconventional Cooper pairing in topological materials. These topics define the primary scope of our review article. We have endeavoured to mention key ideas and developments in the overlapping areas such as specific models and realisations of Majorana zero modes, their tunneling spectroscopy and related transport phenomena. These topics have been covered in several review articles. Further details of the underlying physics and a survey of the results can be found easily in the cited references. 9.1. Chiral superconductivity. A primer (cid:35) (cid:34) Superconductivity originates from attractive pairwise interactions between electrons in a metal. As is common in many-body physics, the pairing interaction can be treated in the mean-field approximation, allowing the description of a superconducting state by the Bogoliubov- de Gennes (BdG) Hamiltonian (cid:34) (cid:35) uk vk ∆k Hk † k −H∗ ∆ −k . , (20) Ψk = Hk = It is a 2 × 2 matrix in which Hk is a single-particle Hamiltonian of the normal system, while the off- † diagonal entry ∆k and its hermitian conjugate ∆ k account for the pairing interaction. The state Ψk has two (Nambu) components, uk and vk, being a particle- and a hole-like wave functions of the normal system, respectively. Coming from the pairing interaction, the matrix structure of the BdG Hamiltonian is crucial for emer- gent topological superconductivity. A paradigmatic example is a 2D superconductor with an odd k - parity Figure 15. Schematic of an equal-spin p - wave electron pairing assumed in Eqs. (21) and (22). gap function [350, 351]: ∆k = ∆(cid:48)(kx + iky), (21) where the constant ∆(cid:48) is taken real. Equation (21) describes equal-spin pairs in the orbital p - wave state with the quantum number m(cid:96) = 1 (see also Fig. 15). It has a partner with the quantum number m(cid:96) = −1. These are the 2D analogues of the A-phase of superfluid 3He [366, 367]. We shall see later that the kx + iky - pairing (21) is also an effective model for the hybrid structures of quantum anomalous Hall insulators and conventional superconductors. As for the normal-state Hamiltonian, here we choose the simplest, parabolic band conductor with Hk = k2 2m − µ, where µ is the chemical potential. Then, the BdG Hamiltonian reads Hk = k2 2m − µ ∆(cid:48)(kx − iky) −( k2 ∆(cid:48)(kx + iky) 2m − µ) , (22) (cid:34) (cid:35) (cid:21) (cid:20) or, in the basis of the Pauli matrices τ = [τ1, τ2, τ3], k2 2m − µ ∆(cid:48) kx, −∆(cid:48) . ky, Hk = τ · dk, dk = This equation resembles the spin Hamiltonian of a ferromagnet, with vector dk playing the role of the magnetization. In this case, dk defines the texture of the Nambu pseudospin τ /2 in the 2D momentum space (see Figs. 16 and 17). (23) The topology of the pseudospin texture can be characterized by the winding number (cid:90) (cid:18) ∂nk (cid:19) C = 1 4π nk· ∂kx × ∂nk ∂ky dkxdky, nk = , (24) dk dk in differential geometry, is the first Chern which, invariant of a principal U (1) bundle over a torus [149, 368]. Geometrically, C is the number of times the unit vector nk sweeps a unit sphere as k covers the entire momentum space. The Chern invariant acquires nontrivial values C = ±1 under condition µm > 0. In this case, the configuration of the vector dk defines the skyrmion, a topological defect that compactifies the physical space on a sphere (see Fig. 16). The superconducting state is, therefore, topologically nontrivial as opposed to the case µm < 0 (Fig. 17) in which no d - vector winding takes place, hence C = 0. CONTENTS 31 Figure 16. Schematic of the Nambu pseudospin texture in a topological superconducting state for µm > 0 in Eq. (23). The texture has the skyrmion topology with a nontrivial Chern number C = ±1. The above considerations have clear parallels with Chern insulators and the quantum anomalous Hall effect in zinc-blende materials (see recent review in Ref. In a Chern insulator, C corresponds to the TKNN invariant [149, 368] and can be obtained from the Berry curvature of the electronic bands as [369]). C = 1 2 [sgn(−µ) − sgn(m)]. (25) This analytical result confirms the picture of the winding of dk in momentum space shown in Figs. 16 and 17. As in Chern insulators, the nontrivial number C indicates a chiral gapless state at the edge of a TS (see Fig. 18). Furthermore, due to the generic particle- hole symmetry CH(x, y)C† = −H(x, y), (26) the edge modes of TSs mimic Majorana fermions of the relativistic quantum theory. Above, H(x, y) is the real-space BdG Hamiltonian, and C is the particle-hole conjugation operation. In non-topological superconductors, C converts a particle into a hole and vice versa. The Majorana edge state, ΨM (x, y, t), transforms to itself under C: C M (x, y, t) = ΨM (x, y, t), Ψ (27) (cid:34) ΨM (x, y, t) = (cid:35)(cid:88) that is, emergent Majorana fermions in TSs are particles and holes at the same time. For the BdG Hamiltonian (23), the conjugation operation is C = τ1K, so the Majorana edge solution is a real eigenstate of τ1. For a "hard-wall" boundary [369], it is given by (cid:104) κ−(kx)y(cid:105) Nkx × cos [kxx − E(kx)t/] . (29) It is assumed that the TS occupies the half-space y ≤ 0, and the edge mode propagates along x. It has a linear dispersion E(kx) = ∆(cid:48) κ+(kx)y − e (30) (28) kx, 1 1 kx e Figure 17. Schematic of the Nambu pseudospin texture in an ordinary superconducting state for µm < 0 in Eq. (23). In this case, the Chern number is trivial, C = 0. (cid:112) m2∆(cid:48)2 + k2 and is localized on the scale given by κ± = m∆(cid:48) x − k2 ± F . The coefficients Nkx of the sum The (28) are the normalizing factors. in Eq. normalizability requires that kx < kF . 9.2. Search for intrinsic chiral superconductivity and related phenomena It is a quasi - 2D material The model discussed above illustrates the principal possibility and essential attributes of the topological chiral p - wave superconductivity. Perhaps the first material for which such a possibility has been considered is the layered perovskite ruthenate Sr2RuO4. in which ruthenium oxides tend to become ferromagnetic, which favours a spin-triplet order parameter [370, 371]. If the spin-triplet superconductivity has a definite chirality, the time-reversal symmetry should be broken below the critical temperature Tc. The expectation that Sr2RuO4 harbors such chiral superconductivity was strengthened by the muon spin rotation experiments [372] which detected internal magnetic fields below Tc. While the scenario of the chiral p - wave superconductivity in Sr2RuO4 may be debatable, the tunneling experiment [373] has reported an enhanced zero-bias conductance consistent with the existence of the boundary modes. The zero-bias peak in the tunneling spectrum was attributed to the surface Andreev bound states (ABSs), although this would not generally tell whether the order parameter is chiral Figure 18. Schematic of a chiral Majorana edge state in a TS [see also Eqs. (27) -- (29)]. kxkydkEkxkyEdkCONTENTS 32 or helical [374]. Initially, the theory of the tunneling spectroscopy of the surface ABSs was developed for d-wave superconductors [347, 349]. Reference [375] has extended the theory to Sr2RuO4, using a three- band model and the recursive Green's function method. The tunneling spectra with both zero-energy peaks and zero-energy dips were found, depending on the spatial dimensionality of the model and the presence or absence of SOC. Another recent theoretical work [376] has proposed to identify the chiral p-wave superconductivity by the electronic states on a domain wall between the order parameters with opposite If the superconducting order chiralities (kx ± iky). parameter breaks the time-reversal symmetry, domains with different chiralities and opposite edge currents are expected to form akin to the ferromagnetic domains. An experimental attempt to detect such chiral domains has been reported in Ref. [377]. A recent review on the point-contact spectroscopy as a means of detecting topological superconductivity is given in Ref. [378]. Beside the chiral phase in Sr2RuO4, there has been a theoretical proposal for a topological crystalline superconductor phase in this material [379]. It is characterized by a pair of Majorana modes each protected by the mirror symmetry of the Sr2RuO4 crystal structure. Reference [379] discussed a magnetic-field-induced transition into the superconducting state accompanied by a rotation of the Balian-Wertheimer d vector parametrizing the triplet order parameter. topological crystalline If, however, [388] ‡. Apart from Sr2RuO4, a number of other candidate materials to host topological superconductivity have been identified in the past decade, most notably CuxBi2Se3 [380, 381, 382], Sn1−xInxTe [383, 384] as well as some noncentrosymmetric superconductors in which the p - wave gap is larger than the s - wave one [385, 386, 387]. As opposed to the chiral superconductivity, these materials are expected to host time-reversal symmetric (helical) topological phases the time-reversal symmetry is broken by an external magnetic field, noncentrosymmetric low-dimensional superconductors may turn into chiral TSs [389, 390, 391]. Normally, this requires a magnetic field that is by far larger than the upper critical field Hc2. The way to overcome this problem is to apply the field parallel to the basal plane, reducing the Meissner currents in favour of the Zeeman splitting. The theory [389] has examined such a possibility for the superconducting interface between LaAlO3 and SrTiO3, assuming the Rashba SOC and a three-band model. More recently, 1D structures at LaAlO3/SrTiO3 oxide interfaces have been found to support Majorana modes [390]. In fact, metallic superconducting films ‡ The state of the art is well captured in the review article [363]. Figure 19. Schematic of a hybrid structure created by placing a singlet superconductor (S) on top of a topological insulator (TI) material. Andreev reflection at the S/TI boundary gives rise to the superconducting proximity effect. grown on a substrate and subject to an in-plane magnetic field may have all the ingredients required to achieve chiral superconductivity, i.e. Copper pairing, broken inversion symmetry, and broken time-reversal symmetry. This expectation has been supported by the density functional theory calculations for ultrathin Pb and β-Sn [391]. 9.3. TI materials as platform for topological superconductivity and Majorana fermions The above discussion pertains to intrinsic supercon- ductivity when the symmetry breaking order parame- ter occurs spontaneously below a certain Tc (∼ 1.5 K for Sr2RuO4). An alternative to that is the induced superconductivity which occurs in a normal conduc- tor brought into electric contact with an intrinsic su- perconductor (see also Fig. 19). Although the nor- mal conductor has no pairing interaction of its own, it acquires the superconducting correlations through the proximity effect. Microscopically, this can be under- stood in terms of Andreev reflection [392, 393] whereby a particle in the normal system is converted into a hole (and vice versa), while a Cooper pair passes through the interface, as sketched in Fig. 19. The particle- hole conversion is most efficient when the thickness of the normal region is smaller than the phase coherence length. system. the normal The superconducting proximity effect offers an attractive alternative to intrinsic superconductivity, as topological phases can be "engineered", using broken symmetries of A prominent example is the theoretical proposal [394] for a chiral TS and Majorana fermions at the surface of a 3DTI proximitized by a conventional (singlet In the past s-wave) decade, impressive progress has been achieved in fabricating and characterizing hybrid structures of superconductors and TI materials [395, 396, 397, 398, 399, 400, 401, 402, 403, 404, 405, 406, 407, 408, 409, superconductor. CONTENTS 33 410, 411, 412, 413, 414, 415, 416, 417, 418, 419, 420, 421, 422, 423, 424, 425, 426, 427, 428, 429, 430, 431, 432, 433, 434]. Most of these experiments have used the tetradymite compounds Bi2Se3 and Bi2Te3, ternary tetradymites (e.g., Bi2Te2Se) or later generations of Bi-based compounds such as Bi2−xSbxTe3−ySey [435]. Other types of the TI materials include thick strained HgTe layers [403, 405, 414, 417], HgTe quantum wells [408, 421, 422, 423], topological crystalline insulator SnTe [428, 429], and Cr-doped (Bi,Se)2Te3 thin films [424, 431]. The latter are magnetic topological insulators that, in the absence of the superconducting pairing, exhibit the quantum anomalous Hall effect [436]. That is, Cr-doped (Bi,Se)2Te3 thin films with Nb contacts, such as in Refs. [424, 431], are prototypes of the quantum anomalous Hall insulator (QAHI) - superconductor devices. [437] has A related paper reported an ob- servation of 2D topological superconductivity in a Pb/Co/Si(111) structure. It was modeled as a Rashba system with a mixed singlet-triplet pairing and an ex- change interaction. In fact, TI materials with induced superconduct- ing and magnetic orders have long been a fertile ground for theoretical modeling of the chiral TS and related phenomena (see, e.g., [394, 438, 439, 440, 441, 442, 443, 444, 445, 446, 447, 448, 449, 450, 451, 452, 453, 454, 455, 456, 457, 458]). This includes the issues of the observability of neutral Majorana fermions in quantum interferometry [438, 439], fractional Joseph- son effect [440], resonant Andreev reflection [441], mag- netic proximity effect [442], backscattering processes [446], current noise [448], tunneling spectroscopy [450], crossed Andreev reflection [452], half-integer longitu- dinal conductance [453], edge-state-induced Andreev oscillations [456], just to name a few. In particular, in the series of papers [444, 446] and [453], a kx + iky phase with a chiral Majorana edge mode has been discovered theoretically in QAHI/superconductor structures. Its expected transport signatures are Majorana backscattering and the half-integer longitudinal conductance. In the following, these ideas are discussed in some more detail. 9.4. Chiral TS with a single Majorana edge mode in QAHI/supercondictor structures nian for a magnetic thin film, H =(cid:80) To set the scene, we define the normal-state Hamilto- † kHkck, where k c Hk is an effective 4-band Hamiltonian [453]: Hk = A(σxky − σykx)ν3 + Mkν1 + λσz − µ, k↓]. Here, the operator ct,b and ck = [ct kσ annihilates an electron with momentum k and spin σ =↑,↓, and the superscripts t and b refer to the top and bottom surface layers of the film, respectively. σi k↑, cb k↓, cb k↑, ct (31) (with i = x, y, z) and νj (with j = 1, 2, 3) are Pauli matrices in spin and layer subspaces, respectively. The first term in Eq. (31) is the Hamiltonian of the two surface layers, where the constant A (assumed positive throughout) determines the surface velocity. The second term Mkν1 introduces the coupling between the layers, opening a hybridization gap at the Γ point (k = 0). The hybridization energy is Mk = M0 + M1(k2 x + k2 y), where M0 yields the half of the gap between the conduction and valence bands, while M1 accounts for the band curvature. The third term λσz is the mean- field exchange Hamiltonian due to the ferromagnetic ordering, with λ being the exchange energy. The Hamiltonian (31) decouples into two Chern subsystems with the Dirac masses λ ± Mk, where the signs ± are dictated by the time-reversal symmetry. The corresponding Chern number is the sum of the Chern numbers for the two subsystems [cf. Eq. (25)], (cid:88) ν=± C = = 1 2 1 2 [sgn(λ + νM0) − sgn(νM1)] [sgn(λ + M0) + sgn(λ − M0)]. (32) (33) (34) (cid:34) (cid:35) topological number A similar is encountered in Haldane's model [16]. Taking for simplicity M0 > 0 and λ > 0, we see that the system undergoes a topological phase transition from an ordinary insulator with C = 0 for λ < M0 to a QAHI with C = 1 for λ > M0: (cid:26) 0 λ < M0, 1, λ > M0. C = In the latter case, there is a chiral edge mode realizing a gapless Dirac fermion. In contact with a conventional superconductor, placed on top of the structure, the magnetic film can be described at low energies by the BdG Hamiltonian (20) with the singlet pairing ∆k = ∆tiσy 0 0 ∆biσy , (35) 11 for a microscopic theory of the see also Sec. superconducting proximity effect. Here, ∆k is a matrix in the layer subspace where ∆t and ∆b denote the pair potentials in the top and bottom layers. As found in Ref. [453], the essential condition for realizing a chiral TS is to have unequal pairing amplitudes ∆t (cid:54)= ∆b. This point is best illustrated in the special case where ∆t = ∆, ∆b = −∆, µ = 0, (36) and ∆ is real. In this case, the BdG Hamiltonian takes a compact form Hk = A(Σxky − Σykx) + (λ + MkV1 + ∆T1)Σz, (37) CONTENTS 34 Figure 20. Chern number of a QAHI/superconductor hybrid N (43) as function of exchange energy λ in units of M0. For a finite pairing energy ∆, a plateau develops at N = 1, corresponding to a chiral kx + iky state with a single Majorana edge mode (29). where Σx, Σy, Σz, V1, and T1 are the 8 × 8 matrices (38) T1 = τ1ν0σz, Σx = τ3ν3σx, Σy = τ3ν3σy, Σz = τ0ν0σz, V1 = τ0ν1σz, and ν0 and τ0 are the unit matrices in layer and Nambu subspaces. Now, the sum λ + MkV1 + ∆T1 in Eq. (37) is the Dirac mass matrix. Furthermore, V1 and T1 commute with each other and with any of Σi, so in the basis of the common eigenstates of V1 and T1 the mass matrix has a diagonal structure with the entries (39) ν, τ = ±1. λ + νMk + τ ∆, (40) ν and τ are the eigenvalues of V1 and T1, respectively. Therefore, the BdG model for the QAHI decouples into four Chern subsystems or, equivalently, four species of the 2D kx + iky - superconductor. Accordingly, the total Chern number is (cid:88) ν,τ =± N = 1 2 [sgn(λ + νM0 + τ ∆) − sgn(νM1)] = 1 [sgn(λ + M0 + ∆) + sgn(λ − M0 + ∆) 2 + sgn(λ + M0 − ∆) + sgn(λ − M0 − ∆)]. To distinguish the superconducting case, we use here the notation N instead of C (cf. Ref. [453]). The superconducting pairing ∆ allows for specific phase transitions that are absent in the normal case (see Fig. 20). For low-Tc superconductor structures, we can safely assume ∆ < M0. Then, for positive parameters, the possible values of the Chern number (42) are (41) (42) (43)  0 N = 1 M0 − ∆ < λ < M0 + ∆, 2, λ < M0 − ∆, λ > M0 + ∆. [453, 424]). Figure 21. Schematic of the chiral edge modes and Majorana backscattering in QAHI/superconductor (S) structure (after Refs. (a) For a large enough exchange field λ, a pair of incident Majorana edge modes (making up a single Dirac edge mode) match those in the superconducting region and pass almost perfectly through the device. (b) As the exchange field λ decreases, the state of the superconducting region switches to the chiral (N = 1) TS, such that one of the paired Majorana edge modes vanishes. Consequently, only one of the incident Majorana edge modes is transmitted to the superconducting region, whereas the other Majorana mode is almost perfectly reflected, resulting in the half-integer 0.5 e2/h longitudinal conductance [453, 424]. In other words, an increasing exchange field λ induces a series of phase transitions from an ordinary superconductor with N = 0 to the topological phases with N = 1 and N = 2. The latter has two chiral Majorana edge modes which correspond to a single Dirac mode, so the N = 2 phase matches the QAHI with C = 1. The truly new phase is that with the odd Chern number N = 1 (43), which is nothing else as the kx + iky TS with a single chiral Majorana edge mode such as discussed above [cf. Eqs. (23) and (28)]. By reversing the magnetization λ → −λ, one can also access the opposite-chirality states N = −1 and N = −2. [453], the N = 1 phase can be identified by a half-integer plateau 0.5 e2/h in the longitudinal conductance as a function of the exchange field. The proposal relies on the backscattering of Majorana edge modes [446] which is expected in a QAHI/superconductor device at the topological transition to the N = 1 phase. The basic setup consists of a magnetic thin film and a superconducting bar place across it, as depicted in Fig. 21. For a large enough exchange field (λ > M0 + ∆, see Fig. 21a), the normal regions are the As argued in Ref. (cid:68)(cid:61)0(cid:68)(cid:68)(cid:68)(cid:68)(cid:68)(cid:68)(cid:68)(cid:68)(cid:68)(cid:185)0012Λ12NCONTENTS 35 fields during magnetization reversals in an external out-of-plane magnetic field. Some of the experimental data on the sample characterization and the occurrence of the half-integer plateaus are shown in Fig. 22. Following the experiment [424], alternative the- oretical and experimental interpretations of the half- integer longitudinal conductance have appeared. Ref- erence [459] suggested a mechanism for the 0.5 e2/h conductance plateau without 1D chiral Majorana fermions. It was argued that such plateaus could be a feature of a good electric contact between quantum Hall and superconducting films, and could therefore indicate neither the existence nor absence of 1D chiral Majorana fermions. The experiment [434] reported high-probability Andreev reflection in QAHI/superconductor structures, attributing their findings to high contact transparency and interpreting in this context the origin of the 0.5 e2/h conductance plateau. The theory [460] argued that a nearly flat con- ductance plateau, similar to that in Ref. [424], could also arise from the percolation of quantum Hall edges well before the onset of the topological superconduc- tivity or at temperatures much above the TS gap. shown that quasi-1D QAHI Another line of the theoretical research has dealt with the issues of control and manipulation of chiral Majorana fermions for possible practical applications of realistic QAHI/superconductors devices. Reference [461] has structures could exhibit a broad topological regime supporting localized Majorana zero energy modes and proposed to implement networks of such quasi-1D QAHI systems for scalable topological quantum computation. Since the Majorana fermion is a charge-neutral particle, the direct effect of an electric field on them should fail. The recent study [462] has proposed a magnetic flux control of the transport of chiral Majorana fermions in topological superconducting devices with Josephson junctions. Reference [463] has found that the propagation of chiral Majorana fermions could lead to the same unitary transformation as that in the braiding of Majorana zero modes, suggesting a platform to perform quantum computation with chiral Majorana fermions. The theoretical work [464] has suggested an interferometer for chiral Majorana modes where the interference effect was caused and controlled by a Josephson junction of proximity- induced TSs. Another recent paper [465] elaborates on the deterministic creation and braiding of chiral edge vortices in hybrid structures. 10. Topological weak superconductivity and the fractional Josephson effect. Striking manifestations of topological superconduc- tivity are expected to occur in weak links between Figure 22. Half-integer longitudinal conductance as a signature of single chiral Majorana edge modes (From [424]. Reprinted with permission from the American Association In particular, C shows for the Advancement of Science.) the longitudinal conductance σ12 as a function of external perpendicular magnetic field measured at 20 mK. When superconductivity is induced on the top surface of the QHAI, σ12 shows additional half-integer plateaus ( 0.5e2/h) between the transitions of the C = ±1 QAHI and the normal insulator. (Lower plot in C) Derivative of σ12 with respect to the magnetic field. Topological transitions are indicated by dashed lines and arrows. For full details of the presented data, see Ref. [424]. C = 1 QAHI with a Dirac edge state propagating along the sample boundary, while the superconducting region supports two Majorana edge modes, forming the phase with N = 2. Since a Dirac fermion is composed of two Majorana ones, we can think of two incident Majorana edge modes which match those in the superconducting region, getting transmitted almost perfectly through the device. Consequently, the longitudinal conductance, σ12, between contacts 1 and 2 reaches the quantum e2/h. Upon lowering the exchange field to M0 < λ < M0 + ∆, the state of the superconducting region switches to the chiral TS with N = 1, such that one of the paired Majorana edge modes vanishes, while the normal regions are still the C = 1 QAHI (see Fig. 21b). In this regime, only one of the incident Majorana edge modes can be transmitted to the superconducting region, whereas the other Majorana mode is almost perfectly reflected. This is what Ref. [453] called the separation of the two Majorana modes at the superconducting boundary, resulting in a half-integer plateau 0.5 e2/h in the longitudinal conductance. Following this theoretical prediction, the experimental paper [424] reported the observation of the half-integer longitudinal conductance plateaus close to the coercive CONTENTS 36 (cid:26) 1 0 low-dimensional systems supporting Majorana modes. Some illustrative examples of such systems are 2D d-wave superconductors [348], 1D p-wave supercon- ductors [352, 466, 467], superconductor/semiconductor wires [468, 469, 470, 471], Shiba chains [472, 473], RKKY wires [474], just to name a few. The bound- aries of 1D TSs host a pair of Majorana zero modes (MZMs). These 0D cousins of the chiral Majorana edge mode appear at the midgap energy, i.e. at exactly zero energy relative to the Fermi level. An interesting implication of the MZMs is the degeneracy of the ground state. For a pair of MZMs, there are two possible ground states corresponding to the eigenvalues ±1 of the hermitian operator iγ1γ2, † where γi are self-adjoint fermionic operators γi = γ i that square to 1 (Majorana operators). If we combine them into a usual fermion c = (γ1 + iγ2)/2, the two ground states have different occupation numbers † c c = 1 2 (1 + iγ1γ2) = , (44) fermion parities. hence different In Josephson junctions (JJs) of two TSs brought into electric contact, a change of the Josephson phase difference by 2π effectively causes swapping the MZMs and a transition between the ground states [352]. This leads to the 4π - periodicity of the MZMs, as another phase advance of 2π is needed to recover the same ground state. Systems supporting the MZMs may prove useful in braiding-based topological quantum computation in which computing operations are performed by unitary transformations within a degenerate set of ground states [354, 475, 476, 477, 362]. However, the MZMs are not readily available in solids, and much effort has been put into engineering and detecting them in accessible materials and structures (see review articles [359, 360, 361, 362, 363, 364, 365]). A growing number of experiments has been testing the existence of the MZMs in various superconducting structures, using tunneling spectroscopy (see Refs. [478, 479, 480, 415, 481, 482, 483, 484, 485]) and Josephson effects (JEs) (see Refs. [486, 417, 422, 423, 487, 488, 489, 490, 433]). The JE diagnostics of topological weak links relies on the so-called fractional JE associated with the ground state degeneracy. First proposed for model p- wave superconductors [352, 466, 467] the fractional JE is also achievable in hybrid structures of conventional superconductors and normal SOC materials, which has caused the recent surge of interest in this and related phenomena. Here, we review this topic from the theoretical perspective, aiming to give some background on the fractional JE and summarize the findings of different models. 10.1. Majorana zero modes in a 1D chiral TS As an exemplary model, we can choose a 1D version of the BdG Hamiltonian of the kx + iky TS [see Eq. (20)]. Alternatively, we can think of the chiral TS in the QAHI/superconductor structures at the transition to the N = 1 state [see Eqs. (37) and (43)]. The reported observability of such transitions [424, 431] gives us extra reason for this choice. With the x-axis parallel to the system, we can write (cid:35) (cid:34) H = = τ3 ∆(cid:48)eiϕkx 2m − µ) k2 2m − µ x (cid:16) k2 ∆(cid:48)e−iϕkx −( k2 + τ1∆(cid:48) 2m − µ (cid:17) e x x −iτ3ϕkx, (45) (46) (cid:16) k2 (cid:17) where kx = −i∂x and ϕ is the phase of the order parameter. It is convenient to make a unitary transformation of the BdG wave function, Ψ(x) → eiτ3ϕ/2Ψ(x), bringing the Hamiltonian to the form (47) x + τ1∆(cid:48) −iτ3ϕ/2Heiτ3ϕ/2 = τ3 2m − µ H → e It maps to a 1D Dirac fermion model with a mass term k2 2m − µ. Furthermore, akin to Jackiw - Rebbi model [491] the chiral symmetry τ2H(x)τ2 = −H(x) ensures the existence of the MZMs as eigenstates of the chirality matrix τ2: (48) kx. x C C (x), τ2Ψ (x) = τ Ψ τ2Ψ(x) = τ Ψ(x), (49) where τ = ±1 are the eigenvalues. The eigenstates of τ2 are self-adjoint [cf. Eq. (27)], satisfying the equation H(x)Ψ(x) = 0 or, equivalently, x + 2µm − 2τ2∆(cid:48) (∂2 m∂x)Ψ(x) = 0. (50) (cid:112)(∆(cid:48)m)2 − 2µm. In view of Eq. (49), this is an ordinary differential equation with simple solutions in the half space (say, x ≥ 0) for the boundary condition Ψ(0) = 0. The substitution Ψ(x) ∝ e−κx yields two solutions for the decay constant κ± = −τ ∆(cid:48) (51) m ± For ∆(cid:48)m > 0, the normalizability condition κ± > 0 is met for τ = −1 in the parameter range 0 < 2µm < (∆(cid:48) This defines the topological regime with a single MZM at x = 0: m)2. (52) 1 −i −κ+x − e ΨM (x) = N where N is the normalizing factor. By the same token, the other eigenstate of τ2 would correspond to an MZM localized at the opposite end of the TS. (53) (cid:34) (cid:35)(cid:0)e −κ−x(cid:1) , CONTENTS 37 10.2. Fractional Josephson effect. Phenomenology (cid:34) (cid:34) (cid:34) (cid:35) The fractional JE is caused by the coupling of the MZMs across a weak link between two TSs. The phenomenology is rather independent of the details of the TSs, and can be illustrated by an effective junction Hamiltonian HJ = HR V † V HL , V = , (54) V0 0 0 −V ∗ 0 (cid:35) (cid:35) where HR and HL are the Hamiltonians of the right and left TSs, while operator V models the coupling between them. V is a diagonal Nambu matrix where V0 is the normal-state coupling (generally complex, which would break time-reversal symmetry). The unitary transformation (47) in each TS yields HJ → HR V eiτ3φ/2 V †e−iτ3φ/2 HL , φ = ϕR − ϕL, (55) where the coupling acquires the dependence on the phase difference φ between the TSs, whereas the transformed HR and HL are both phase-independent akin to the Hamiltonian in Eq. (48). The energy spectrum of the JJ is obtained from the BdG equations † −iτ3φ/2ΨL, e (56) EΨR = HRΨR + V EΨL = V eiτ3φ/2ΨR + HLΨL, (57) assuming the normalization condition (cid:104)ΨL,RΨL,R(cid:105) = 1. In the lowest order in V , the energy levels can be expressed through the MZMs by putting HRΨR = 0 and HLΨL = 0 and projecting the BdG equations on the bra states (cid:104)ΨR,L. Denoting the solution by E+, we have −iτ3φ/2ΨL(cid:105). E+ = (cid:104)ΨLV eiτ3φ/2ΨR(cid:105) = (cid:104)ΨRV Since (cid:104)ΨRV †e−iτ3φ/2ΨL(cid:105) = (cid:104)ΨLV eiτ3φ/2ΨR(cid:105) (58) ∗, the (58) just means that the e † second equality in Eq. solution is real. The particle-hole symmetry ensures the existence irrespective of other sym- C L and of another solution, E−, metries of the system. Replacing ΨL → Ψ ΨR → Ψ E− = (cid:104)Ψ where the matrix element can be evaluated as follows C R, we have C LV eiτ3φ/2Ψ C R(cid:105), (59) C LV eiτ3φ/2Ψ (cid:104)Ψ C R(cid:105) = (cid:104)ΨLC† V eiτ3φ/2CΨR(cid:105) = (cid:104)ΨLC† V Ceiτ3φ/2ΨR(cid:105) = − (cid:104)ΨLV eiτ3φ/2ΨR(cid:105). (60) (61) (62) We use the particle-hole symmetry C†V C = −V , due to which the second level comes with the opposite sign E− = −(cid:104)ΨLV eiτ3φ/2ΨR(cid:105). (63) Figure 23. MZMs [see Eq. (64)]. 4π - periodic ABSs from hybridization of two Physically, E± are the levels of the Andreev bound states (ABSs) formed by the two hybridized MZMs. The corresponding wave functions Ψ± are the linear combinations of the right and left MZMs. Remarkably, the topological ABSs are 4π - periodic in the Josephson phase difference φ. This is a qualitative distinction from usual JJs where the ABSs are 2π - periodic [492]. For a time-reversal-invariant coupling V = V0τ3, the terms ∝ sin(φ/2) vanish because of the orthogonality of the MZMs, so the phase dependence of the energy levels is reduced to E±(φ) = ±(cid:104)ΨLV ΨR(cid:105) cos(φ/2), see also Fig. 23. It is worth noting that microscopic calculations assuming a potential barrier in the JJ [348, 466, 467, 440] give a similar result for the ABS spectrum (64) E±(φ) = ±(∆(cid:48) kF )√D cos(φ/2), (65) where D is the barrier transparency, and kF is the Fermi wave number. Two aspects of the topological ABSs merit special attention. First, their 4π - periodicity harbors the topological degeneracy due to the underlying MZMs. A phase translation φ → φ + 2π brings the JJ into a state with the same energy E+(φ + 2π) = E−(φ), (66) (44). but with the opposite fermion parity, since the relative sign of ΨR and ΨL has changed [cf. Eq. (47)], which corresponds to switching the occupation number in Eq. Second, each ABS carries a 4π - periodic supercurrent. Since the phase and the particle number are conjugate dynamical variables [493], a phase dependent coupling energy gives rise to a current flow between the systems. In sufficiently short JJs, a major contribution to the current-phase relation (CPR) comes from the ABSs [492]. At equilibrium, the CPR J(φ) is given by the thermodynamic formula E(cid:43)E(cid:45)2Π4ΠΦCONTENTS 38 J(φ) = J+(φ) + J−(φ), J±(φ) = e  ∂E±(φ) ∂φ n[E±(φ)], (67) (68) where J±(φ) are the contributions of the two ABS levels occupied according to the Fermi distribution n(E). Using Eq. (64) and specially chosen occupations n[E+(φ)] = 0 and n[E−(φ)] = 1, one has J(φ) = J−(φ) = e(cid:104)ΨLV ΨR(cid:105) 2 sin(φ/2). (69) This example illustrates the fractional JE which is characterized by a subharmonic CPR with the frequency 1/2. References [466, 467] proposed that CPR (69) could be observed in voltage-biased JJs where the phase difference evolves with time as φ(t) = 2eU t/, producing the current J(t) = e(cid:104)ΨLV ΨR(cid:105) 2 sin(eU t/), (70) oscillating at half the usual AC Josephson frequency 2eU/ at bias voltage U . That is, in topological JJs the Josephson current is carried by single electrons, rather than by Cooper pairs, or, in other words, an MZM is, loosely speaking, half the fermion. 10.3. Recent theories of the fractional JE. Non-equilibrium dynamics More recent theories have revisited the fractional JE in the context of MZMs in hybrid structures combining conventional superconductors with topological insula- tors or semiconductor nanowires [440, 448, 494, 495, 496, 497, 498, 499, 500, 501, 502, 503, 504, 505, 506, 507, 455, 508, 509, 510, 457, 511, 512, 513, 514, 515, 516, 517, 518, 519, 520, 521, 522, 523, 524]. Theoreti- cally, the 4π - periodic CPR (69) has been proposed for ferromagnetic weak links in quantum spin-Hall insu- lator/superconductor structures under assumption of the local fermion-parity conservation [440]. An explicit calculation of the 4π - periodic CPR from a parity- constraint free energy has been carried out in Ref. [501]. A related fractional JE occurs when the MZMs are spatially separated by a superconducting barrier [494]. A number of fermion parity. studies have opted for non- equilibrium dynamics as a more accessible alternative to fixing the [448] has identified signatures of the fractional JE in the finite-frequency current noise in a quantum spin-Hall insulator/superconductor structure with a ferromagnetic barrier. Particular attention has been paid to JJs between finite-length topological wires [495, Reference (71) 496, 497, 498, 514, 516, 519] where the hybridization of the end MZMs opens a gap in the ABS spectrum, rendering it 2π - periodic [495]. The fractional JE is recovered by biasing the JJ and thereby inducing the Landau-Zener transitions [496, 497, 498, 516, 519]. Several theoretical works have looked at the Shapiro steps in the current-voltage characteristics of dynamically driven JJs [496, 499, 502, 504, 512, 513, 517]. When a conventional JJ is exposed to an AC field with frequency ω a DC voltage develops, showing a series of steps at UDC = nω/2e, where n is an integer [525]. For topological JJs, where the current is carried by single electrons, the size of the Shapiro steps is expected to be twice larger UDC = nω/e = (2n)ω/2e, which corresponds to the even steps of conventional JJs. That expectation has been tested in the calculations using the resistively and capacitively shunted junction (RCSJ) model [496, 512, 513] supplemented with an appropriate CPR. The even steps were reproduced along with some additional features, such as odd and fractional steps, depending on the details of the input CPR and the parameter Instead of the RCSJ model and adiabatic choice. analysis, Ref. [517] has used the non-equilibrium Green's functions technique revealing a crossover from conventional Shapiro steps at high frequencies to a pattern with the missing odd steps at low frequencies. An applied bias leads to a finite lifetime and dynamics of the occupation of the ABS due to its non- adiabatic coupling to the continuum spectrum. As argued in Refs. [502, 503], the 4π periodicity manifests itself by an even-odd effect in Shapiro steps only if the ABS lifetime is longer than the phase adjustment time determined by the environment. However, another indicator of the 4π periodicity, a peak in the current noise spectrum at half the Josephson frequency, was found to be more robust against the environment. Qualitatively, the predicted noise spectrum is seU/π (72) (ω ∓ eU/)2 + (seU/π)2 , S(ω) ∝ for ω ∓ eU/ (cid:28) eU/. the usual Josephson frequency ω = ±eU/ which manifest the fractional JE in the regime when the ABS occupation switches faster than the phase adjustment time (corresponding to a small parameter s (cid:28) 1 in the equation above). It has peaks at half 10.4. Equilibrium tests of the fractional JE Driving JJs out of equilibrium brings about also unwanted effects that may hinder access to the topological physics. One of them is Joule overheating. According to Ref. it may be responsible for [489], CONTENTS 39 Figure 24. Schematic of a topological JJ created by placing two superconducting films across the edge of a 2DTI. The spreading of the edge state into the 2DTI bulk (on length-scale κ−1) results in the dependence of the Josephson transport on the magnetic flux Φ enclosed in the effective junction area (indicated by the dashed contour), which shows the 2Φ0 periodicity; w is the width of the superconducting contact to the 2DTI. higher order odd Shapiro steps seen in the experiments (see, e.g., [486, 417, 487]), although all odd steps should be missing in the fractional AC JE. Besides, Landau- Zener tunneling between the 2π - periodic branches of non-topological ABSs can emulate the fractional JE (see, e.g., [499]). In order to rule out such a possibility and avoid heating, it would also be desirable to be able to test the fractional JE at equilibrium, ideally when the topological ABSs are decoupled from the continuum. The difficulty is that the equilibrium CPR (67) is 2π periodic, as the two contributions there simply swap upon a 2π phase advance. Although the fractional JE cannot be easily inferred from such equilibrium CPRs, they, nevertheless, diagnose unconventional superconductivity in topological materials which has become a subject of intense effort on its own [443, 526, 527, 500, 528, 529, 530, 531, 532, 533, 534, 535, 536, 537, 538, 539, 540, 541, 542, 543, 544, 545, 546, 425, 458, 547, 548, 549]. In order to trace the fractional JE at equilibrium, one may look at the effect of an external magnetic field. Reference [505] has found an anomalous Fraunhofer pattern due to hybridized Majorana channels [394] at the top and bottom surfaces of a TI film. Another specific interference effect has been proposed in Ref. [507] for two finite-length 1D TSs forming a loop thread by a magnetic flux. The 4π periodicity translates into the magnetic-flux dependence with the period 2Φ0, where Φ0 = h/2e is the magnetic flux quantum. However, parity-switching events were found to spoil the 2Φ0 -periodicity of the critical current, causing instead a behaviour similar to that in π - junctions. Very often, wire-like TSs are treated as strictly 1D systems with zero width. This approximation misses an orbital magnetic-field effect on the wire, thereby overlooking a possible mechanism for the 2Φ0 periodicity in topological JJs Let consider a weak link between us, for example, [523]. Figure 25. Critical current of topological edge JJs for different values of the contact width w (see also Fig. 24 and [523]). The topological 2Φ0 - spaced oscillations are clearly visible if w is not too large compared to the edge-state width κ−1. 24). the edge of a two superconducting channels at 2DTI/superconductor hybrid (see Fig. It is essential that in real space the edge states are quasi-2D, spreading exponentially into the 2DTI bulk. For typical band-structure parameters of the inverted HgTe quantum wells, the edge-state spreading can be estimated as κ−1 ∼ 10 nm. This finite length-scale makes a topological ABS nonlocal in the sense that it picks up a magnetic flux, Φ, enclosed in the effective area of the JJ. Qualitatively, the critical current is given by (cid:20) e 2 (cid:18) (cid:19)(cid:21) Φ Φ0 −κw,(73) π ∆ + ∆κ cos , ∆κ ∼ ∆e Jc(Φ) ≈ where ∆ is the proximity-induced s - wave gap, while ∆κ accounts for the exponential spreading of the edge state underneath the superconducting contact. A more detailed analysis shows [523] that the 2Φ0 - spaced oscillations of Jc(Φ) occur on top of a monotonic decrease, as depicted in Fig. 25. The topological ABS levels show similar oscillations due to the gauge invariance of the 4π - periodic JE. A different type of the magnetic-field dependence Jc(B) has been predicted for semiconductor topological JJs [514]. In that case, the Zeeman effect of the applied field leads to magnetic oscillations of the critical current indicating the splitting of the MZMs in finite-length wires. An external magnetic field can also modify the shape of the equilibrium CPR J(φ), exposing the hidden fractional JE despite the conventional 2π - periodicity in φ. A recent example is the chiral CPR proposed in Ref. [524] for 2DTI-based JJs. This is a CPR of the form (cid:12)(cid:12)(cid:12)(cid:12)sin (cid:12)(cid:12)(cid:12)(cid:12) , φ 2 J(φ) = e∆ 2 C C = ±1, (74) Jclargerwsmallerw(cid:45)8(cid:45)6(cid:45)4(cid:45)202468(cid:70)(cid:70)00.51CONTENTS 40 10.5. Beyond the 4π periodicity So war, we have discussed the fractional JE associated with a double ground state degeneracy leading to a 4π periodicity due to the underlying MZMs. To conclude this section, let us mention an interesting generalization of the 4π periodic JE which comes into play when, in addition to Cooper pairing, other electronic interaction are present. Such electronic interactions can cause further fermion fractionalization due to induced many-body level splitting in topological JJs. The theory [506] has proposed that electron- electron interactions lead to a fourfold ground state degeneracy and, consequently, to a 8π - periodic JE associated with the weak tunneling of charge e/2 quasiparticles. A series of theoretical papers [509, 510, 511, 518] has addressed further aspects of electron interactions and "fractional" MZMs in topological JJs. 11. Unconventional superconductivity 11.1. Mixed-parity superconducting order parameter. Phenomenology Intrinsic noncentrosymmetric superconductors (NCSs) as well as many proximity structures of conventional superconductors and topological materials lack a center of inversion symmetry. Such superconducting systems do not fit into traditional classification of superconducting states which invokes definite (even or odd) spatial parity of the Cooper-pair wave function. Two examples of odd-parity states, the kx + iky - and kx - TSs, were discussed in the preceding sections. Both intrinsic NCSs and the mentioned proximity structures (dubbed, for convenience, 'proximity NCSs' here) exhibit an antisymmetric SOC which mixes the even-parity (spin-singlet) and the odd-parity (spin- triplet) Cooper pairs, producing an unconventional, mixed-parity superconducting order parameter. Unambiguous verification of the mixed-parity superconducting order remains one of the outstanding challenges in the NCS research [551, 552, 553, 387, 554]. Among intriguing physical consequences of the parity mixing are magnetoelectric effects [555] manifested in the conversion of a charge current into spin magnetization and vice versa (see, e.g., [556, 557, 558, 559]), the nonuniform (helical) superconducting order [560, 561] as well as topological bulk and surface properties (see recent reviews in Refs. [387] and [554]). For intrinsic NCSs, candidate order parameters can be classified according to their behaviour under the symmetry elements (space group) of the crystal. This is discussed extensively in literature (see, e.g., [562, 563, 551, 552, 554]). We will take a different route and derive the mixed-parity order parameter from a microscopic model for a proximity NCS. It Figure 26. Comparison between the chiral and ballistic CPRs, Eqs. (74) and (75), respectively. The arrows indicate the discontinuities of the derivative J(cid:48)(φ) caused by the fermion parity switching. describing a unidirectional supercurrent with the chirality C at T = 0. Precisely speaking, C coincides with the Chern number of the occupied spin band of the 2DTI. Noteworthy is a non-analytic phase dependence of Eq. (74) which clearly harbors This the 4π - periodic CPR J(φ) ∝ sin(φ/2). non-analyticity reflects a discontinuous topological transition associated with the change of the ground- state fermion parity and is inherent to the fractional JE. In Fig. 26, we compare the chiral CPR (74) with the CPR of a 1D ballistic JJ at T = 0: J(φ) = e∆ 2 sin φ 2 sgn cos . (75) (cid:18) (cid:19) φ 2 The shape of this CPR is largely independent of the type of the superconductors provided that the JJ is fully transparent (cf. Refs. [550] and [467]). The V- shaped minima of the chiral CPR indicate the fermion parity switching at 2π, 4π, ..., whereas the ballistic CPR is continuous at these points (see Fig. 26). The above discussion of topological weak super- conductivity misses a number of factors that can be op- erational in realistic topological JJs. The recent work [521] has scrutinized the role of various realistic physi- cal effects, such as a finite wire length, gap suppression, non-topological Andreev bound states, or chemical po- tential variations, in Majorana nanowire systems. As argued in [521], the system may exhibit 2π or 4π JEs or a combination of both, without a clear indication of the topological physics or emphasizing only some aspects of it. Only in a rather idealized situation (a very long wire with no chemical potential fluctuations or gap suppression) one could establish the 4π (resp. 2π) oscillations in the Josephson effect as being reliable evidence for topological (resp. ordinary) superconduc- tivity. These issues need to be understood better for the JEs as diagnostics of topological or trivial super- conducting states. chiralCPRballisticCPRJ2Π4ΠΦCONTENTS 41 Figure 27. A pictorial representation of a mixed-parity proximity effect in a SOC 2D material (2DM) contacted by a conventional (s-wave singlet) superconductor (S). The in-plane spin-momentum-locking facilitates conversion of singlet Cooper pairs into a mixture of singlet and triplet states in the SOC 2DM. is assumed that superconductivity is induced in a 2D SOC system by an overlying s - wave singlet superconductor, as depicted in Fig. 27. We can, for example, think of the surface of a 3DTI which should exhibit a pronounced parity mixing owing to extraordinary large SOC in these materials. In such proximity structures, the SOC forces the electron spins in a tunneling singlet Cooper pair to follow the electron momentum in the plane of the normal system, which causes a spin flip, hence an admixture of spin-triplet odd-parity Cooper pairs. Since each spin in a singlet pair can be flipped, both up - and down - spin pairs are induced with no net spin magnetization. This phenomenology is behind many microscopic studies of the proximity effect in systems with broken spin rotation symmetry, such as Rashba systems and TI materials[564, 565, 566, 567, 568, 569, 526, 570, 571, 572, 573, 574, 575, 537, 576, 577, 578, 538, 579, 580, 581, 582, 583, 584, 546, 585, 586, 425, 587, 549]. 11.2. Theory of the mixed-parity proximity effect More insight can be gained from the weak-coupling model of the superconducting proximity effect used earlier for various low-dimensional systems without SOC (see, e.g., [588, 589, 590, 591, 592, 593, 594, 595]) and later for TI surface states (see, e.g., [565, 526, 573, 574]). In this model, the proximity of the superconductor is accounted for by a tunneling self- energy ΣT in the equation of motion for the Green's function of the normal system Gk: [E−H(0) k −ΣT ]Gk = I, H(0) k = , (76) (cid:34) (cid:35) Hk 0 0 −H∗ −k k where H(0) is the bare Hamiltonian of the normal system in the Nambu representation, and I is the (cid:20) corresponding unit matrix. The self-energy is a matrix in the Nambu space with the following structure (cid:21) . (77) ΣT = −iΓ(E) −∆∗(E)iσy −iΓ(E) ∆(E)iσy Figure 28. Energy spectrum of a proximity NCS from Eqs. (90) and Eqs. (91) Its off-diagonal entries yield the induced singlet pair potential, while the diagonal elements account for the shift of the spectrum due to the tunneling: Γ(E) = Γ0gS (E) = Γ0 , Γ0 = πT 2ρS , (78) ∆(E) = iΓ0fS (E) = iΓ0 . (79) E(cid:112)E2 − ∆2 ∆S(cid:112)E2 − ∆2 S S (cid:35) (cid:34) Here, gS (E) and fS (E) are the momentum-integrated quasiparticle and condensate Green functions of the overlying superconductor which has the gap energy ∆S . The energy Γ0 is determined by the single-particle tunneling rate depending on the normal-state density of the states in the superconducting metal, ρS , and the tunneling matrix element T . At low energies E (cid:28) ∆S , in the main approximation Γ = 0 and ∆ = Γ0, hence the effective BdG Hamiltonian Hk = H(0) k + ΣT = Hk ∆iσy −∆iσy −H∗ −k . (80) The normal system is described by a 2D Hamiltonian Hk = σ · γk − µ, with an antisymmetric SOC field γk. (81) The pair potential in the self-energy (77) should not be confused with the induced order parameter. The latter can be characterized by the matrix pair amplitude (the anomalous average), which is a 2 × 2 spin matrix with the elements (cid:68) c↑k(t)c↑−k(t) c↓k(t)c↑−k(t) (cid:69) c↑k(t)c↓−k(t) c↓k(t)c↓−k(t) = f0(t, k)iσy (82) + f (t, k) · σiσy.(83) E(2)kE(1)kkFΔCONTENTS 42 Figure 29. Schematic of the triplet pairing in a 2D proximity NCS. There are two species of opposite-spin pairs in orbital (kx ± iky) states. the brackets Here, (cid:104)...(cid:105) denote the ground-state expectation value. Also, we use the singlet-triplet basis, with the singlet pair amplitude f0(t, k) and the triplet vector (cid:21) (cid:20) f↓↓ − f↑↑ 2 f (t, k) = , f↑↑ + f↓↓ 2i , f↑↓+↓↑ , (84) combining the amplitudes f↑↑(t, k), f↓↓(t, k), and f↑↓+↓↑(t, k) of the triplet pair states with the total spin projections Sz = 1,−1, and 0. In proximity NCSs, the role of the f vector is similar to that of the Balian- Wertheimer d vector in intrinsic NCSs. All pairing amplitudes can be obtained from the Green function of Eq. (76), which is the Nambu matrix (cid:20) G(E, k) F (E, k) (cid:21) . F †(E, k) G(E, k) G(E, k) = Here, each entry is a 2×2 matrix in spin space: G(E, k) and G(E, k) are the quasiparticle Green functions, while F (E, k) is the anomalous (condensate) Green function given by (85) F (E, k) = [f0(E, k) + f (E, k) · σ]iσy, where f0(E, k) and f (E, k) are the orbital amplitudes for the singlet and triplet pairing at given energy. From Eqs. (76) and (80) one readily finds (see, e.g., [582]) (86) ∆ Π(E, k) 2µ∆ f0(E, k) = (E2 − µ2 − ∆2 − γ2 k), γk, Π(E, k) f (E, k) = − Π(E, k) = (cid:2)E2 − (µ − γk2 − ∆2(cid:3) (cid:2)E2 − (µ + γk)2 − ∆2(cid:3) . × Noteworthy is the information about the energy spectrum and the order parameter of a proximity NCS. The energy spectrum is given by the roots of Π(E, k) (89) and consists of two spin-split BCS-like branches (cid:112)(µ − γk)2 + ∆2, (cid:112)(µ + γk)2 + ∆2. E(1) k = ± E(2) k = ± Figure 30. (a) Current-voltage characteristics as a function of magnetic field and temperature for Al/Bi2Te3/Al JJs (from Ref. [425]). The curves are shifted in voltage by a value proportional to the magnetic field. The dark points yield the magnetic pattern of the JJ with a pronounced dip at B = 0. (b) Evolution of the magnetic pattern shown in (a) as a function of the temperature. The dip flattens out at a temperature close to 100 mK. These are plotted in Fig. 28. As for the order parameter, it has a mixed parity, with the even singlet f0(E, k) (87) and odd triplet f (E, k) (88) components. The triplet admixture is proportional to the induced singlet pair potential ∆ and the SOC vector γk. That is, the SOC converts some of the tunneling singlet pairs into triplets, as depicted in Fig. 27. For example, for the Rashba SOC with (92) γk = αso(k × z) = αso[ky,−kx, 0] the f vector lies in the basal plane, describing the equal-spin triplets with the orbital (kx ± iky) symmetries (see also Fig. 29): f↑↑,↓↓ ∝ αso(kx ∓ iky), where αso is the SOC constant. The Sz = 0 triplet is absent by time-reversal symmetry. A similar triplet admixture occurs in intrinsic NCSs of the tetragonal group [562, 386, 552]. We note that Eq. (88) is not limited to the linear - in - k SOC. For example, the theory [584] has discussed the role of the hexagonal warping of the Fermi surface, which is relevant for the surface states of tetradymite compounds. f↑↓+↓↑ = 0, (93) The role of disorder deserves separate comment. In dirty TIs, the p - wave component (88) was found to be suppressed relative to the s - wave pairing when the elastic mean-free path was much smaller than the superconducting coherence length [574]. The suppression is due to the generic nonlocality of the odd-parity Cooper pairs, which makes them sensitive to the electron mean-free path in a disordered system. In cleaner TIs, however, the p -wave component can be comparable to the s - wave one and should therefore be observable despite the presence of a modest amount of disorder (e.g., random impurity potential). Reference [585] has developed the quasiclassical theory for the proximity effect in impure Dirac materials. Non- equilibrium Eilenberger and Usadel equations were (87) (88) (89) (90) (91) CONTENTS 43 or, explicitly, f (E, k) = 2∆ Π(E, k) [−µγk + Eh + iγk × h]. (97) We have now two new contributions. The term linear in energy E corresponds to the odd-frequency triplet pairing found a while ago in proximity structures of ferromagnets and s-wave superconductors [596, 597, 598]. Such odd-frequency Cooper pairs have the s- wave spatial symmetry and are robust to disorder and other spatial inhomogeneities, thus offering both new interesting physics and application potential [599, 600]. The triplet component (97) was calculated in Ref. [526] for a TI surface state with the linear Dirac spectrum. A growing body of work has been dealing with different aspects of the odd-frequency pairing in hybrid structures involving 3DTIs [572, 573, 576, 584, 587], 2DTI [577, 586] and related Rashba materials [578, 581]. [586] has looked into the emergence of the odd-frequency s-wave pairing at the edge of a 2DTI without any magnetism. In particular, Ref. Also, [603, 604]). superconductors of various The odd-frequency proximity effects have been symmetry studied for classes [601, 602]. the connection to the Majorana modes has been pointed out (see, e.g., Refs. Various other findings have been summarized in the review articles [597, 358, 605]. While the literature on the odd-frequency superconductivity is abundant, the emergence and physical consequences of the imaginary term in Eqs. (96) and (97) have gone largely unnoticed. This type of pairing is an analogue of the paradigmatic nonunitary pairing in triplet superfluids [366] and superconductors [606] with a complex triplet order parameter. We elaborate on this point below. 11.4. Nonunitary triplet pairing and charge-spin conversion As mentioned above, the standard classification of the Cooper pairing which invokes the Balian-Wertheimer d vector is not applicable to proximity NCSs where no pairing interaction takes place. For that purpose, we employ the matrix condensate Green function F (E, k) (86) which has proved useful in diverse proximity structures [597, 607, 571, 608, 609] and driven superconductors [610]. the pairing nonunitary if the product F F † is not proportional to a unit spin matrix. Using Eq. (86), we find 0 f +f0f∗)·σ+(if×f∗)·σ, (98) F F where 1 stands for the unit spin matrix (the arguments E and k are suppressed for brevity). The second and third terms above indicate the nonunitary pairing due † = (f02+f·f∗)1+(f [366], we call Following Ref. ∗ Figure 31. Sketch of a JJ at the surface of Bi2Te3 to probe induced mixed-parity superconductivity (from Ref. [425]). (a) The S electrodes (Al) induce a mixed s + p - wave superconductivity at the surface of Bi2Te3 (only the (px + ipy) - component is shown). In close proximity to low transparency interfaces, the px + ipy symmetry changes to a py one. (b) Top view of the device illustrating π coupling. In the presence of scattering, a quasi-particle trajectory emerging from the negative py - lobe on one side of the JJ (blue arrow) couples to a trajectory associated with the positive py - lobe on the other side of the JJ (red arrow). In the case of a scattering-free transport, the quasi- particle trajectories probe the same phase in both electrodes. For more details, see Ref. [425]. derived to first order in quantities small compared to the Fermi energy for Dirac edge and surface states with spin-momentum locking. The experiment [425] has reported an observation of the induced unconventional superconductivity at the surface of Bi2Te3 in phase-sensitive measurements on nanoscale JJs. The magnetic field pattern of the junctions was found to have a dip at zero applied magnetic field (see Fig. 30), presumably, due to the simultaneous existence of the 0 and π couplings across the junction provided by a mixed s + p - wave order parameter. The π coupling was attributed to the combined effect of a sign-changing p - component of the order parameter and scattering in the JJ (see also Fig. 31). 11.3. Odd-frequency triplet superconductivity Breaking the time-reversal symmetry enriches the un- conventional proximity effect in topological materials. An informative generalization of the above model is achieved by adding an exchange (or Zeeman) spin field h: γ(cid:48) k = γk + h. The total spin field γ(cid:48) momentum: −k (cid:54)= γ(cid:48) γ(cid:48) k is no longer antisymmetric in (94) (95) k, which leads to the following generalization of the triplet f vector in Eq. (88): f (E, k) = ∆ Π(E, k) [−µ(γ(cid:48) k − γ(cid:48) −k) + E(γ(cid:48) + iγ(cid:48) k + γ(cid:48) k × γ(cid:48) −k], −k) (96) CONTENTS 44 Figure 32. Sketch of charge-spin conversion in a SOC 2DM proximitized by a current-biased conventional superconductor. An applied electric current generates an in- plane spin polarization (cid:104)S(cid:105) ∝ if × f∗ reflecting the triplet pairing with the total spin projection Sz = 0 on the SOC plane. This can be interpreted as tilting the pair spins ↑ and ↓ such that they acquire a common in-plane projection [see also Eqs. (101) and (102)]. to the lack of inversion and time-reversal, respectively. We are interested in the latter case where, by analogy with triplet superfluids, Cooper pairs have a net spin polarization: (cid:104)S(cid:105) ∝ if × f∗ . (99) For example, for a 2D NCS in a perpendicular spin field h, Eq. (97) yields the following result for the axial vector if × f∗ at the Fermi level: if (0, k) × f∗(0, k) = 8µ∆2γ2 k Π2(0, k) h. (100) As expected, the pair spin polarization is parallel to the spin field h, indicating an imbalance between the equal-spin triplets ↑↑ and ↓↓ (see also Fig. 29). In the above example, the f vector formalism allows us to extend the notion of the nonunitary pairing beyond its original context [366], viz. to treat proximiy-induced superconductivity. Furthermore, the nonunitary pairing does not generally require the spin field h. The pair spin polarization if × f∗ can be induced just by an electric current via charge-spin conversion [582, 611]. To illustrate this point let us consider the BdG Hamiltonian (cid:21) (cid:20) σ · γk+q − µ ∆iσy , −∆iσy −(σ · γ−k+q − µ)∗ Hk = where the wave-vector shift q accounts for the presence of a superconducting phase gradient associated with a dissipationless electric current. The current is applied to the overlying superconductor and is weak enough to disregard the depairing effects in ∆. The corresponding triplet f vector is given by [582] Figure 33. Vector plot of the pair spin polarization (102) for the Rashba SOC. Vector if × f∗ shows the average polarization (103). where we used γ±k+q = ±γk + γq for any linear SOC. The combined effect of the SOC and the supercurrent produces a Zeeman-like field γq, thereby generating both odd-frequency and nonunitary pairing akin to h. Generally, the direction of γq depends on the type of the structural or lattice asymmetry behind the SOC, so does the pair spin polarization 8µ∆2 Π2(0, k) if (0, k) × f∗(0, k) = − γk × (γk × γq). (102) For the Rashba SOC, the spin polarization is carried by the triplet state with the total spin projection Sz = 0 on the SOC plane. Loosely speaking, the supercurrent tilts the pair spins ↑ and ↓ such that they acquire a common in-plane projection, as sketched in Fig. 32. Importantly, the spin polarization does not vanish upon averaging over the directions of the wave vector k. Using Eq. (92), one finds if (0, k) × f∗(0, k) = 4µ∆2α3 sok2 Π2(0, k) (q × z), (103) the where the bar means the average value. This result just means that an unpolarized charge current is converted into spin magnetization of superconducting condensate, a form of the magnetoelectric effect pioneered in normal metals [612] and studied later in NCSs [556]. the magnetoelectric effect refers to the spin magnetization induced by a phase gradient of the order parameter, while in the inverse magnetoelectric effect the magnetic polarization causes charge and spin flows in a variety of situations [557, 613, 614, 615, 616, 617, 618, 619, 620, 535, 621, 622]. In both effects, the magnetoelectric coupling is characterized by the SOC constant αso. Here, f (E, k) = 2∆ Π(E, k) [−µγk + Eγq + iγk × γq], (101) Equation (103) resembles the current-induced thermodynamic magnetization of a Rashba NCS [556, qiffCONTENTS 45 557, 558, 559]. The averaged spin polarization retains the dependence on the structural or lattice asymmetry. For the Rashba SOC, the average polarization direction is perpendicular to an applied supercurrent in the SOC plane (see also Fig. 33). For NCSs of the cubic crystal group we expect a different result. In this case, the SOC vector is simply parallel to the momentum, γk = αsok = αso[kx, ky, kz], and Eq. (102) yields the following result if (0, k) × f∗(0, k) = 16µ∆2α3 sok2 3Π2(0, k) q. (104) As we see, in cubic NCSs the pair spin polarization is locked parallel to the applied current. The charge-spin conversion is an indicator of the unconventional, mixed-parity order parameter. Still, the direct magnetoelectric effect has not been verified experimentally yet despite a diverse range of other observed properties §. On the other hand, a growing body of theoretical predictions may help in planning a decisive experiment. Some specific predictions include spin Hall effects and nonequilibrium spin accumulation in superconducting structures [623, 624, 625, 542], electrically controllable spin filtering in TI surfaces states [581], equal-spin Andreev reflection due to the induced nonunitary pairing [582, 611], magnetoelectric 0− π transitions in quantum spin Hall insulators [544], the generation of a transverse spin supercurrent by a charge supercurrent [626], and the long-range effect of a Zeeman field on the electric current in an Andreev interferometer [627]. 12. Outlook We give a brief overview of future directions. The contrast between quantum anomalous Hall research, which focuses on dissipationless transport at very low temperatures, and spin-orbit torque devices, which are aiming for room-temperature operation, is noteworthy. As a result of the latter several gaps remain in our understanding of spin-orbit torques at very low temperatures, for example the role of quantum interference effects such as weak localisation and anti-localisation. The exact origin of the anti- damping torque continues to be hotly debated, with possibilities including the spin Hall effect, the Berry curvature anomalous Hall term, which may however be overwhelmed by disorder, and spin-orbit scattering mechanisms that have not been fully explained. A related question, to date unsettled, is whether it is possible for an anti-damping torque to exist without the spin-Hall effect. in simple analytical models, the § A review article [554] gives a detailed account of the ongoing theoretical and experimental studies of intrinsic NCSs. final result to this question is very sensitive to the starting Hamiltonian. Furthermore, whereas the bulk of attention has focused on ferromagnets, antiferromagnets are also generating excitement [628, 629] and will no doubt witness considerable growth. Traditional problems from magnetism, such as current-driven domain wall motion, have yet to take off in topological materials. Spin-momentum locking offers new functionalities for magneto-resistive devices such as spin valves, since electrons travelling in a specific direction have a fixed spin orientation determined by their momentum. Experimentally, there is a lot of space to investigate van der Waals heterostructures where experiments are just beginning [630]. A fundamental gap in the theoretical approach to spin-orbit torques as well as non-linear response is the method used to handle the spin current. It is well known that in spin-orbit coupled systems the proper definition of the spin current is not (1/2){si, vj}, where s and v represent the spin and velocity operators respectively, as that current is not conserved, but (1/2)(d/dt){si, rj}, with r the position operator. This is motivated by spin non-conservation in the presence of spin-orbit coupling (an enlightening strategy for circumventing these ambiguities is described in [631]). Whereas the position operator is a difficult quantity to handle, in particular in bases of Bloch states, in which most such calculations are attempted, a complete understanding of spin-orbit torques in topological materials will remain elusive until the magnitude of the proper spin current is determined. Important unanswered questions in anomalous Hall transport include the role of spin-charge cor- relations, which has only recently begun to receive attention [176]. Moreover, according to a number of experiments, the sign of anomalous Hall conduc- tivity can be the same [632, 633, 634] as or differ [168, 632, 635, 636, 637] from that of the intrinsic con- tribution [638, 639, 640, 641], depending on the mag- netic doping concentration. This puzzling observation is thus far unexplained. Likewise, the role of disor- der in the anomalous Hall contribution due to Fermi arcs in WSMs has not been elucidated. The anomalous Hall effect is being explored in Dirac systems with spin, pseudospin and valley degrees of freedom [642], whose behaviour is qualitatively different from TIs. With the possibility of achieving strong spin-orbit coupling in graphene, one can envisage further research on this topic. In skyrmions studies it is assumed the skyrmion texture remains unaffected by the electrons at the in- terface, an assumption that remains to be verified by further research. Finally, anomalous Hall transport af- fects Coulomb drag [643], which must be explored fur- ther. CONTENTS 46 The understanding of interband coherence effects on the non-linear optical response is in its infancy. This applies both to interband transitions induced by Berry curvature terms as well as the role of scattering, whether by disorder, phonons or magnons, and the examination of effects known to be important in transport such as skew scattering, side-jump, localisation and Kondo physics. To add to this, the vast majority of work on optical systems has focused on the case of an undoped conduction band, while potential Fermi surface effects in doped systems have not been explored. Another exciting avenue for future research is the exploration of chiral superconductivity, non-Abelian excitations and unconventional Cooper pairing in topological materials. These topics remain a subject of intense experimental and theoretical effort. The experimental quest for chiral superconductivity in both intrinsic and proximity-induced superconductors continues. Speaking of hybrid proximity structures, we have seen that the chiral superconductivity can be understood as the duality between a kx + iky - superconductor and a Chern insulator. Is there more new physics beyond this intricate duality? The answer to this question is not only of theoretical interest, but may also uncover hitherto unexplored routes towards topological quantum computation and superconducting spintronics. Acknowledgments DC would like to thank Branislav Nikoli´c for a series of instructive discussions and to acknowledge the stimulating input of Di Xiao, Shulei Zhang, Giovanni Vignale, Allan MacDonald, and Oleg Tretiakov. YL thanks N. Liu, X. F. Niu and Z. C. Wang for technical assistance, and X. Dai, Z. Fang, K. He, H. Z. Lu, A. D. Mirlin, S. Q. Shen, J. R. Shi, H. M. Weng, P. Xiong and X. C. Xie for valuable discussions. GT thanks Ulrich Eckern, Sebastian Bergeret, Alexander Golubov, Yukio Tanaka and Dieter Weiss for their valuable comments. This work was supported by the Australian Research Council Centre of Excellence in Future Low-Energy Electronics Technologies funded by the Australian Government, by the National Science Foundation of China (Project 61425015), National Basic Research Program No. of China (Project No. 2015CB921102), National Key Research and Development Program (Project No. 2016YFA0300600), and Strategic Priority Research Program of Chinese Academy of Sciences (Project No. XDB28000000) and by the German Research Foundation (DFG) through TRR 80. References [1] Ren Y, Qiao Z and Niu Q 2016 Rep. Prog. Phys. 79 066501 [2] Culcer D and Geresdi A 2019 arXiv:1907.02625 [3] Shen S Q 2017 Topological Insulators: Dirac Equation in Condensed Matter (Singapore: Springer) [4] Ando Y 2013 J. Phys. Soc. Jpn. 82 102001 [5] Armitage N, Mele E and Vishwanath A 2018 Rev. Mod. Phys. 90 015001 [6] Xiao D, Liu G B, Feng W, Xu X and Yao W 2012 Phys. Rev. Lett. 108(19) 196802 [7] Zhang G, Li X, Wu G, Wang J, Culcer D, Kaxiras E and Zhang Z 2014 Nanoscale 6 3259 [8] Li X, Zhang G, Wu G, Chen H, Culcer D and Zhang Z 2013 Chin. Phys. B 22 097306 [9] Okugawa R and Murakami S 2014 Phys. Rev. B 89 235315 [10] Collins J L, Tadich A, Wu W, Gomes L C, Rodrigues J N B, Liu C, Hellerstedt J, Ryu H, Tang S, Mo S K, Adam S, Yang S A, Fuhrer M S and Edmonds M T 2018 Nature 564 390 [11] Rostami H, Guinea F, Polini M and Roldan R 2018 npj 2D Materials and Applications 2 15 [12] Wang Y, Zhu D, Wu Y, Yang Y, Yu J, Ramaswamy R, Mishra R, Shi S, Elyasi M, Teo K L, Wu Y and Yang H 2017 Nature Comm. 8 1364 [13] Avila J, Pearanda F, Prada E, San-Jose P and Aguado R 2018 arXiv:1807.04677 [14] Herring C 1937 Phys. Rev. 52(4) 365 -- 373 URL https: //link.aps.org/doi/10.1103/PhysRev.52.365 [15] Berry M V and Wilkinson M 1984 Proceedings of the Royal Society of London A. 392 15 -- 43 [16] Haldane F D M 1988 Phys. Rev. Lett. 61 2015 [17] Kane C L and Mele E J 2005 Phys. Rev. Lett. 95 226801 [18] Kane C 2013 Topological Insulators: Chapter 1. Topo- logical Band Theory and the Z2 Invariant Contempo- rary Concepts of Condensed Matter Science (Elsevier Science) ISBN 9780128086827 URL https://books. google.com.au/books?id=PawZDAAAQBAJ [19] Bernevig B and Hughes T 2013 Topological Insulators and Topological Superconductors (Princeton University Press) ISBN 9780691151755 URL https://books. google.com.au/books?id=wOn7JHSSxrsC [20] Nechaev I A and Krasovskii E E 2016 Phys. Rev. B 94 201410 [21] Yan B and Felser C 2017 Annual Review of Condensed Matter Physics 8 337 -- 354 (Preprint https://doi. org/10.1146/annurev-conmatphys-031016-025458) URL annurev-conmatphys-031016-025458 https://doi.org/10.1146/ [22] Goswami P and Tewari S 2013 Phys. Rev. B 88(24) 245107 URL https://link.aps.org/doi/10.1103/PhysRevB. 88.245107 [23] Chen Y, Wu S and Burkov A A 2013 Phys. Rev. B 88(12) 125105 URL https://link.aps.org/doi/10. 1103/PhysRevB.88.125105 [24] Zyuzin A A and Burkov A A 2012 Phys. Rev. B 86(11) 115133 URL https://link.aps.org/doi/10. 1103/PhysRevB.86.115133 [25] Kormanyos A, Zolyomi V, Drummond N D, Rakyta P, Burkard G and Fal'ko V I 2013 Phys. Rev. B 88 045416 [26] Kormanyos A, G Burkard G, Gmitra M, Fabian J, Zolyomi V and Falko N D D V 2015 2D Materials 2 022001 [27] Gong Z, Liu G B, Yu H, Xiao D, Cui X, Xu X and Yao W 2013 Nature Commun. 4 2053 [28] Culcer D, Sekine A and MacDonald A H 2017 Phys. Rev. B 96 035106 [29] Sekine A, Culcer D and MacDonald A H 2017 Phys. Rev. B 96 235134 [30] Li Q, Rossi E and Das Sarma S 2012 Phys. Rev. B 86 235443 CONTENTS 47 [31] Xiao C, Xiong B and Xue F 2018 J. Phys.: Condens. Matter 30 415002 [32] He H T, Wang G, Zhang T, Sou I K, Wong G K L, Wang J N, Lu H Z, Shen S Q and Zhang F C 2011 Phys. Rev. Lett. 106(16) 166805 [33] Lu H Z, Shi J and Shen S Q 2011 Phys. Rev. Lett. 107(7) 076801 [34] Shan W Y, Lu H Z and Shen S Q 2012 Phys. Rev. B 86(12) 125303 [35] Lu H Z and Shen S Q 2014 Phys. Rev. Lett. 112(14) 146601 [36] Liu W E, Liu H and Culcer D 2014 Phys. Rev. B 89 195417 [37] Adroguer P, Liu W E, Culcer D and Hankiewicz E M 2015 Phys. Rev. B 92 241402 [38] Liu W E, Hankiewicz E M and Culcer D 2017 Phys. Rev. B 96 045307 [39] Liu W E, Hankiewicz E M and Culcer D 2017 Materials 10 807 [40] Culcer D 2011 Phys. Rev. B 84(23) 235411 [41] Nandy S and Sodemann I 2019 arXiv 1901:04467 [42] Kane C L and Mele E J 2005 Phys Rev Lett 95 146802 ISSN 0031-9007 (Print) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/16241681 [43] Kane C L and Mele E J 2005 Phys. Rev. Lett. 95(22) 226801 URL https://link.aps.org/doi/10. 1103/PhysRevLett.95.226801 [44] Bernevig B A, Hughes T L and Zhang S C 2006 Science 314 1757 -- 1761 ISSN 0036-8073 [45] Koenig M, Wiedmann S, Brune C, Roth A, Buhmann H, Molenkamp L W, Qi X L and Zhang S C 2007 Science 318 766 -- 70 ISSN 1095-9203 (Electronic) 0036- 8075 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/17885096 [46] Roth A, Brune C, Buhmann H, Molenkamp L W, Maciejko J, Qi X L and Zhang S C 2009 Science 325 294 -- 7 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/19608911 [47] Schopfer F and Poirier W 2007 Journal of Applied Physics 102 054903 URL https://aip.scitation.org/doi/ abs/10.1063/1.2776371 [48] Jiang H, Cheng S, Sun Q F and Xie X C 2009 Phys Rev Lett 103 036803 ISSN 0031-9007 (Print) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/19659306 [49] Vayrynen J I, Goldstein M and Glazman L I 2013 Phys Rev Lett 110 216402 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/23745899 [50] Vayrynen J I, Goldstein M, Gefen Y and Glazman L I 2014 Physical Review B 90 ISSN 1098-0121 URL <GotoISI>://WOS:000342127700004 [51] Ma E Y, Calvo M R, Wang J, Lian B, Muhlbauer M, Brune C, Cui Y T, Lai K, Kundhikanjana W, Yang Y, Baenninger M, Konig M, Ames C, Buhmann H, Leubner P, Molenkamp L W, Zhang S C, Goldhaber-Gordon D, Kelly M A and Shen Z X 2015 Nat Commun 6 7252 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/26006728 [52] Gusev G M, Kvon Z D, Shegai O A, Mikhailov N N, Dvoretsky S A and Portal J C 2011 Physical Review B 84 ISSN 1098-0121 URL <GotoISI>://WOS: 000294777400001 [53] Liu C, Hughes T L, Qi X L, Wang K and Zhang S C 2008 Phys Rev Lett 100 236601 ISSN 0031-9007 (Print) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/18643529 [54] Knez I, Du R R and Sullivan G 2011 Phys Rev Lett 107 136603 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/22026882 [55] Du L, Knez I, Sullivan G and Du R R 2015 Phys Rev Lett 114 096802 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/25793839 [56] Li C A, Zhang S B and Shen S Q 2018 Physical Review B 97 ISSN 2469-9950 URL <GotoISI>://WOS: 000423121900005 [57] Skolasinski R, Pikulin D I, Alicea J and Wimmer M 2018 Physical Review B 98 ISSN 2469-9950 URL <GotoISI>://WOS:000450139700005 [58] Hart S, Ren H, Wagner T, Leubner P, Muhlbauer M, Brune C, Buhmann H, Molenkamp L and Yacoby A 2014 Nature Physics 10 638 -- 643 ISSN 1745-2473 URL <GotoISI>://WOS:000341820700015 [59] Nichele F, Suominen H J, Kjaergaard M, Marcus C M, Sajadi E, Folk J A, Qu F M, Beukman A J A, de Vries F K, van Veen J, Nadj-Perge S, Kouwenhoven L P, Nguyen B M, Kiselev A A, Yi W, Sokolich M, Manfra M J, Spanton E M and Moler K A 2016 New Journal of Physics 18 ISSN 1367-2630 URL <GotoISI>://WOS: 000388580400001 [60] Akiho T, Irie H, Onomitsu K and Muraki K 2019 Physical Review B 99 ISSN 2469-9950 URL <GotoISI>://WOS: 000462900200001 [61] Shojaei B, McFadden A P, Pendharkar M, Lee J S, Flatte M E and Palmstrom C J 2018 Physical Review Materials 2 ISSN 2475-9953 URL <GotoISI>://WOS: 000434770100001 [62] Liu C C, Feng W and Yao Y 2011 Phys Rev Lett 107 076802 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/21902414 [63] Xu Y, Yan B, Zhang H J, Wang J, Xu G, Tang P, Duan W and Zhang S C 2013 Phys Rev Lett 111 136804 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/24116803 [64] Reis F, Li G, Dudy L, Bauernfeind M, Glass S, Hanke W, Thomale R, Schafer J and Claessen R 2017 Science 357 287 -- 290 ISSN 1095-9203 (Electronic) 0036- 8075 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/28663438 [65] Weng H, Dai X and Fang Z 2014 Physical Review X 4 ISSN 2160-3308 [66] Qian X, Liu J, Fu L and Li J 2014 Science 346 1344 -- 7 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/25504715 [67] Yang F, Miao L, Wang Z F, Yao M Y, Zhu F, Song Y R, Wang M X, Xu J P, Fedorov A V, Sun Z, Zhang G B, Liu C, Liu F, Qian D, Gao C L and Jia J F 2012 Phys Rev Lett 109 016801 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/23031123 [68] Ren Y, Qiao Z and Niu Q 2016 Rep Prog Phys 79 066501 ISSN 1361-6633 (Electronic) 0034-4885 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/27176924 [69] Zhu F F, Chen W J, Xu Y, Gao C L, Guan D D, Liu C H, Qian D, Zhang S C and Jia J F 2015 Nat Mater 14 1020 -- 5 ISSN 1476-1122 (Print) 1476- 1122 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/26237127 [70] Drozdov I K, Alexandradinata A, Jeon S, Nadj-Perge S, Ji H W, Cava R J, Bernevig B A and Yazdani A 2014 Nature Physics 10 664 -- 669 ISSN 1745-2473 URL <GotoISI>://WOS:000341820700020 [71] Takayama A, Sato T, Souma S, Oguchi T and Takahashi T 2015 Phys Rev Lett 114 066402 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/25723232 [72] Tang S J, Zhang C F, Wong D, Pedramrazi Z, Tsai H Z, Jia C J, Moritz B, Claassen M, Ryu H, Kahn S, Jiang CONTENTS 48 J, Yan H, Hashimoto M, Lu D H, Moore R G, Hwang C C, Hwang C, Hussain Z, Chen Y L, Ugeda M M, Liu Z, Xie X M, Devereaux T P, Crommie M F, Mo S K and Shen Z X 2017 Nature Physics 13 683 -- + ISSN 1745-2473 URL <GotoISI>://WOS:000404629900018 [73] Peng L, Yuan Y, Li G, Yang X, Xian J J, Yi C J, Shi Y G and Fu Y S 2017 Nat Commun 8 659 ISSN 2041- 1723 (Electronic) 2041-1723 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/28939864 [74] Fei Z Y, Palomaki T, Wu S F, Zhao W J, Cai X H, Sun B S, Nguyen P, Finney J, Xu X D and Cobden D H 2017 Nature Physics 13 677 -- + ISSN 1745-2473 URL <GotoISI>://WOS:000404629900017 [75] Wu R, Ma J Z, Nie S M, Zhao L X, Huang X, Yin J X, Fu B B, Richard P, Chen G F, Fang Z, Dai X, Weng H M, Qian T, Ding H and Pan S 2016 Physical Review X 6 ISSN 2160-3308 [76] Li X B, Huang W K, Lv Y Y, Zhang K W, Yang C L, Zhang B B, Chen Y B, Yao S H, Zhou J, Lu M H, Sheng L, Li S C, Jia J F, Xue Q K, Chen Y F and Xing D Y 2016 Phys Rev Lett 116 176803 ISSN 1079- 7114 (Electronic) 0031-9007 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/27176532 [77] Shi Y, Kahn J, Niu B, Fei Z, Sun B, Cai X, Francisco B A, Wu D, Shen Z X, Xu X, Cobden D H and Cui Y T 2019 Sci Adv 5 eaat8799 ISSN 2375-2548 (Electronic) 2375- 2548 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/30783621 [78] Wu S, Fatemi V, Gibson Q D, Watanabe K, Taniguchi T, Cava R J and Jarillo-Herrero P 2018 Science 359 76 -- 79 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/29302010 [79] Tang F, Ren Y, Wang P, Zhong R, Schneeloch J, Yang S A, Yang K, Lee P A, Gu G, Qiao Z and Zhang L 2019 Nature 569 537 -- 541 ISSN 1476-4687 (Electronic) 0028-0836 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/31068693 [80] Hasan M Z and Kane C L 2010 Reviews of Modern Physics 82 3045 -- 3067 ISSN 0034-6861 1539-0756 [81] Qi X L and Zhang S C 2011 Reviews of Modern Physics 83 1057 -- 1110 ISSN 0034-6861 1539-0756 [82] Ando Y 2013 Journal of the Physical Society of Japan 82 ISSN 0031-9015 1347-4073 [83] Hsieh D, Qian D, Wray L, Xia Y, Hor Y S, Cava R J and Hasan M Z 2008 Nature 452 970 -- 4 ISSN 1476- 4687 (Electronic) 0028-0836 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/18432240 [84] Zhang H, Liu C X, Qi X L, Dai X, Fang Z and Zhang S C 2009 Nature Physics 5 438 -- 442 ISSN 1745-2473 1745-2481 [85] Ren Z, Taskin A A, Sasaki S, Segawa K and Ando Y 2010 Physical Review B 82 ISSN 1098-0121 1550-235X [86] Jia S, Ji H, Climent-Pascual E, Fuccillo M K, Charles M E, Xiong J, Ong N P and Cava R J 2011 Physical Review B 84 ISSN 1098-0121 1550-235X [87] Yang W M, Lin C J, Liao J and Li Y Q 2013 Chinese Physics B 22 ISSN 1674-1056 [88] Zhang J, Chang C Z, Zhang Z, Wen J, Feng X, Li K, Liu M, He K, Wang L, Chen X, Xue Q K, Ma X and Wang Y 2011 Nat Commun 2 574 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/22146393 [89] Kong D, Chen Y, Cha J J, Zhang Q, Analytis J G, Lai K, Liu Z, Hong S S, Koski K J, Mo S K, Hussain Z, Fisher I R, Shen Z X and Cui Y 2011 Nat Nanotechnol 6 705 -- 9 ISSN 1748-3395 (Electronic) 1748-3387 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/21963714 [90] Ren Z, Taskin A A, Sasaki S, Segawa K and Ando Y 2011 Physical Review B 84 ISSN 1098-0121 1550-235X [91] Xia Y, Qian D, Hsieh D, Wray L, Pal A, Lin H, Bansil A, Grauer D, Hor Y S, Cava R J and Hasan M Z 2009 Nature Physics 5 398 -- 402 ISSN 1745-2473 1745-2481 [92] Chen Y L, Chu J H, Analytis J G, Liu Z K, Igarashi K, Kuo H H, Qi X L, Mo S K, Moore R G, Lu D H, Hashimoto M, Sasagawa T, Zhang S C, Fisher I R, Hussain Z and Shen Z X 2010 Science 329 659 -- 62 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/20689013 [93] Roushan P, Seo J, Parker C V, Hor Y S, Hsieh D, Qian D, Richardella A, Hasan M Z, Cava R J and Yazdani A 2009 Nature 460 1106 -- 9 ISSN 1476-4687 (Electronic) 0028-0836 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/19668187 [94] Zhang T, Cheng P, Chen X, Jia J F, Ma X, He K, Wang L, Zhang H, Dai X, Fang Z, Xie X and Xue Q K 2009 Phys Rev Lett 103 266803 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/20366330 [95] Alpichshev Z, Analytis J G, Chu J H, Fisher I R, Chen Y L, Shen Z X, Fang A and Kapitulnik A 2010 Phys Rev Lett 104 016401 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/20366373 [96] Cheng P, Song C, Zhang T, Zhang Y, Wang Y, Jia J F, Wang J, Wang Y, Zhu B F, Chen X, Ma X, He K, Wang L, Dai X, Fang Z, Xie X, Qi X L, Liu C X, Zhang S C and Xue Q K 2010 Phys Rev Lett 105 076801 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/20868065 [97] Dai J, West D, Wang X, Wang Y, Kwok D, Cheong S W, Zhang S B and Wu W 2016 Phys Rev Lett 117 106401 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/27636482 [98] Qu D X, Hor Y S, Xiong J, Cava R J and Ong N P 2010 Science 329 821 -- 4 ISSN 1095-9203 (Electronic) 0036- 8075 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/20671155 [99] Analytis J G, McDonald R D, Riggs S C, Chu J H, Boebinger G S and Fisher I R 2010 Nature Physics 6 960 -- 964 ISSN 1745-2473 1745-2481 [100] Xiong J, Petersen A C, Qu D, Hor Y S, Cava R J and Ong N P 2012 Physica E: Low-dimensional Systems and Nanostructures 44 917 -- 920 ISSN 13869477 [101] Peng H, Lai K, Kong D, Meister S, Chen Y, Qi X L, Zhang S C, Shen Z X and Cui Y 2010 Nat Mater 9 225 -- 9 ISSN 1476-1122 (Print) 1476-1122 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/20010826 [102] Altshuler B, Aronov A and Spivak B 1981 JETP Letters 33 94 [103] Yu S D and V S Y 1981 JETP Letters 34 272 -- 275 [104] Bachtold A, Strunk C, Salvetat J P, Bonard J M, Forr L, Nussbaumer T and Schnenberger C 1999 Nature 397 673 -- 675 ISSN 0028-0836 1476-4687 URL https: //doi.org/10.1038/17755 [105] Washburn S and Webb R A 1986 Advances in Physics 35 375 -- 422 ISSN 0001-8732 1460-6976 [106] Zhang Y and Vishwanath A 2010 Phys Rev Lett 105 206601 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/21231253 ISSN 1079-7114 (Electronic) [107] Bardarson J H, Brouwer P W and Moore J E 2010 Phys Rev Lett 105 156803 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/21230927 [108] Jauregui L A, Pettes M T, Rokhinson L P, Shi L and Chen Y P 2016 Nat Nanotechnol 11 345 -- 51 ISSN 1748- 3395 (Electronic) 1748-3387 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/26780658 [109] Hong S S, Zhang Y, Cha J J, Qi X L and Cui Y 2014 Nano Lett 14 2815 -- 21 ISSN 1530-6992 (Electronic) CONTENTS 49 1530-6984 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/24679125 [110] Cho S, Dellabetta B, Zhong R, Schneeloch J, Liu T, Gu G, Gilbert M J and Mason N 2015 Nat Commun 6 7634 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/26158768 [111] Hikami S, Larkin A I and Nagaoka Y 1980 Progress of Theoretical Physics 63 707 -- 710 ISSN 0033-068X 1347- 4081 [112] Bergmann G 1984 Physics Reports-Review Section of Physics Letters 107 1 -- 58 ISSN 0370-1573 URL <GotoISI>://WOS:A1984TA69300001 [113] Chakravarty S and Schmid A 1986 Physics Reports 140 193 -- 236 ISSN 03701573 [114] Chen J, Qin H J, Yang F, Liu J, Guan T, Qu F M, Zhang G H, Shi J R, Xie X C, Yang C L, Wu K H, Li Y Q and Lu L 2010 Phys Rev Lett 105 176602 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/21231064 [115] Kim Y S, Brahlek M, Bansal N, Edrey E, Kapilevich G A, Iida K, Tanimura M, Horibe Y, Cheong S W and Oh S 2011 Physical Review B 84 ISSN 1098-0121 1550-235X [116] King P D, Hatch R C, Bianchi M, Ovsyannikov R, Lupulescu C, Landolt G, Slomski B, Dil J H, Guan D, Mi J L, Rienks E D, Fink J, Lindblad A, Svensson S, Bao S, Balakrishnan G, Iversen B B, Osterwalder J, Eberhardt W, Baumberger F and Hofmann P 2011 Phys Rev Lett 107 096802 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/21929260 [117] Suzuura H and Ando T 2002 Phys Rev Lett 89 266603 ISSN 0031-9007 (Print) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/12484845 [118] Garate I and Glazman L 2012 Physical Review B 86 ISSN 1098-0121 1550-235X [119] Lin C J, He X Y, Liao J, Wang X X, Iv V S, Yang W M, Guan T, Zhang Q M, Gu L, Zhang G Y, Zeng C G, Dai X, Wu K H and Li Y Q 2013 Physical Review B 88 ISSN 1098-0121 1550-235X [120] Chen J, He X Y, Wu K H, Ji Z Q, Lu L, Shi J R, Smet J H and Li Y Q 2011 Physical Review B 83 ISSN 1098-0121 1550-235X [121] Zhang G, Qin H, Chen J, He X, Lu L, Li Y and Wu K 2011 21 2351 -- 2355 ISSN 1616-301X [122] Checkelsky J G, Hor Y S, Cava R J and Ong N P 2011 Phys Rev Lett 106 196801 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/21668185 [123] Steinberg H, Lalo J B, Fatemi V, Moodera J S and Jarillo- Herrero P 2011 Physical Review B 84 ISSN 1098-0121 1550-235X [124] Kim D, Syers P, Butch N P, Paglione J and Fuhrer M S 2013 Nat Commun 4 2040 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/23800708 [125] Brahlek M, Koirala N, Salehi M, Bansal N and Oh S 2014 Phys Rev Lett 113 026801 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/25062217 [126] Liao J, Ou Y, Liu H, He K, Ma X, Xue Q K and Li Y 2017 Nat Commun 8 16071 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/28695894 [127] Xu Y, Miotkowski I, Liu C, Tian J, Nam H, Alidoust N, Hu J, Shih C K, Hasan M Z and Chen Y P 2014 Nature Physics 10 956 -- 963 ISSN 1745-2473 1745-2481 [128] Yoshimi R, Tsukazaki A, Kozuka Y, Falson J, Takahashi K S, Checkelsky J G, Nagaosa N, Kawasaki M and Tokura Y 2015 Nat Commun 6 6627 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/25868494 [129] Xu Y, Miotkowski I and Chen Y P 2016 Nat Com- mun 7 11434 ISSN 2041-1723 (Electronic) 2041- 1723 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/27142344 [130] Koirala N, Brahlek M, Salehi M, Wu L, Dai J, Waugh J, Nummy T, Han M G, Moon J, Zhu Y, Dessau D, Wu W, Armitage N P and Oh S 2015 Nano Lett 15 8245 -- 9 ISSN 1530-6992 (Electronic) 1530-6984 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/26583739 [131] Moon J, Koirala N, Salehi M, Zhang W, Wu W and Oh S 2018 Nano Lett 18 820 -- 826 ISSN 1530-6992 (Electronic) 1530-6984 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/29313354 [132] Xie F, Zhang S, Liu Q, Xi C, Kang T T, Wang R, Wei B, Pan X C, Zhang M, Fei F, Wang X, Pi L, Yu G L, Wang B and Song F 2019 Physical Review B 99 ISSN 2469-9950 2469-9969 [133] Taskin A A, Legg H F, Yang F, Sasaki S, Kanai Y, Matsumoto K, Rosch A and Ando Y 2017 Nature Comm. 8 1340 [134] Chiba T, Takahashi S and Bauer G E W 2017 Phys. Rev. B 95 094428 [135] Scharf B, Matos-Abiague A, Han J E, Hankiewicz E M and Zutic I 2016 Phys. Rev. Lett. 117 166806 [136] Huang H, Gu J, Ji P, Wang Q, Hu X, Qin Y, Wang J and Zhang C 2018 Applied Physics Letters 113 ISSN 0003-6951 1077-3118 [137] Wu B, Pan X C, Wu W, Fei F, Chen B, Liu Q, Bu H, Cao L, Song F and Wang B 2018 Applied Physics Letters 113 ISSN 0003-6951 1077-3118 [138] Nandy S, Sharma G, Taraphder A and Tewari S 2017 Phys Rev Lett 119 176804 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/29219428 [139] Nandy S, Taraphder A and Tewari S 2018 Sci. Rep. 8 14983 [140] Huang X, Zhao L, Long Y, Wang P, Chen D, Yang Z, Liang H, Xue M, Weng H, Fang Z, Dai X and Chen G 2015 Physical Review X 5 ISSN 2160-3308 [141] Burkov A A 2017 Physical Review B 96 ISSN 2469-9950 URL <GotoISI>://WOS:000405184900001 [142] Dai X, Du Z Z and Lu H Z 2017 Phys Rev Lett 119 166601 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/29099204 [143] Taskin A A, Legg H F, Yang F, Sasaki S, Kanai Y, Matsumoto K, Rosch A and Ando Y 2017 Nat Commun 8 1340 ISSN 2041-1723 (Electronic) 2041- 1723 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/29109397 [144] Qi X L, Hughes T L and Zhang S C 2008 Physical Review B 78 ISSN 1098-0121 URL <GotoISI>://WOS: 000262607800110 [145] Qi X L, Li R, Zang J and Zhang S C 2009 Science 323 1184 -- 7 ISSN 1095-9203 (Electronic) 0036- 8075 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/19179491 [146] Wu L, Salehi M, Koirala N, Moon J, Oh S and Armitage N P 2016 Science 354 1124 -- 1127 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/27934759 [147] Dziom V, Shuvaev A, Pimenov A, Astakhov G V, Ames C, Bendias K, Bottcher J, Tkachov G, Hankiewicz E M, Brune C, Buhmann H and Molenkamp L W 2017 Nat Commun 8 15197 ISSN 2041-1723 (Electronic) 2041- 1723 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/28504268 [148] Okada K N, Takahashi Y, Mogi M, Yoshimi R, Tsukazaki A, Takahashi K S, Ogawa N, Kawasaki M and Tokura Y 2016 Nat Commun 7 12245 ISSN 2041-1723 CONTENTS 50 (Electronic) 2041-1723 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/27436710 [149] Thouless D J, Kohmoto M, Nightingale M P and den Nijs M 1982 Physical Review Letters 49 405 -- 408 ISSN 0031- 9007 [150] Klass U, Dietsche W, Vonklitzing K and Ploog K 1992 Surface Science 263 97 -- 99 ISSN 0039-6028 URL <GotoISI>://WOS:A1992HF18600018 [151] Ado I A, Dmitriev I A, Ostrovsky P M and Titov M 2015 EPL (Europhysics Letters) 111 ISSN 0295-5075 1286- 4854 [152] Lu H Z, Shi J and Shen S Q 2011 Phys Rev Lett 107 076801 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/21902413 [153] He K, Wang Y and Xue Q K 2018 Annual Review of Condensed Matter Physics 9 329 -- 344 ISSN 1947-5454 1947-5462 [154] Tokura Y, Yasuda K and Tsukazaki A 2019 Nature Reviews Physics 1 126 -- 143 ISSN 2522-5820 [155] Yu R, Zhang W, Zhang H J, Zhang S C, Dai X and Fang Z 2010 Science 329 61 -- 4 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/20522741 [156] Xu S Y, Neupane M, Liu C, Zhang D, Richardella A, Andrew Wray L, Alidoust N, Leandersson M, Balasubramanian T, Snchez-Barriga J, Rader O, Landolt G, Slomski B, Hugo Dil J, Osterwalder J, Chang T R, Jeng H T, Lin H, Bansil A, Samarth N and Zahid Hasan M 2012 Nature Physics 8 616 -- 622 ISSN 1745-2473 1745-2481 [157] Chang C Z, Zhang J, Liu M, Zhang Z, Feng X, Li K, Wang L L, Chen X, Dai X, Fang Z, Qi X L, Zhang S C, Wang Y, He K, Ma X C and Xue Q K 2013 Adv Mater 25 1065 -- 70 ISSN 1521-4095 (Electronic) 0935- 9648 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/23334936 [158] Chang C Z, Zhao W, Kim D Y, Zhang H, Assaf B A, Heiman D, Zhang S C, Liu C, Chan M H and Moodera J S 2015 Nat Mater 14 473 -- 7 ISSN 1476-1122 (Print) 1476-1122 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/25730394 [159] Lee I, Kim C K, Lee J, Billinge S J L, Zhong R, Schneeloch J A, Liu T, Valla T, Tranquada J M, Gu G and Davis J C S 2015 Proceedings of the National Academy of Sciences 112 1316 -- 1321 ISSN 0027-8424 (Preprint https://www.pnas.org/content/ 112/5/1316.full.pdf) URL https://www.pnas.org/ content/112/5/1316 [160] Chang C Z, Zhang J, Feng X, Shen J, Zhang Z, Guo M, Li K, Ou Y, Wei P, Wang L L, Ji Z Q, Feng Y, Ji S, Chen X, Jia J, Dai X, Fang Z, Zhang S C, He K, Wang Y, Lu L, Ma X C and Xue Q K 2013 Science 340 167 -- 70 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/23493424 [161] Fox E J, Rosen I T, Yang Y, Jones G R, Elmquist R E, Kou X, Pan L, Wang K L and Goldhaber-Gordon D 2018 Phys Rev B 98 ISSN 2469-9950 (Print) URL https://www.ncbi.nlm.nih.gov/pubmed/30984899 [162] Gtz M, Fijalkowski K M, Pesel E, Hartl M, Schreyeck S, Winnerlein M, Grauer S, Scherer H, Brunner K, Gould C, Ahlers F J and Molenkamp L W 2018 Applied Physics Letters 112 ISSN 0003-6951 1077-3118 [163] Mogi M, Yoshimi R, Tsukazaki A, Yasuda K, Kozuka Y, Takahashi K S, Kawasaki M and Tokura Y 2015 Applied Physics Letters 107 ISSN 0003-6951 1077-3118 [164] Ou Y, Liu C, Jiang G, Feng Y, Zhao D, Wu W, Wang X X, Li W, Song C, Wang L L, Wang W, Wu W, Wang Y, He K, Ma X C and Xue Q K 2018 Adv Mater 30 ISSN 1521-4095 (Electronic) 0935-9648 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/29125706 [165] Mogi M, Kawamura M, Yoshimi R, Tsukazaki A, Kozuka Y, Shirakawa N, Takahashi K S, Kawasaki M and Tokura Y 2017 Nat Mater 16 516 -- 521 ISSN 1476-1122 (Print) 1476-1122 (Linking) URL https://www.ncbi. nlm.nih.gov/pubmed/28191899 [166] Mogi M, Kawamura M, Tsukazaki A, Yoshimi R, Takahashi K S, Kawasaki M and Tokura Y 2017 Sci Adv 3 eaao1669 ISSN 2375-2548 (Electronic) 2375- 2548 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/28989967 [167] Xiao D, Jiang J, Shin J H, Wang W, Wang F, Zhao Y F, Liu C, Wu W, Chan M H W, Samarth N and Chang C Z 2018 Phys Rev Lett 120 056801 ISSN 1079- 7114 (Electronic) 0031-9007 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/29481164 [168] Checkelsky J G, Ye J, Onose Y, Iwasa Y and Tokura Y 2012 Nature Physics 8 729 EP -- URL https://doi. org/10.1038/nphys2388 [169] Zhang D, Richardella A, Rench D W, Xu S Y, Kandala A, Flanagan T C, Beidenkopf H, Yeats A L, Buckley B B, Klimov P V, Awschalom D D, Yazdani A, Schiffer P, Hasan M Z and Samarth N 2012 Physical Review B 86 ISSN 1098-0121 1550-235X [170] Zhang M, Lv L, Wei Z, Yang L, Yang X and Zhao Y 2014 International Journal of Modern Physics B 28 ISSN 0217-9792 1793-6578 [171] Sanchez-Barriga J, Varykhalov A, Springholz G, Steiner H, Kirchschlager R, Bauer G, Caha O, Schierle E, Weschke E, Unal A A, Valencia S, Dunst M, Braun J, Ebert H, Minar J, Golias E, Yashina L V, Ney A, Holy V and Rader O 2016 Nat Commun 7 10559 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/26892831 [172] Liu N, Teng J and Li Y 2018 Nat Commun 9 1282 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/29599425 [173] Lee J S, Richardella A, Rench D W, Fraleigh R D, Flanagan T C, Borchers J A, Tao J and Samarth N 2014 Physical Review B 89 ISSN 1098-0121 1550-235X [174] Zhang J, Chang C Z, Tang P, Zhang Z, Feng X, Li K, Wang L L, Chen X, Liu C, Duan W, He K, Xue Q K, Ma X and Wang Y 2013 Science 339 1582 -- 6 ISSN 1095-9203 (Electronic) 0036-8075 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/23539598 [175] Zhang Z, Feng X, Guo M, Li K, Zhang J, Ou Y, Feng Y, Wang L, Chen X, He K, Ma X, Xue Q and Wang Y 2014 Nat Commun 5 4915 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/25222696 [176] Keser A C, Raimondi R and Culcer D 2019 arXiv 1902:09605 [177] Fu L and Kane C L 2008 Phys Rev Lett 100 096407 ISSN 0031-9007 (Print) 0031-9007 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/18352737 [178] Fu L and Kane C L 2009 Phys Rev Lett 102 216403 ISSN 0031-9007 (Print) 0031-9007 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/19519119 [179] Akhmerov A R, Nilsson J and Beenakker C W 2009 Phys Rev Lett 102 216404 ISSN 0031-9007 (Print) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/19519120 [180] Wei P, Katmis F, Assaf B A, Steinberg H, Jarillo- Herrero P, Heiman D and Moodera J S 2013 Phys Rev Lett 110 186807 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/23683236 [181] Yang Q I, Dolev M, Zhang L, Zhao J, Fried A D, Schemm E, Liu M, Palevski A, Marshall A F, Risbud S H and Kapitulnik A 2013 Physical Review B 88 ISSN 1098- CONTENTS 51 0121 1550-235X [182] Kandala A, Richardella A, Rench D W, Zhang D M, Flanagan T C and Samarth N 2013 Applied Physics Letters 103 202409 (Preprint https://doi.org/10. 1063/1.4831987) URL https://doi.org/10.1063/1. 4831987 [183] Lang M, Montazeri M, Onbasli M C, Kou X, Fan Y, Upadhyaya P, Yao K, Liu F, Jiang Y, Jiang W, Wong K L, Yu G, Tang J, Nie T, He L, Schwartz R N, Wang Y, Ross C A and Wang K L 2014 Nano Lett 14 3459 -- 65 ISSN 1530-6992 (Electronic) 1530-6984 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/24844837 [184] Yang W, Yang S, Zhang Q, Xu Y, Shen S, Liao J, Teng J, Nan C, Gu L, Sun Y, Wu K and Li Y 2014 Applied Physics Letters 105 ISSN 0003-6951 1077-3118 [185] Alegria L D, Ji H, Yao N, Clarke J J, Cava R J and Petta J R 2014 Applied Physics Letters 105 ISSN 0003-6951 1077-3118 [186] Tang C, Chang C Z, Zhao G, Liu Y, Jiang Z, Liu C X, McCartney M R, Smith D J, Chen T, Moodera J S and Shi J 2017 Sci Adv 3 e1700307 ISSN 2375-2548 (Electronic) 2375-2548 (Linking) URL https://www. ncbi.nlm.nih.gov/pubmed/28691097 [187] Buchenau S, Sergelius P, Wiegand C, B ler S, Zierold R, Shin H S, Rbhausen M, Gooth J and Nielsch K 2017 2D Materials 4 ISSN 2053-1583 [188] Huang S Y, Chong C W, Tung Y, Chen T C, Wu K C, Lee M K, Huang J C, Li Z and Qiu H 2017 Sci Rep 7 2422 ISSN 2045-2322 (Electronic) 2045-2322 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/28546637 [189] Katmis F, Lauter V, Nogueira F S, Assaf B A, Jamer M E, Wei P, Satpati B, Freeland J W, Eremin I, Heiman D, Jarillo-Herrero P and Moodera J S 2016 Nature 533 513 -- 6 ISSN 1476-4687 (Electronic) 0028- 0836 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/27225124 [190] Li M, Song Q, Zhao W, Garlow J A, Liu T H, Wu L, Zhu Y, Moodera J S, Chan M H W, Chen G and Chang C Z 2017 Physical Review B 96 ISSN 2469-9950 2469-9969 [191] Lee J S, Richardella A, Fraleigh R D, Liu C x, Zhao W and Samarth N 2018 npj Quantum Materials 3 ISSN 2397-4648 [192] Mogi M, Nakajima T, Ukleev V, Tsukazaki A, Yoshimi R, Kawamura M, Takahashi K S, Hanashima T, Kakurai K, Arima T h, Kawasaki M and Tokura Y 2019 Physical Review Letters 123 ISSN 0031-9007 1079-7114 [193] Luo W D and Qi X L 2013 Physical Review B 87 ISSN 1098-0121 URL <GotoISI>://WOS:000315146500016 [194] Eremeev S V, Men'shov V N, Tugushev V V, Echenique P M and Chulkov E V 2013 Physical Review B 88 ISSN 1098-0121 URL <GotoISI>://WOS:000326389400005 [195] Hirahara T, Eremeev S V, Shirasawa T, Okuyama Y, Kubo T, Nakanishi R, Akiyama R, Takayama A, Hajiri T, Ideta S I, Matsunami M, Sumida K, Miyamoto K, Takagi Y, Tanaka K, Okuda T, Yokoyama T, Kimura S I, Hasegawa S and Chulkov E V 2017 Nano Lett 17 3493 -- 3500 ISSN 1530-6992 (Electronic) 1530- 6984 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/28545300 [196] Otrokov M M, Rusinov I P, Blanco-Rey M, Hoffmann M, Vyazovskaya A Y, Eremeev S V, Ernst A, Echenique P M, Arnau A and Chulkov E V 2019 Phys Rev Lett 122 107202 ISSN 1079-7114 (Electronic) 0031- 9007 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/30932645 [197] Li J, Li Y, Du S, Wang Z, Gu B L, Zhang S C, He K, Duan W and Xu Y 2019 Sci Adv 5 eaaw5685 ISSN 2375-2548 (Electronic) 2375-2548 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/31214654 [198] Deng Y, Yu Y, Zhu Shi M, Wang J, Chen X H and Zhang Y 2019 arXiv e-prints arXiv:1904.11468 (Preprint 1904. 11468) [199] Nikolic B K, Dolui K, Petrovic M D, Plechac P, Markussen T and Stokbro K 2018 First-Principles Quantum Transport Modeling of Spin-Transfer and Spin-Orbit Torques in Magnetic Multilayers (Cham: Springer) [200] Manchon A, Miron I, Jungwirth T, Sinova J, Zelezny J, Thiaville A, Garello K and Gambardella P arXiv 1801.09636 [201] Li C H, van 't Erve O M, Robinson J T, Liu Y, Li L and Jonker B T 2014 Nat Nanotechnol 9 218 -- 24 ISSN 1748-3395 (Electronic) 1748-3387 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/24561354 [202] Mellnik A, Lee J, Richardella A, Grab J, Mintun P, Fischer M, Vaezi A, Manchon A, Kim E, Samarth N and Ralph D 2014 Nature 511 449 [203] Wang Y, Deorani P, Banerjee K, Koirala N, Brahlek M, Oh S and Yang H 2015 Phys Rev Lett 114 257202 ISSN 1079-7114 (Electronic) 0031-9007 (Linking) URL https://www.ncbi.nlm.nih.gov/pubmed/26197141 [204] Jamali M, Lee J S, Jeong J S, Mahfouzi F, Lv Y, Zhao Z, Nikolic B K, Mkhoyan K A, Samarth N and Wang J P 2015 Nano Lett 15 7126 -- 32 ISSN 1530- 6992 (Electronic) 1530-6984 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/26367103 [205] Kondou K, Yoshimi R, Tsukazaki A, Fukuma Y, Matsuno J, Takahashi K S, Kawasaki M, Tokura Y and Otani Y 2016 Nature Physics 12 1027 -- 1031 ISSN 1745-2473 1745-2481 [206] Fanchiang Y T, Chen K H M, Tseng C C, Chen C C, Cheng C K, Yang S R, Wu C N, Lee S F, Hong M and Kwo J 2018 Nat Commun 9 223 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/29335558 [207] Fan Y, Upadhyaya P, Kou X, Lang M, Takei S, Wang Z, Tang J, He L, Chang L T, Montazeri M, Yu G, Jiang W, Nie T, Schwartz R N, Tserkovnyak Y and Wang K L 2014 Nat Mater 13 699 -- 704 ISSN 1476-1122 (Print) 1476-1122 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/24776536 [208] Yasuda K, Tsukazaki A, Yoshimi R, Kondou K, Takahashi K, Otani Y, Kawasaki M and Tokura Y 2017 Phys. Rev. Lett. 119 137204 [209] Dc M, Grassi R, Chen J Y, Jamali M, Reifsnyder Hickey D, Zhang D, Zhao Z, Li H, Quarterman P, Lv Y, Li M, Manchon A, Mkhoyan K A, Low T and Wang J P 2018 Nat Mater 17 800 -- 807 ISSN 1476-1122 (Print) 1476- 1122 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/30061733 [210] Han J, Richardella A, Siddiqui S A, Finley J, Samarth N and Liu L 2017 Phys Rev Lett 119 077702 ISSN 1079- 7114 (Electronic) 0031-9007 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/28949690 [211] Wang Y, Zhu D, Wu Y, Yang Y, Yu J, Ramaswamy R, Mishra R, Shi S, Elyasi M, Teo K L, Wu Y and Yang H 2017 Nat Commun 8 1364 ISSN 2041-1723 (Electronic) 2041-1723 (Linking) URL https://www.ncbi.nlm.nih. gov/pubmed/29118331 [212] Khang N H D, Ueda Y and Hai P N 2018 Nat Mater 17 808 -- 813 ISSN 1476-1122 (Print) 1476- 1122 (Linking) URL https://www.ncbi.nlm.nih.gov/ pubmed/30061731 [213] Liu L, Pai C F, Li Y, Tseng H W, Ralph D C and Buhrman R A 2012 Science 336 555 -- 8 ISSN 1095- 9203 (Electronic) 0036-8075 (Linking) URL https:// www.ncbi.nlm.nih.gov/pubmed/22556245 [214] Mounet N, Gibertini M, Schwaller P, Campi D, Merkys A, Marrazzo A, Sohier T, Castelli I E, Cepellotti A, Pizzi G and Marzari N 2018 NATURE NANOTECHNOLOGY 13 246+ ISSN 1748-3387 CONTENTS 52 [215] Aronov A and Lyanda-Geller Y 1989 JETP Lett. 50 431 [216] Aronov A, Lyanda-Geller Y and Pikus G 1991 JETP 73 537 [217] Ramaswamy R, Lee J M, Cai K M and Yang H 2018 Applied Physics Reviews 5 ISSN 1931-9401 URL <GotoISI>://WOS:000446117000007 [218] Ghosh S and Manchon A 2018 Phys. Rev. B 97 134402 [219] MacNeill D, Stiehl G, Guimaraes M, Buhrman R, Park J [252] Fang C, Gilbert M J and Bernevig B A 2014 Phys. Rev. Lett. 112 046801 [253] Zhong P, Ren Y, Han Y, Zhang L and Qiao Z 2017 Phys. Rev. B 96 241103 [254] Serlin M, Tschirhart C L, Polshyn H, Zhang Y, Zhu J, Watanabe K, Taniguchi T, Balents L and Young A F 2019 arXiv:1907.00261 [255] Sabzalipour A, Abouie J and Abedinpour S H 2015 J. and Ralph D 2017 Nature Phys. 13 300 Phys.: Condens. Matter 27 115301 [220] Li P, Wu W, Wen Y, Zhang C, Zhang J, Zhang S, Yu Z, Yang S A, Manchon A and xiang Zhang X 2018 Nature Commun. 9 3990 [256] Zarezad A N and Abouie J 2018 Phys. Rev. B 98 155413 [257] Ado I A, Dmitriev I A, Ostrovsky P M and Titov M 2015 Europhys. Lett. 111 37004 [221] Chang P H, Markussen T, Smidstrup S, Stokbro K and [258] Ado I A, Dmitriev I A, Ostrovsky P M and Titov M 2017 Nikolic B K 2015 Phys. Rev. B 92 201406 Phys. Rev. B 96 235148 [222] Akzyanov R S and Rakhmanov A L 2018 Phys. Rev. B 97 075421 [223] Marmolejo-Tejada J M, Dolui K, Lazic P, Chang P H, Smidstrup S, Stradi D, Stokbro K and Nikolic B K 2017 Nano Letters 17 5626 [224] Inoue J i, Kato T, Ishikawa Y, Itoh H, Bauer G E W and Molenkamp L W 2006 Phys. Rev. Lett. 97(4) 046604 [225] Inoue J, Bauer G E W and Molenkamp L W 2004 Phys. Rev. B 70 041303 [226] Ado I A, Tretiakov O A and Titov M 2017 Phys. Rev. B 95 094401 [227] Xiao C and Niu Q 2017 Phys. Rev. B 96 045428 [228] Fischer M H, Vaezi A, Manchon A and Kim E A 2016 Phys. Rev. B 93 125303 [229] Ndiaye P B, Akosa C A, Fischer M H, Vaezi A, Kim E A and Manchon A 2017 Phys. Rev. B 96 014408 [230] Jin K H and Jhi S H 2013 Phys. Rev. B 87 075442 [231] Rodriguez-Vega M, Schwiete G, Sinova J and Rossi E 2017 [259] Tse W K 2016 Phys. Rev. B 94 125430 [260] Lee W R and Tse W K 2017 Phys. Rev. B 95 201411 [261] Araki Y and Nomura K 2017 Phys. Rev. B 96 165303 [262] Andrikopoulos D and Soree B 2017 Sci. Rep. 7 17871 [263] Okuma N, Masir M R and MacDonald A H 2017 Phys. Rev. B 95 165418 [264] Imai Y and Kohno H 2018 J. Phys. Soc. Jpn. 87 073709 [265] Yasuda K, Tsukazaki A, Yoshimi R, Takahashi K, Kawasaki M and Tokura Y 2016 Phys. Rev. Lett. 117 127202 [266] Konig E J, Ostrovsky P M, Protopopov I V, Gornyi I V, Burmistrov I S and Mirlin A D 2014 Phys. Rev. B 90 165435 [267] Zhang S B, Lu H Z and Shen S Q 2015 Sci. Rep. 5 13277 [268] Li X, Zhang F and Niu Q 2013 Phys. Rev. Lett. 110 066803 [269] Cazalilla M A, Ochoa H and Guinea F 2014 Phys. Rev. Lett. 113 077201 [270] Li X, Zhang F and MacDonald A 2016 Phys. Rev. Lett. Phys. Rev. B 96 235419 116 026803 [232] Zhang Q, Chan K S and Li J 2018 Sci. Rep. 8 4343 [233] Song K, Soriano D, Cummings A W, Robles R, Ordejon P and Roche S 2018 Nano Lett. 18 2033 [271] Spataru C D and Lonard F 2014 Phys. Rev. B 90 085115 [272] Kotulla M and Zuelicke U 2017 New J. Phys. 19 073025 [273] Wang J, Meir Y and Gefen Y 2017 Phys. Rev. Lett. 118 [234] Huang C L, Chong Y D and Cazalilla M A 2017 Phys. 046801 Rev. Lett. 119 136804 [274] Tanaka Y, Furusaki A and Matveev K A 2011 Phys. Rev. [235] He K, Wang Y and Xue Q K 2018 Annu. Rev. Condens. Lett. 106 236402 Matter Phys. 9 329 [236] Yu R, Zhang W, Zhang H J, Zhang S C, Dai X and Fang Z 2010 Science 329 61 -- 64 (Preprint http://www. sciencemag.org/content/329/5987/61.full.pdf) [237] Rosenberg G and Franz M 2012 Phys. Rev. B 85 195119 [238] Wang J and Culcer D 2013 Phys. Rev. B 88(12) 125140 [239] Principi A, Vignale G and Rossi E 2015 Phys. Rev. B 92 041107 [275] Maciejko J 2009 Phys. Rev. Lett. 102 256803 [276] Altshuler B L, Aleiner I L and Yudson V I 2013 Phys. Rev. Lett. 111 086401 [277] Black-Schaffer A M and Balatsky A V 2012 Phys. Rev. B 85 121103 [278] Zhang Y Y, Shen M, An X T, Sun Q F, Xie X C, Chang K and Li S S 2014 Phys. Rev. B 90 054205 [279] Novelli P, Taddei F, Geim A K and Polini M 2019 Phys. [240] Kernreiter T, Governale M, Zuelicke U and Hankiewicz E Rev. Lett. 122 016601 2016 Phys. Rev. X 6 021010 [280] Vayrynen J I, Goldstein M and Glazman L I 2013 Phys. [241] Kim J, Wang H and Wu R 2018 Phys. Rev. B 97 125118 [242] Qi S, Qiao Z, Deng X, Cubuk E D, Chen H, Zhu W, Kaxiras E, Zhang S, Xu X and Zhang Z 2016 Phys. Rev. Lett. 117 056804 [243] Kim J, Jhi S H, MacDonald A H and Wu R 2017 Phys. Rev. Lett. 110 216402 [281] Vayrynen J I, Goldstein M, Gefe Y and Glazman L I 2014 Phys. Rev. B 90 115309 [282] Liu C, Edmonds M T and Fuhrer M S 2019 arXiv:1906.01214 Rev. B 96 140410 [283] Kosmider K, Gonzalez J W and Fernandez-Rossier J 2013 [244] Men'shov V N, Tugushev V V, Eremeev S V, Echenique Phys. Rev. B 88 245436 P M and Chulkov E V 2013 Phys. Rev. B 88 224401 [284] Kosmider K and Fernandez-Rossier J 2013 Phys. Rev. B [245] Eremeev S V, Men'shov V N, Tugushev V V, Echenique 87 075451 P M and Chulkov E V 2013 Phys. Rev. B 88 144430 [285] Gibertini M, Pellegrino F M D, Marzari N and Polini M [246] Nagaosa N, Sinova J, Onoda S, MacDonald A H and Ong 2014 Phys. Rev. B 90 245411 N P 2010 Rev. Mod. Phys. 82(2) 1539 [286] Xiao Y M, Xu W, Duppen B V and Peeters F M 2016 [247] Culcer D and Das Sarma S 2011 Phys. Rev. B 83(24) Phys. Rev. B 94 155432 245441 [287] Mak K F, McGill K L, Park J and McEuen P L 2014 [248] Lu H Z, Zhao A and Shen S Q 2013 Phys. Rev. Lett. 111 Science 344 1489 146802 [249] Dai X, Du Z Z and Lu H Z 2017 Phys. Rev. Lett. 119 [288] Lee J, Mak K F and Shan J 2016 Nature Nano. 11 421 [289] Beconcini M, Taddei F and Polini M 2016 Phys. Rev. B 166601 94 121408 [250] Tse W K and Das Sarma S 2006 Phys. Rev. B 74 245309 [251] Bi X, He P, Hankiewicz E M, Winkler R, Vignale G and [290] Feng W, Yao Y, Zhu W, Zhou J, Yao W and Xiao D 2012 Phys. Rev. B 86 165108 Culcer D 2013 Phys. Rev. B 88 035316 [291] Shan W Y, Lu H Z and Xiao D 2013 Phys. Rev. B 88 CONTENTS 125301 [292] Zhang L, Gong K, Chen J, Liu L, Zhu Y, Xiao D and Guo H 2014 Phys. Rev. B 90 195428 [293] Tahir M, Vasilopoulos P and Peeters F M 2016 Phys. Rev. B 93 035406 53 [326] Gao Y and Xiao D 2018 Phys. Rev. B 98 060402 [327] Sodemann I and Fu L 2015 Phys. Rev. Lett. 115 216806 [328] Facio J I, Efremov D, Koepernik K, You J S, Sodemann I and van den Brink J 2018 Phys. Rev. Lett. 121 246403 [329] Konig E J, Dzero M, Levchenko A and Pesin D A 2019 [294] Tse W K, Saxena A, Smith D and Sinitsyn N 2014 Phys. Phys. Rev. B 99 155404 Rev. Lett. 113 046602 [330] You J S, Fang S, Xu S Y, Kaxiras E and Low T 2018 Phys. [295] Sharma G, Economou S E and Barnes E 2017 Phys. Rev. Rev. B 98 121109 B 96 125201 [296] Zhou T, Zhang J, Jiang H, Zutic I and Yang Z 2018 npj Quant Mater 3 39 [297] Mak K F and Shan J 2016 Nature Photonics 10 216 [298] Mak K F, Xiao D and Shan J 2018 Nature Photonics 12 [331] Zhang Y, Sun Y and Yan B 2018 Phys. Rev. B 97 041101 [332] Matsyshyn O and Sodemann I 2019 arXiv:1907.02532 [333] Isobe H, Xu S Y and Fu L 2018 arXiv 1812:08162 [334] Du Z Z, Wang C M, Lu H Z and Xie X C 2018 arXiv 1812:08377 451 [335] Yang Z K, Wang J R and Liu G Z 2018 Phys. Rev. B 98 [299] Tahir M, Krstajic P M and Vasilopoulos P 2017 Europhys. 195123 Lett. 118 17001 [300] Kovalev V M, Tse W K, Fistul M V and Savenko I G 2018 New J. Phys 20 083007 [301] Lensky Y D, Song J C, Samutpraphoot P and Levitov L S 2015 Phys. Rev. Lett. 114 256601 [302] Kormanyos A, Zolyomi V, Fal'ko V I and Burkard G 2018 [336] Zhang S S L and Vignale G 2019 arXiv 1808:06339 [337] He P, Zhang S S L, Zhu D, Liu Y, Wang Y, Yu J, Vignale G and Yang H 2018 Nature Phys. 14 495 [338] Hipolito F and Pereira V M 2017 2D Materials 4 021027 [339] Sipe J E and Shkrebtii A I 2000 Phys. Rev. B 61 5337 [340] Kim K W, Morimoto T and Nagaosa N 2017 Phys. Rev. Phys. Rev. B 98 035408 B 95 035134 [303] Abergel D, Edge J and Balatsky A 2014 New J. Phys. 16 [341] Hamamoto K, Ezawa M, Kim K W, Morimoto T and 065012 Nagaosa N 2017 Phys. Rev. B 95 224430 [304] Sekine A and MacDonald A H 2018 Phys. Rev. B 97 [342] Zhang Y, van den Brink J, Felser C and Yan B 2018 2D 201301 Materials 5 044001 [305] Hosur P 2012 Phys. Rev. B 86 195102 [306] Wilson J H, Pixley J H, Huse D A, Refael G and Das [343] Rostami H and Polini M 2018 Phys. Rev. B 97 195151 [344] Salomaa M M and Volovik G E 1987 Rev. Mod. Phys. 59 Sarma S 2018 Phys. Rev. B 97 235108 533 [307] Inoue H, Gyenis A, Wang Z, Li J, Oh S W, Jiang S, Bernevig B A and Yazdani A 2016 Science 351 1184 [308] Gorbar E V, Miransky V A, Shovkovy I A and Sukhachov P O 2016 Phys. Rev. B 93 235127 [309] Resta G, Pi S T, Wan X and Savrasov S Y 2018 Phys. Rev. B 97 085142 [310] Zhao X M, Kong X, Guo C X, Wu Y J and Kou S P 2018 Europhys. Lett. 120 47004 [311] Potter A C, Kimchi I and Vishwanath A 2014 Nature Comm. 5 5161 [312] Moll P J, Nair N L, Helm T, Potter A C, Kimchi I, Vishwanath A and Analytis J G 2016 Nature 535 266 [313] Wang C M, Sun H P, Lu H Z and Xie X C 2017 Phys. [345] Salomaa M M and Volovik G E 1988 Phys. Rev. B 37 9298 [346] Wen X G and Zee A 1991 Phys. Rev. B 44 274 [347] Tanaka Y and Kashiwaya S 1995 Phys. Rev. Lett. 74 345 [348] Tanaka Y and Kashiwaya S 1996 Phys. Rev. B 53 9371 [349] Kashiwaya S and Tanaka Y 2000 Rep. Prog. Phys. 63 1641 [350] Read N and Green D 2000 Phys. Rev. B 61 10267 [351] Ivanov D A 2001 Phys. Rev. Lett. 86 268 [352] Kitaev A Y 2001 Phys. Usp. 44 131 [353] Kitaev A 2003 Ann. Phys. 303 2 [354] Nayak C, Simon S H, Stern A, Freedman M and Das Sarma S 2008 Rev. Mod. Phys. 80 1083 [355] Schnyder A P, Ryu S, Furusaki A and Ludwig A W W 2008 Phys. Rev. B 78 195125 Rev. Lett. 119 136806 [356] Morimoto T, Furusaki A and Mudry C 2015 Phys. Rev. B [314] Baireuther P, Hutasoit J A, Tworzydlo J and Beenakker 92 125104 C W J 2016 New Jour. Phys. 18 045009 [357] Chiu C K, Teo J C Y, Schnyder A P and Ryu S 2016 Rev. [315] k Shi L and Song J C W 2017 Phys. Rev. B 96 081410 [316] Pellegrino F M D, Katsnelson M I and Polini M 2015 Phys. Mod. Phys. 88 035005 [358] Tanaka Y, Sato M and Nagaosa N 2012 J. Phys. Soc. Jpn. Rev. B 92 201407 81 011013 [317] Hofmann J and Das Sarma S 2016 Phys. Rev. B 93 241402 [318] Song J C W and Rudner M S 2017 Phys. Rev. B 96 205443 [319] Andolina G M, Pellegrino F M D, Koppens F H L and Polini M 2018 Phys. Rev. B 97 125431 [359] Alicea J 2012 Rep. Prog. Phys. 75 076501 [360] Beenakker C 2013 Annu. Rev. Con. Mat. Phys. 4 113 [361] Elliott S R and Franz M 2015 Rev. Mod. Phys. 87 137 [362] Sarma S D, Freedman M and Nayak C 2015 NPJ Quantum [320] Misawa T, Yokoyama T and Murakami S 2011 Phys. Rev. Inf. 1 15001 B 84 165407 [321] Tokura Y and Nagaosa N 2018 Nature Communications 9 3740 [322] Ma Q, Xu S Y, Shen H, MacNeill D, Fatemi V, Chang T R, Valdivia A M M, Wu S, Du Z, Hsu C H, Fang S, Gibson Q D, Watanabe K, Taniguchi T, Cava R J, Kaxiras E, Lu H Z, Lin H, Fu L, Gedik N and Jarillo-Herrero P 2019 Nature 565 337 [323] Kang K, Li T, Sohn E, Shan J and Mak K F 2019 Nature Mater. 18 324 [324] Xu S Y, Ma Q, Shen H, Fatemi V, Wu S, Chang T R, Chang G, Valdivia A M M, Chan C K, Gibson Q D, Zhou J, Liu Z, Watanabe K, Taniguchi T, Lin H, Cava R J, Fu L, Gedik N and Jarillo-Herrero P 2018 Nature Physics 14 900 [325] Gao Y, Yang S A and Niu Q 2014 Phys. Rev. Lett. 112 166601 [363] Sato M and Ando Y 2017 Rep. Prog. Phys. 80 076501 [364] Aguado R 2017 La Rivista del Nuovo Cimento 40 523 [365] Haim A and Oreg Y 2018 Time-reversal-invariant topological superconductivity (Preprint arXiv:1809. 06863) [366] Leggett A J 1975 Rev. Mod. Phys. 47 331 [367] Vollhardt D and Wolfle P 1990 The Superfluid Phases of Helium 3 (New York: Taylor and Francis) [368] Kohmoto M 1985 Ann. Phys. 160 343 [369] Liu C X, Zhang S C and Qi X L 2016 Annu. Rev. Condens. Matter Phys. 7 301 [370] Mackenzie A P and Maeno Y 2003 Rev. Mod. Phys. 75 657 [371] Rice T M and Sigrist M 1995 J. Phys.: Condens. Matter 7 L643 [372] Luke G M et al. 1998 Nature 394 558 [373] Kashiwaya S, Kashiwaya H, Kambara H, Furuta T, CONTENTS 54 Yaguchi H, Tanaka Y and Maeno Y 2011 Phys. Rev. Lett. 107 077003 [374] Scaffidi T, Romers J C and Simon S H 2014 Phys. Rev. B 89 220510 [375] Yada K, Golubov A A, Tanaka Y and Kashiwaya S 2014 J. Phys. Soc. Jpn. 83 074706 [376] Zhang L F, Becerra V F, Covaci L and Milosevi´c M V 2016 Phys. Rev. B 94 024520 [407] Sochnikov I, Bestwick A J, Williams J R, Lippman T M, Fisher I R, Goldhaber-Gordon D, Kirtley J R and Moler K A 2013 Nano Lett. 13 3086 [408] Hart S, Ren H, Wagner T, Leubner P, Muhlbauer M, Brune C, Buhmann H, Molenkamp L W and Yacoby A 2014 Nat. Phys. 10 638 [409] Snelder M et al. 2014 Supercond. Sci. Technol. 27 104001 [410] Molenaar C G, Leusink D P, Wang X L and Brinkman A [377] Wang H, Luo J, Lou W, Ortmann J E, Mao Z Q, Liu Y 2014 Supercond. Sci. Technol. 27 104003 and Wei J 2017 New J. Phys. 19 053001 [378] Wang H, Ma L and Wang J 2018 Sci. Bull. 63 1141 [379] Ueno Y, Yamakage A, Tanaka Y and Sato M 2013 Phys. Rev. Lett. 111 087002 [380] Hor Y S, Williams A J, Checkelsky J G, Roushan P, Seo J, Xu Q, Zandbergen H W, Yazdani A, Ong N P and Cava R J 2010 Phys. Rev. Lett. 104 057001 [381] Fu L and Berg E 2010 Phys. Rev. Lett. 105 097001 [382] Sasaki S, Kriener M, Segawa K, Yada K, Tanaka Y, Sato M and Ando Y 2011 Phys. Rev. Lett. 107 217001 [383] Sasaki S, Ren Z, Taskin A A, Segawa K, Fu L and Ando Y 2012 Phys. Rev. Lett. 109 217004 [384] Novak M, Sasaki S, Kriener M, Segawa K and Ando Y 2013 Phys. Rev. B 88 140502 [385] Sato M and Fujimoto S 2009 Phys. Rev. B 79 094504 [386] Tanaka Y, Yokoyama T, Balatsky A V and Nagaosa N 2009 Phys. Rev. B 79 060505 [387] Schnyder A P and Brydon P M R 2015 J. Phys.: Condens. Matter 27 243201 [388] Yada K, Sato M, Tanaka Y and Yokoyama T 2011 Phys. Rev. B 83 064505 [389] Loder F, Kampf A P and Kopp T 2015 Sci. Rep. 5 15302 [390] Mazziotti M V, Scopigno N, Grilli M and Caprara S 2018 Condens. Matt. 3 37 [391] Lei C, Chen H and MacDonald A H 2018 Phys. Rev. Lett. 121 227701 [392] Andreev A F 1964 Sov. Phys. JETP. 19 1228 [393] Blonder G E, Tinkham M and Klapwijk T M 1982 Phys. Rev. B 25 4515 [394] Fu L and Kane C L 2008 Phys. Rev. Lett. 100 096407 [395] Zhang D, Wang J, DaSilva A M, Lee J S, Gutierrez H R, Chan M H W, Jain J and Samarth N 2011 Phys. Rev. B 84 165120 [396] Koren G, Kirzhner T, Lahoud E, Chashka K B and Kanigel A 2011 Phys. Rev. B 84 224521 [397] Sacepe B, Oostinga J B, Li J, Ubaldini A, Couto N J G, Giannini E and Morpurgo A F 2011 Nat. Commun. 2 575 [398] Veldhorst M, Snelder M, Hoek M, Gang T, Guduru V K, Wang X L, Zeitler U, van der Wiel W G, Golubov A A, Hilgenkamp H and Brinkman A 2012 Nat. Mater. 11 417 [399] Qu F, Yang F, Shen J, Ding Y, Chen J, Ji Z, Liu G, Fan J, Jing X, Yang C and Lu L 2012 Sci. Rep. 2 339 [400] Wang M X et al. 2012 Science 336 52 [401] Williams J R, Bestwick A J, Gallagher P, Hong S S, Cui Y, Bleich A S, Analytis J G, Fisher I R and Goldhaber- Gordon D 2012 Phys. Rev. Lett. 109 056803 [402] Yang F, Qu F, Shen J, Ding Y, Chen J, Ji Z, Liu G, Fan J, Yang C, Fu L and Lu L 2012 Phys. Rev. B 86 134504 [403] Maier L, Oostinga J B, Knott D, Brune C, Virtanen P, Tkachov G, Hankiewicz E M, Gould C, Buhmann H and Molenkamp L W 2012 Phys. Rev. Lett. 109 186806 [404] Cho S, Dellabetta B, Yang A, Schneeloch J, Xu Z, Valla T, Gu G, Gilbert M J and Mason N 2013 Nat. Commun. 4 1689 [405] Oostinga J B, Maier L, Schuffelgen P, Knott D, Ames C, Brune C, Tkachov G, Buhmann H and Molenkamp L W 2013 Phys. Rev. X 3 021007 [406] Koren G, Kirzhner T, Kalcheim Y and Millo O 2013 EPL [411] Galletti L 2014 Phys. Rev. B 89 134512 [412] Kurter C, Finck A D K, Ghaemi P, Hor Y S and Van Harlingen D J 2014 Phys. Rev. B 90 014501 [413] Finck A D K, Kurter C, Hor Y S and Van Harlingen D J 2014 Phys. Rev. X 4 041022 [414] Sochnikov I et al. 2015 Phys. Rev. Lett. 114 066801 [415] Xu J P et al. 2015 Phys. Rev. Lett. 114 017001 [416] Stehno M P, Orlyanchik V, Nugroho C D, Ghaemi P, Brahlek M, Koirala N, Oh S and Van Harlingen D J 2016 Phys. Rev. B 93 035307 [417] Wiedenmann J et al. 2016 Nat. Commun. 7 10303 [418] Tikhonov E S, Shovkun D V, Snelder M, Stehno M P, Huang Y, Golden M S, Golubov A A, Brinkman A and Khrapai V S 2016 Phys. Rev. Lett. 117 147001 [419] Pang Y et al. 2016 Chin. Phys. B 25 117402 [420] Dayton I M, Sedlmayr N, Ramirez V, Chasapis T C, Loloee R, Kanatzidis M G, Levchenko A and Tessmer S H 2016 Phys. Rev. B 93 220506 [421] Hart S, Ren H, Kosowsky M, Ben-Shach G, Leubner P, Brune C, Buhmann H, Molenkamp L W, Halperin B I and Yacoby A 2016 Nat. Phys. 13 87 [422] Bocquillon E, Deacon R S, Wiedenmann J, Leubner P, Klapwijk T M, Brune C, Ishibashi K, Buhmann H and Molenkamp L W 2016 Nat. Nanotechnol. 12 137 [423] Deacon R S et al. 2017 Phys. Rev. X 7 021011 [424] He Q L et al. 2017 Science 357 294 [425] Charpentier S et al. 2017 Nat. Commun. 8 2019 [426] Jauregui L A, Kayyalha M, Kazakov A, Miotkowski I, Rokhinson L P and Chen Y P 2018 Appl. Phys. Lett. 112 093105 [427] Ghatak S, Breunig O, Yang F, Wang Z, Taskin A A and Ando Y 2018 Nano Lett. 18 5124 [428] Klett R et al. 2018 Nano Letters 18 1264 [429] Snyder R A, Trimble C J, Rong C C, Folkes P A, Taylor P J and Williams J R 2018 Phys. Rev. Lett. 121 097701 [430] Lyu Z et al. 2018 Phys. Rev. B 98 155403 [431] Shen J et al. 2018 Spectroscopic evidence of chiral ma- jorana modes in a quantum anomalous hall insula- tor/superconductor heterostructure (Preprint arXiv: 1809.04752) [432] Kurter C et al. 2019 Nano Lett. 19 38 [433] Kayyalha M, Kargarian M, Kazakov A, Miotkowski I, Galitski V M, Yakovenko V M, Rokhinson L P and Chen Y P 2019 Phys. Rev. Lett. 122 047003 [434] Kayyalha M et al. 2019 Non-majorana origin of the half- quantized conductance plateau in quantum anomalous hall insulator and superconductor hybrid structures (Preprint arXiv:1904.06463) [435] Taskin A A, Ren Z, Sasaki S, Segawa K and Ando Y 2011 Phys. Rev. Lett. 107 016801 [436] Chang C Z et al. 2013 Science 340 167 [437] M´enard G C, Guissart S, Brun C, Leriche R T, Trif M, Debontridder F, Demaille D, Roditchev D, Simon P and Cren T 2017 Nat. Commun. 8 2040 [438] Fu L and Kane C L 2009 Phys. Rev. Lett. 102 216403 [439] Akhmerov A R, Nilsson J and Beenakker C W J 2009 Phys. Rev. Lett. 102 216404 [440] Fu L and Kane C L 2009 Phys. Rev. B 79 161408 [441] Law K T, Lee P A and Ng T K 2009 Phys. Rev. Lett. 103 237001 103 67010 [442] Tanaka Y, Yokoyama T and Nagaosa N 2009 Phys. Rev. CONTENTS 55 Lett. 103 107002 [443] Linder J, Tanaka Y, Yokoyama T, Sudbø A and Nagaosa N 2010 Phys. Rev. B 81 184525 [444] Qi X L, Hughes T L and Zhang S C 2010 Phys. Rev. B 82 184516 [445] Ioselevich P A and Feigel'man M V 2011 Phys. Rev. Lett. 106 077003 [446] Chung S B, Qi X L, Maciejko J and Zhang S C 2011 Phys. Rev. B 83 100512 [447] Ii A, Yada K, Sato M and Tanaka Y 2011 Phys. Rev. B 83 224524 Shtrikman H 2012 Nat. Phys. 8 887 [480] Nadj-Perge S, Drozdov I K, Li J, Chen H, Jeon S, Seo J, MacDonald A H, Bernevig B A and Yazdani A 2014 Science 346 602 [481] Deng M T, Vaitiekenas S, Hansen E B, Danon J, Leijnse M, Flensberg K, Nygard J, Krogstrup P and Marcus C M 2016 Science 354 1557 [482] Albrecht S M, Higginbotham A P, Madsen M, Kuemmeth F, Jespersen T S, Nygard J, Krogstrup P and Marcus C M 2016 Nature 531 206 [483] Jeon S, Xie Y, Li J, Wang Z, Bernevig B A and Yazdani [448] Badiane D M, Houzet M and Meyer J S 2011 Phys. Rev. A 2017 Science 358 772 Lett. 107 177002 [449] Golub A, Kuzmenko I and Avishai Y 2011 Phys. Rev. Lett. 107 176802 [484] Zhang H et al. 2018 Nature 556 74 [485] Ren H et al. 2019 Nature 569 93 [486] Rokhinson L P, Liu X and Furdyna J K 2012 Nature [450] Ii A, Yamakage A, Yada K, Sato M and Tanaka Y 2012 Physics 8 795 Phys. Rev. B 86 174512 [451] Weithofer L and Recher P 2013 New J. Phys. 15 085008 [452] He J J, Wu J, Choy T P, Liu X J, Tanaka Y and Law K T 2014 Nat. Commun. 5 3232 [487] Li C, de Boer J C, de Ronde B, Ramankutty S V, van Heumen E, Huang Y, de Visser A, Golubov A A, Golden M S and Brinkman A 2018 Nature Materials 17 875 [488] Wang A Q, Li C Z, Li C, Liao Z M, Brinkman A and Yu [453] Wang J, Zhou Q, Lian B and Zhang S C 2015 Phys. Rev. D P 2018 Phys. Rev. Lett. 121 237701 B 92 064520 [454] Akzyanov R S, Rakhmanov A L, Rozhkov A V and Nori F 2015 Phys. Rev. B 92 075432 [455] Shapiro D S, Shnirman A and Mirlin A D 2016 Phys. Rev. B 93 155411 [456] Lian B, Wang J and Zhang S C 2016 Phys. Rev. B 93 161401 [489] Le Calvez K, Veyrat L, Gay F, Plaindoux P, Winkelmann C B, Courtois H and Sac´ep´e B 2019 Commun. Phys. 2 4 [490] Laroche D et al. 2019 Nat. Commun. 10 245 [491] Jackiw R and Rebbi C 1976 Phys. Rev. D 13 3398 [492] Golubov A A, Kupriyanov M Y and Il'ichev E 2004 Rev. Mod. Phys. 76 411 [457] Shapiro D S, Feldman D E, Mirlin A D and Shnirman A 2017 Phys. Rev. B 95 195425 [493] Anderson P W 1966 Rev. Mod. Phys. 38 298 [494] Jiang L, Pekker D, Alicea J, Refael G, Oreg Y and von [458] Virtanen P, Bergeret F S, Strambini E, Giazotto F and Oppen F 2011 Phys. Rev. Lett. 107 236401 Braggio A 2018 Phys. Rev. B 98 020501 [459] Ji W and Wen X G 2018 Phys. Rev. Lett. 120 107002 [460] Huang Y, Setiawan F and Sau J D 2018 Phys. Rev. B 97 [495] Pikulin D I and Nazarov Y V 2012 JETP Lett. 94 693 [496] Dom´ınguez F, Hassler F and Platero G 2012 Phys. Rev. B 86 140503 100501 [497] San-Jose P, Prada E and Aguado R 2012 Phys. Rev. Lett. [461] Chen C Z, Xie Y M, Liu J, Lee P A and Law K T 2018 108 257001 Phys. Rev. B 97 104504 [462] Zhou Y F, Hou Z, Lv P, Xie X and Sun Q F 2018 Sci. China Phys. Mech. Astron. 61 127811 [463] Lian B, Sun X Q, Vaezi A, Qi X L and Zhang S C 2018 Proc. Nat. Acad. Sci. USA 115 10938 [464] Li C A, Li J and Shen S Q 2019 Phys. Rev. B 99 100504 [465] Beenakker C W J, Baireuther P, Herasymenko Y, Adagideli I, Wang L and Akhmerov A R 2019 Phys. Rev. Lett. 122 146803 [498] Pikulin D I and Nazarov Y V 2012 Phys. Rev. B 86 140504 [499] Sau J D, Berg E and Halperin B I 2012 On the possibility of the fractional ac josephson effect in non-topological conventional superconductor-normal- superconductor junctions (Preprint arXiv:1206.4596) [500] Tkachov G and Hankiewicz E M 2013 Phys. Status Solidi (b) 250 215 [501] Beenakker C W J, Pikulin D I, Hyart T, Schomerus H and Dahlhaus J P 2013 Phys. Rev. Lett. 110 017003 [466] Kwon H J, Sengupta K and Yakovenko V M 2004 Eur. [502] Houzet M, Meyer J S, Badiane D M and Glazman L I 2013 Phys. J. B 37 349 Phys. Rev. Lett. 111 046401 [467] Kwon H J, Yakovenko V M and Sengupta K 2004 Low [503] Badiane D M, Glazman L I, Houzet M and Meyer J S 2013 Temp. Phys. 30 613 C. R. Phys. 14 840 [468] Sau J D, Lutchyn R M, Tewari S and Das Sarma S 2010 Phys. Rev. Lett. 104 040502 [469] Alicea J 2010 Phys. Rev. B 81 125318 [470] Lutchyn R M, Sau J D and Das Sarma S 2010 Phys. Rev. [504] Virtanen P and Recher P 2013 Phys. Rev. B 88 144507 [505] Potter A C and Fu L 2013 Phys. Rev. B 88 121109 [506] Zhang F and Kane C L 2014 Phys. Rev. Lett. 113 036401 [507] Lee S P, Michaeli K, Alicea J and Yacoby A 2014 Phys. Lett. 105 077001 Rev. Lett. 113 197001 [471] Oreg Y, Refael G and von Oppen F 2010 Phys. Rev. Lett. [508] Dmytruk O, Trif M and Simon P 2016 Phys. Rev. B 94 105 177002 115423 [472] Nadj-Perge S, Drozdov I K, Bernevig B A and Yazdani A [509] Peng Y, Vinkler-Aviv Y, Brouwer P W, Glazman L I and 2013 Phys. Rev. B 88 020407 [473] Pientka F, Glazman L I and von Oppen F 2013 Phys. Rev. B 88 155420 von Oppen F 2016 Phys. Rev. Lett. 117 267001 [510] Hui H Y and Sau J D 2017 Phys. Rev. B 95 014505 [511] Vinkler-Aviv Y, Brouwer P W and von Oppen F 2017 [474] Klinovaja J, Stano P, Yazdani A and Loss D 2013 Phys. Phys. Rev. B 96 195421 Rev. Lett. 111 186805 [475] Flensberg K 2011 Phys. Rev. Lett. 106 090503 [476] Bonderson P and Lutchyn R M 2011 Phys. Rev. Lett. 106 130505 [477] Hyart T, van Heck B, Fulga I C, Burrello M, Akhmerov A R and Beenakker C W J 2013 Phys. Rev. B 88 035121 [478] Mourik V, Zuo K, Frolov S M, Plissard S R, Bakkers E P A M and Kouwenhoven L P 2012 Science 336 1003 [479] Das A, Ronen Y, Most Y, Oreg Y, Heiblum M and [512] Dom´ınguez F et al. 2017 Phys. Rev. B 95 195430 [513] Pic´o-Cort´es J, Dom´ınguez F and Platero G 2017 Phys. Rev. B 96 125438 [514] Cayao J, San-Jose P, Black-Schaffer A M, Aguado R and Prada E 2017 Phys. Rev. B 96 205425 [515] Trif M, Dmytruk O, Bouchiat H, Aguado R and Simon P 2018 Phys. Rev. B 97 041415 [516] Sun D and Liu J 2018 Phys. Rev. B 97 035311 [517] Li Y H, Song J, Liu J, Jiang H, Sun Q F and Xie X C CONTENTS 56 2018 Phys. Rev. B 97 045423 Rev. B 92 125443 [518] Sticlet D, Sau J D and Akhmerov A 2018 Phys. Rev. B 98 125124 [519] Feng J J, Huang Z, Wang Z and Niu Q 2018 Phys. Rev. B [559] Tokatly I V 2017 Phys. Rev. B 96 060502 [560] Mineev V P and Samokhin K V 1994 JETP 78 401 [561] Mineev V P and Samokhin K V 2008 Phys. Rev. B 78 98 134515 144503 [520] Mohanta N, Kampf A P and Kopp T 2018 Phys. Rev. B 97 214507 [521] Chiu C K and Das Sarma S 2019 Phys. Rev. B 99 035312 [522] Rodr´ıguez-Mota R, Vishveshwara S and Pereg-Barnea T 2019 Phys. Rev. B 99 024517 [523] Tkachov G 2019 J. Phys.: Condens. Matter 31 175301 [524] Tkachov G 2019 Phys. Rev. B 100 035403 [525] Shapiro S 1963 Phys. Rev. Lett. 11 80 [526] Yokoyama T 2012 Phys. Rev. B 86 075410 [527] Olund C T and Zhao E 2012 Phys. Rev. B 86 214515 [528] Snelder M, Veldhorst M, Golubov A A and Brinkman A [562] Sigrist M 2009 AIP AIP Conf. Proc. 1162 55 [563] Sergienko I A and Curnoe S H 2004 Phys. Rev. B 70 214510 [564] Edelstein V M 2003 Phys. Rev. B 67 020505 [565] Stanescu T D, Sau J D, Lutchyn R M and Das Sarma S 2010 Phys. Rev. B 81 241310 [566] Potter A C and Lee P A 2011 Phys. Rev. B 83 184520 [567] Lababidi M and Zhao E 2011 Phys. Rev. B 83 184511 [568] Ito Y, Yamaji Y and Imada M 2011 Journal of the Physical Society of Japan 80 063704 [569] Khaymovich I M, Chtchelkatchev N M and Vinokur V M 2013 Phys. Rev. B 87 104507 2011 Phys. Rev. B 84 075142 [529] Tkachov G and Hankiewicz E M 2013 Phys. Rev. B 88 075401 [570] Virtanen P and Recher P 2012 Phys. Rev. B 85 035310 [571] Annunziata G, Manske D and Linder J 2012 Phys. Rev. B [530] Zhang K H, Zhu Z G, Wang Z C, Zheng Q R and Su G 86 174514 2014 Phys. Lett. A 378 3131 [572] Black-Schaffer A M and Balatsky A V 2012 Phys. Rev. B [531] Bai C and Yang Y 2014 Nanoscale Res. Lett. 9 515 [532] Khezerlou M and Goudarzi H 2015 Int. J. Mod. Phys. B 86 144506 [573] Black-Schaffer A M and Balatsky A V 2013 Phys. Rev. B 29 1550018 87 220506 [533] Khezerlou M and Goudarzi H 2015 Phys. C 508 6 [534] Tkachov G, Burset P, Trauzettel B and Hankiewicz E M 2015 Phys. Rev. B 92 045408 [574] Tkachov G 2013 Phys. Rev. B 87 245422 [575] Zyuzin A A and Loss D 2014 Phys. Rev. B 90 125443 [576] Burset P, Lu B, Tkachov G, Tanaka Y, Hankiewicz E M [535] Dolcini F, Houzet M and Meyer J S 2015 Phys. Rev. B 92 and Trauzettel B 2015 Phys. Rev. B 92 205424 035428 [577] Cr´epin F, Burset P and Trauzettel B 2015 Phys. Rev. B [536] Lu B, Burset P, Yada K and Tanaka Y 2015 Supercond. 92 100507 Sci. Technol. 28 105001 [537] Snelder M, Golubov A A, Asano Y and Brinkman A 2015 J. Phys.: Condens. Matter 27 315701 [538] Tkachov G 2015 Topological Insulators: The Physics of Spin Helicity in Quantum Transport (New York: Pan Stanford Publishing) [539] Zyuzin A, Alidoust M, Klinovaja J and Loss D 2015 Phys. Rev. B 92 174515 [578] Reeg C R and Maslov D L 2015 Phys. Rev. B 92 134512 [579] Yu T and Wu M W 2016 Phys. Rev. B 93 195308 [580] Yu T and Wu M W 2017 Phys. Rev. B 96 155312 [581] Bobkova I V and Bobkov A M 2017 Phys. Rev. B 95 184518 [582] Tkachov G 2017 Phys. Rev. Lett. 118 016802 [583] Vasenko A S, Golubov A A, Silkin V M and Chulkov E V 2017 JETP Lett. 105 497 [540] Song J, Liu H, Liu J, Li Y X, Joynt R, Sun Q f and Xie [584] Vasenko A S, Golubov A A, Silkin V M and Chulkov E V X C 2016 Phys. Rev. B 93 195302 2017 J. Phys.: Condens. Matter 29 295502 [541] Zyuzin A, Alidoust M and Loss D 2016 Phys. Rev. B 93 [585] Hugdal H G, Linder J and Jacobsen S H 2017 Phys. Rev. 214502 B 95 235403 [542] Bobkova I V, Bobkov A M, Zyuzin A A and Alidoust M [586] Cayao J and Black-Schaffer A M 2017 Phys. Rev. B 96 2016 Phys. Rev. B 94 134506 155426 [543] Choudhari T and Deo N 2017 Phys. E 85 238 [544] Tkachov G 2017 Phys. Rev. B 95 245407 [545] Yu Q, Tao Z, Song J, Tao Y C and Wang J 2017 Sci. Rep. 7 10723 [546] Alidoust M and Hamzehpour H 2017 Phys. Rev. B 96 165422 [547] Blasi G, Taddei F, Giovannetti V and Braggio A 2019 [587] Lu B and Tanaka Y 2018 Philos. Trans. Royal Soc. A 376 20150246 [588] Volkov A, Magn´ee P, van Wees B and Klapwijk T 1995 Phys. C 242 261 [589] Tkachov G and Fal'ko V I 2004 Phys. Rev. B 69 092503 [590] Tkachov G 2005 Phys. C 417 127 [591] Fagas G, Tkachov G, Pfund A and Richter K 2005 Phys. Phys. Rev. B 99 064514 Rev. B 71 224510 [548] Rouco M, Tokatly I V and Bergeret F S 2019 Phys. Rev. B 99 094514 [549] Khezerlou M and Goudarzi H 2019 Mass-like gap creation by mixed singlet and triplet state in superconducting topological insulator (Preprint arXiv:1903.01144) [550] Tanaka Y and Kashiwaya S 1997 Phys. Rev. B 56 [551] Bauer E and Sigrist M (eds) 2012 Non-Centrosymmetric Superconductors: Introduction and Overview (Lecture Notes in Physics vol 847) (Berlin: Springer) [552] Yip S 2014 Annu. Rev. Condens. Matter Phys. 5 15 [553] Samokhin K 2015 Ann. Phys. 359 385 [554] Smidman M, Salamon M B, Yuan H Q and Agterberg D F [592] Tkachov G 2007 Phys. Rev. B 76 235409 [593] Rohlfing F, Tkachov G, Otto F, Richter K, Weiss D, Borghs G and Strunk C 2009 Phys. Rev. B 80 220507 [594] Kopnin N B and Melnikov A S 2011 Phys. Rev. B 84 064524 [595] Kopnin N B, Khaymovich I M and Mel'nikov A S 2013 Phys. Rev. Lett. 110 027003 [596] Bergeret F S, Volkov A F and Efetov K B 2001 Phys. Rev. Lett. 86 4096 [597] Bergeret F S, Volkov A F and Efetov K B 2005 Rev. Mod. Phys. 77 1321 [598] Bergeret F S and Tokatly I V 2013 Phys. Rev. Lett. 110 2017 Rep. Prog. Phys. 80 036501 117003 [555] Levitov L S, Nazarov Y V and Eliashberg G M 1985 JETP Lett. 41 445 [556] Edelstein V M 1995 Phys. Rev. Lett. 75 2004 [557] Yip S K 2002 Phys. Rev. B 65 144508 [558] Konschelle F, Tokatly I V and Bergeret F S 2015 Phys. [599] Eschrig M 2015 Rep. Prog. Phys. 78 104501 [600] Linder J and Robinson J W A 2015 Nat. Phys. 11 307 [601] Tanaka Y and Golubov A A 2007 Phys. Rev. Lett. 98 037003 [602] Yokoyama T, Tanaka Y and Golubov A A 2007 Phys. Rev. CONTENTS B 75 134510 [603] Asano Y and Tanaka Y 2013 Phys. Rev. B 87 104513 [604] Tamura S, Hoshino S and Tanaka Y 2019 Phys. Rev. B 99 184512 [605] Linder J and Balatsky A 2017 Odd-frequency supercon- ductivity (Preprint arXiv:1709.03986) [606] Sigrist M and Ueda K 1991 Rev. Mod. Phys. 63 239 [607] Buzdin A I 2005 Rev. Mod. Phys. 77 [608] Fritsch D and Annett J F 2014 J. Phys.: Condens. Matter 26 274212 [609] Fritsch D and Annett J F 2015 Supercond. Sci. Techn. 28 085015 [610] Triola C and Balatsky A V 2016 Phys. Rev. B 94 094518 [611] Tkachov G 2018 J. Phys.: Condens. Matter 31 055301 [612] D'yakonov M I and Perel' V I 1971 JETP Lett. 13 467 [613] Krive I V, Gorelik L Y, Shekhter R I and Jonson M 2004 Low Temp. Phys. 30 398 [614] Buzdin A 2008 Phys. Rev. Lett. 101 107005 [615] Mal'shukov A G, Sadjina S and Brataas A 2010 Phys. Rev. B 81 060502 [616] Black-Schaffer A M and Linder J 2011 Phys. Rev. B 83 220511 [617] Yokoyama T, Eto M and Nazarov Y V 2013 J. Phys. Soc. Jpn. 82 054703 [618] Yokoyama T, Eto M and Nazarov Y V 2014 Phys. Rev. B 89 195407 57 2014 Phys. Rev. B 89(17) 174425 URL https://link. aps.org/doi/10.1103/PhysRevB.89.174425 [637] Liu N, Teng J and Li Y 2018 Nature Communications 9 1282 ISSN 2041-1723 URL https://doi.org/10.1038/ s41467-018-03684-0 [638] Kou X, Guo S T, Fan Y, Pan L, Lang M, Jiang Y, Shao Q, Nie T, Murata K, Tang J, Wang Y, He L, Lee T K, Lee W L and Wang K L 2014 Phys. Rev. Lett. 113(13) 137201 URL https://link.aps.org/doi/10. 1103/PhysRevLett.113.137201 [639] Chang C Z, Zhao W, Kim D Y, Zhang H, Assaf B A, Heiman D, Zhang S C, Liu C, Chan M H W and Moodera J S 2015 Nature Materials 14 473 EP -- URL https://doi.org/10.1038/nmat4204 [640] Checkelsky J G, Yoshimi R, Tsukazaki A, Takahashi K S, Kozuka Y, Fa, Kawasaki M and Tokura Y 2014 Nature Physics 10 731 EP -- URL https://doi.org/10.1038/ nphys3053 [641] Chang C Z, Zhang J, Feng X, Shen J, Zhang Z, Guo M, Li K, Ou Y, Wei P, Wang L L, Ji Z Q, Feng Y, Ji S, Chen X, Jia J, Dai X, Fang Z, Zhang S C, He K, Wang Y, Lu L, Ma X C and Xue Q K 2013 Science 340 167 -- 170 ISSN 0036-8075 http://science.sciencemag. org/content/340/6129/167.full.pdf) URL http: //science.sciencemag.org/content/340/6129/167 (Preprint [642] Offidani M and Ferreira A 2018 Phys. Rev. Lett. 121 [619] Alidoust M and Halterman K 2015 New J. Phys. 17 126802 033001 [643] Liu H, Liu W E and Culcer D 2017 Phys. Rev. B 95 205435 [620] Mironov S V, Mel'nikov A S and Buzdin A I 2015 Phys. Rev. Lett. 114 227001 [621] Mal'shukov A G 2017 Phys. Rev. B 95 064517 [622] Amundsen M and Linder J 2017 Phys. Rev. B 96 064508 [623] Mal'shukov A G and Chu C S 2008 Phys. Rev. B 78 104503 [624] Mal'shukov A G and Chu C S 2011 Phys. Rev. B 84 054520 [625] Bergeret F S and Tokatly I V 2016 Phys. Rev. B 94 180502 [626] Linder J, Amundsen M and Risinggard V 2017 Phys. Rev. B 96 094512 [627] Mal'shukov A G 2018 Phys. Rev. B 97 064515 [628] Ghosh S and Manchon A 2017 Phys. Rev. B 95 035422 [629] Smejkal L, Mokrousov Y, Yan B and MacDonald A H 2018 Nature Phys. 12 242 [630] Novoselov K S, Mishchenko A, Carvalho A and Neto A H C 2016 Science 353 6298 [631] Tatara G 2018 Phys. Rev. B 98 174422 [632] Zhang J, Chang C Z, Tang P, Zhang Z, Feng X, Li K, Wang L l, Chen X, Liu C, Duan W, He K, Xue Q K, Ma X and Wang Y 2013 Science 339 1582 -- 1586 ISSN 0036-8075 http://science.sciencemag. org/content/339/6127/1582.full.pdf) URL http: //science.sciencemag.org/content/339/6127/1582 (Preprint [633] Zhang Z, Feng X, Guo M, Li K, Zhang J, Ou Y, Feng Y, Wang L, Chen X, He K, Ma X, Xue Q and Wang Y 2014 Nature Communications 5 4915 EP -- article URL https://doi.org/10.1038/ncomms5915 [634] Chang C Z, Zhang J, Liu M, Zhang Z, Feng X, Li K, Wang L L, Chen X, Dai X, Fang Z, Qi X L, Zhang S C, Wang Y, He K, Ma X C and Xue Q K 2013 Advanced Mate- rials 25 1065 -- 1070 (Preprint https://onlinelibrary. wiley.com/doi/pdf/10.1002/adma.201203493) URL https://onlinelibrary.wiley.com/doi/abs/10.1002/ adma.201203493 [635] Chang C Z, Tang P, Wang Y L, Feng X, Li K, Zhang Z, Wang Y, Wang L L, Chen X, Liu C, Duan W, He K, Ma X C and Xue Q K 2014 Phys. Rev. Lett. 112(5) 056801 URL https://link.aps.org/doi/10. 1103/PhysRevLett.112.056801 [636] Lee J S, Richardella A, Rench D W, Fraleigh R D, Flanagan T C, Borchers J A, Tao J and Samarth N 

1807.11230

1

1807

Intrinsic structural and electronic properties of the Buffer Layer on Silicon Carbide unraveled by Density Functional Theory

The buffer carbon layer obtained in the first instance by evaporation of Si from the Si-rich surfaces of silicon carbide (SiC) is often studied only as the intermediate to the synthesis of SiC supported graphene. In this work, we explore its intrinsic potentialities, addressing its structural and electronic properties by means Density Functional Theory. While the system of corrugation crests organized in a honeycomb super-lattice of nano-metric side returned by calculations is compatible with atomic microscopy observations, our work reveals some possible alternative symmetries, which might coexist in the same sample. The electronic structure analysis reveals the presence of an electronic gap of ~0.7eV. In-gap states are present, localized over the crests, while near-gap states reveal very different structure and space localization, being either bonding states or outward pointing p orbitals and unsaturated Si dangling bonds. On one hand, he presence of these interface states was correlated with the n-doping of the monolayer graphene subsequently grown on the buffer. On the other hand, the correlation between their chemical character and their space localization is likely to produce a differential reactivity towards specific functional groups with a spatial regular modulation at the nano-scale, opening perspectives for a finely controlled chemical functionalization.

cond-mat.mes-hall

cond-mat

56127 Pisa, Italy Abstract Intrinsic structural and electronic properties of the Buffer Layer on Silicon Carbide unraveled by Density Functional Theory Tommaso Cavallucci and Valentina Tozzini* NEST- Scuola Normale Superiore and Istituto Nanoscienze, Cnr, Piazza San Silvestro 12, The buffer carbon layer obtained in the first instance by evaporation of Si from the Si-rich surfaces of silicon carbide (SiC) is often studied only as the intermediate to the synthesis of SiC supported graphene. In this work, we explore its intrinsic potentialities, addressing its structural and electronic properties by means Density Functional Theory. While the system of corrugation crests organized in a honeycomb super-lattice of nano-metric side returned by calculations is compatible with atomic microscopy observations, our work reveals some possible alternative symmetries, which might coexist in the same sample. The electronic structure analysis reveals the presence of an electronic gap of ~0.7eV. In-gap states are present, localized over the crests, while near-gap states reveal very different structure and space localization, being either bonding states or outward pointing p orbitals and unsaturated Si dangling bonds. On one hand, he presence of these interface states was correlated with the n-doping of the monolayer graphene subsequently grown on the buffer. On the other hand, the correlation between their chemical character and their space localization is likely to produce a differential reactivity towards specific functional groups with a spatial regular modulation at the nano-scale, opening perspectives for a finely controlled chemical functionalization. 1. Introduction The thermal decomposition of silicon carbide (SiC) is a widely used technique to produce supported graphene1. Upon selective evaporation of Si from a Si-rich face of SiC (typically the 111 of 4C or the 0001 of hexagonal polytypes), excess carbon reconstruction produces in the first instance a honeycomb carbon layer covalently bound to the substrate called the buffer layer2 (BL). Due to the non negligible residual amount of sp3 hybridized sites, the BL is not graphene, but the latter can be obtained from it either by further evaporation of Si, producing the so called "epitaxial" graphene monolayer (EM), or by intercalation of hydrogen3 or metals4 underneath the buffer layer (Quasi Free Standing Monolayer Graphene, QFSMG5). In both cases, the (electronic) structure of the BL is a determinant of the graphene properties: in EM, the rippling of the BL determines the corrugation of the graphene sheet6 as observed in Scanning Tunneling Microscopy (STM)7,8. In QFSMG, the interaction with the BL is very weak, but vacancies in the intercalation layer produce localized states, appearing located in regular patterns9,10 which were proposed to descend from the BL symmetry11. Finally, the n-doping observed in EM was attributed to the effects of the surface states of the BL12. a pattern of corrugation appearing organized in crests forming an honeycomb-like lattice of nano-sized edge, whose periodicity was recognized as a "quasi" 6!6 (of SiC). However, analyses at low temperature somehow debated in the literature. Atomically resolved microscopy analyses at high temperature! reveal and different voltages",13,14 reveal different electronic structures, appearing more sharp and localized, and with the experimental observations. For QFSMG,#$,##, we recently proposed a alternative symmetry, different possible symmetries, including the 6"3!6"3 R30, which is the smaller consensus supercell between graphene and SiC when the two lattices are rotated of exactly 30 deg. This supercell is considered the "standard one" in several theoretical studies on BL, EM or QFSMG15,16,17,18,19. In some of these, the BL structure was studied and actually revealed a system of crests of ~1Å high, in agreement The conformation and electronic structure of the BL is therefore of primary importance, and namely "31!"31 R8.95 (of SiC), obtained by a slight rotation of the two lattices, and leading to a 1/3 sized consensus supercell which allows extensive calculations. At variance of those of EM and QFSMG, however, the electronic properties of the BL alone received little attention: only very recently an ARPES20 measurement has revealed the presence of a small band gap (~0.5eV) whose origin was not clearly attributed. The reason of the scarce interest towards this system is that it is usually considered only a support system to generate the most interesting one, that is, graphene. In this work we adopt a different point of view, focusing on the BL itself. We are moved by two aims. The first is to clarify details of the structure of this system, since this influences the structure of EM and QFSMLG and can shed light on the kinetics of their formation processes, still basically unknown. The second is the analysis of the electronic structure of the BL as a system per se, to envision whether the specific properties of the BL could be exploited for applications. Our investigation tool is Density Functional Theory (DFT). The calculation scheme is reviewed in the next section, also including a description of the model systems, especially the non-standard ones, used for here the first time with BL. The subsequent section illustrates the results. Summary and discussion on perspectives for applications are reported in the last section. 2. Models and methods 2.1 Model Systems The two model systems analyzed in this work are reported in Fig 1. The first one is the 6"3!6"3 R30 of SiC (or 13!13 of graphene, Fig 1, top part), considered the "standard" simulation cell for the SiC/graphene systems (hereafter called "L"). We included four SiC layers of the cubic polytype, and saturated the last one with H. In this work we also considered a smaller ("S") supercell recently used and validated for the QFSMG#$,##, the "31!"31 R8.95 of SiC (or 7!7 R21.787 of graphene, Fig 1, bottom part). In this case, the consensus between graphene and SiC supercells is obtained by allowing a small relative rotation (< 1 deg with respect to that in L model) of the two lattices and slightly different contraction of the buffer: in S model the surface density of the buffer atoms is 0.9% larger than in the L model. In the S supercell the BL can have two different stacking conformations## with respect to SiC, allowing respectively one site of type "hollow", i.e. with a Si atom lying under the center of a BL hexagon (Sh model) and one site of type "top", i.e. with Si atom lying under a C atom (St model, see red circles in Fig 1). These models are built locating a flat hexagonal carbon layer on the top of a neatly cut SiC surface at a distance slightly larger than the presumed bond distance and then relaxing (see next section for calculation details). The S supercell includes less than 1/3 atoms of the L one, reducing the computational of roughly one order of magnitude. Fig 1. Model systems used in this work. Top and side views of the large (L) model and small models with "hollow" (Sh) and "top" (St) buffer stacking are reported (red circles indicate the hollow and top sites). The lattice vectors in the xy plane are represented as blue arrows, and their values reported. Orange arrows indicate the crystallographic directions of the SiC lattice, cyan arrows those of graphene, their relative rotation is reported. The thick arrows indicate the projection directions of the side views. L model includes 1310 atoms, and S models include 377 atoms each. Color coding: black = Carbon of the BL, cyan = carbon of SiC, yellow = Si, white = H. The views show the structurally optimized geometries. DFT calculations are performed with a setup previously tested on this kind of systems%. The Rappe-Rabe- 2.2 Calculations Setup Kaxiras-Joannopoulos ultrasoft (RRKJUS) pseudopotentials21 are used with the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional22, van der Waals interactions are treated within the semi-empirical Grimme D2 scheme23 (PBE-D2). The plane wave cutoff energy was set at 30 Ry and the density cutoff at 300 Ry, while the convergence threshold for self-consistency was set at 10-8. The BFGS quasi-Newton algorithm was used for structural optimizations24, with standard convergence criteria, i.e. 10-3 a.u. for the forces and 10-4 a.u. for the energy. # point only is used for structural optimization, while for calculations of the density of states (DoS) we used a 10!10!1 grid generated using the Monkhorst and Pack scheme25 for S supercells and a 5!5!1 grid for L. A Gaussian smearing of 0.01 Ry was set in all calculations. The lattice vectors of the supercells in the xy plane (reported in Fig 1) are chosen in such a way to have a relaxed SiC crystal. The third supercell parameter, in z direction, is the same in all cases (31.8Å), chosen large enough to decouple the periodic copies. The calculations were performed using Quantum ESPRESSO26 (QE, version 5.3.0). Numerical data generated during this study are available from the corresponding author upon reasonable requests. 3. Results In this section the results of this study are reported. The rippling pattern of the L model (part of section 3.1), were previously described by us% and others&. However to our knowledge a deep-in characterization of the structural features of the buffer in L or in the alternative models S was never reported. We believe that it is important to fill this gap, especially for the interpretation of energetic and electronic data, reported in the subsequent subsections. 3.1 General structural properties The main structural properties are summarized in Fig 2. As it can be seen from panel (a), the structure of SiC lattice is rather regular in the bottom layer, and becomes less regular in the layer nearer to the surface, as indicated by the larger spread of the z coordinate distributions. The BL is located at an average distance of 2.4 Å from the Si layer, and displays a large vertical spread due to its rippling. The ripples are organized in specific patterns, clearly visible in panels (b)-(d) of Fig 2 (bright areas are protruding). In L the "crests" form a honeycomb lattice (profiled in yellow in panel (b)) separating three types of hexagonal irregular "tiles". This pattern is compatible with that observed by STM!,!$, although in those experiment can produce different sharpness and contrast",#",#'. Indeed, however, the experimental images the the crests appear smeared. This could be attributed to the thermal fluctuations, not included in these calculations. In addition, it should be considered that STM images depend on voltage, whose variation hexagonal "tiles" appears somehow more regular than those in L model. Interestingly, the corrugation pattern arising from model Sh, here first reported, is more similar in this respect, displaying only one kind of regular tile. The pattern of model St is instead formed by distorted hexagons, appearing "broken" on two sides. This situation appear similar to some what observed in some region of the STM of ref [2], where the crests seem to form rather parallel zig-zag lines weakly connected. Fig 2. Structure of the buffer layer. (a) A side view of the system (specifically L model) with the average z coordinate of the layers reported. On the right, with the same scale, the distribution of the z coordinates of all atoms are reported, for the L, Sh and St model separately (color coding in the plot). Panels (b) to (d) report top views of the buffer layer (in vdW spheres representation), colored according to the z coordinate (bright grey protruding areas, dark=intruding). Yellow lines follow the profile of protruding areas. The cyan circles highlight the groups of benzene-like rings. Superimposed as yellow balls in the top right corner are the locations of the upper layer Si atoms. The inset (e) shows a zoom-in of the white profiled area in panel (c). In (e), location of Si atoms are also indicated as green and pink dots. Pink arrows in panels (b)-(d) indicate the zig-zag direction of graphene.  Locally, the BL appears organized in benzene-like units (highlighted by blue circles in panels (b-d)). In the zoomed inset (panel (e)) the crests appear formed by interconnected chains of those (broken) units, while the tiles include six, seven or eight benzene units, separated by intruding C atoms (colored in black and indicated by the pink dots in (e)). By comparing with the location of the Si atoms of the upper layer (yellow balls), it can be seen that each of the intruding C atom is located on top of a Si atom. As it will be shown in the next sections, they are actually covalently bound. However, not all the Si atoms are bonded to C. For instance, the central Si atom colored in green in panel (d) is found in a hollow position i.e. in the center of the benzene unit, and it is unbound. Top and hollow position roughly alternate, but the mismatch between the lattices unregisters the stacking. This modulated alternating of favorable (top) and unregistered relative C-Si positions creates the specific patterns of bonds of crests and benzene sub-units. 3.2 Bonding patterns Fig 3 (a), reports the z profiles evaluated along different zig-zag lines in the three models shown in Fig 3 (b-d). The Si bonded C atoms are easily identified as those pointing downward, and are regularly separated by groups of C atoms belonging to the benzene units; the crest areas can be recognized as those pointing upwards. As it can be seen, the profiles of the L model (black and blue line in the bottom and top plot respectively) have the super-cell periodicity of ~3.2 nm, while the profiles of S models (magenta, green and red) are only approximately periodic, since the zig-zag line is not perfectly aligned to the crests, due to the rotation angle between substrate and BL. Nevertheless, the crests structures of the three models are locally superimposable (black vs magenta lines, belonging to L and Sh respectively, and red vs blue lines, belonging to L and St). The green profile is the most different of all, because it crosses the "broken crest", which is a peculiarity of the St model. The covalent bonds between BL and the substrate can be identified in different ways: (i) by selecting the buffer layer atoms pointing downwards, or (ii) those at short distance from a Si atom, or (iii) by searching for the effective charge localization between buffer and substrate. Method (i) is the simplest but ambiguous, because the tail of distribution of z coordinates of bound and unbound atoms superimpose. Method (ii) identifies better the bound atoms, being the bond length distributed at ~2Å. Finally, method (iii) gives an immediate idea of the bonding pattern: Fig 3 (f-h) reports a representation of the total charge density evaluated on a plane located mid-way between the buffer and the substrate, as shown in (e). Bonds can be identified as charge density accumulation points (bright spots), located below a sub-set of the BL atoms. These are colored in pink in the superimposed image, and as anticipated, are distributed on a specific hexagonal pattern, alternating with the "benzene" rings (in cyan) appearing as grey hexagon in (b)-(d) and (f)-(h) due to their $ orbitals structure. (The atoms of the crests appear dark in the image because they are unbound and far above the representation plane.) Fig 3 panels (f-h) also confirm the local resemblance of patterns of model L to that of model Sh and St. A comparative analysis of the three methods is reported in the SI. 3.3 Formation energies The evaluation of relative energies of L and S models is not straightforward, because they have different sizes. One possibility is to normalize to the area of the cell, or equivalently, to the number of surface Si atoms. However, the surface density of atoms of the carbon layer is higher in the S model of ~1% (and, correspondingly, the corrugation level is of 2% larger see Table 1, two last columns). Therefore, a second non-equivalent normalization is to the number of atoms the BL. Even the evaluation of the total energy itself is non trivial. The global stability could be evaluated considering the formation energy Ef, i.e. the energy with respect to the sum of energy of all isolated atoms !!!!!"#! !!!!!!!!!!!!"!!" When renormalize to the unit surface, Ef can be used as a measure of the relative stability, although it is not very meaningful per se, being dependent on the number of layers of bulk included in the model. The binding/dissociation energy of the BL was also evaluated, but it is less appropriate considering that the synthesis proceeds from the bulk (details in the SI). The analysis of the energies indicates that Sh model is more stable than St model of about 0.01eV per surface Si atom. This is likely to be due to a more symmetric conformation of the bonding pattern, since the concentration of C atoms and the level of corrugation is basically the same in St and Sh. Additionally, S models turns out more stable of L of ~4.8 eV/Si surface atom. Because the main difference between S and L is the surface C density of the buffer, this points to a preference of the system for a slightly larger concentration, such as in S model. In general, however, the stabilization is likely to arise from the interplay between the different corrugation level (see Table 1) and the different arrangement/symmetry of the Si-C bonds and of the crests. Conversely, when one considers the formation energy per C buffer atom, the trend is inverted and L results lower than S of about 3.9 eV/ C atom. On the formal level, this is trivially due to the different surface concentration of C atoms in the two models and seems to indicate that the BL is more relaxed within the L model. Fig 3. (a) z profiles of the BL atoms along the zig-zag lines depicted in panels (b)-(d), for the models L (b), Sh (c) and St (d) (colors of lines and plots correspond). The profiles are built including all the atoms of a single zig-zag line, the l coordinate being the projection of x-y coordinate onto this line. This imply that the separation of dots along the l coordinate of ~1.23Å. The AFM- like images (b)-(d) are obtained from the iso-density surfaces of the total electronic charge density, colored according to the eight (bright protruding, dynamical range of coloring = 1Å). In panels (e)-(h) the charge density is evaluated on a plane located between the buffer layer and the substrate, as shown in panel (e). Panels (e-f) report the value of the charge density on the plane (bright high density, dark low density), for models L (f), Sh (g) and St (h). Superimposed in red is the BL structure, and in colors the atoms of BL of the three different types: bound to SiC, "benzene-like" rings and crests, colored as shown in the color coding legenda in panel (e). 38.306 38.654 38.654 14.8 15.1 15.2 Ef (eV) NSi NC 338 98 98 Ef/NC (eV) -534.7286 -530.8805 -530.8776 Ef/NSi (eV) -1673.5024 -1678.2675 -1678.2581 Nb Nb /NC NC/S (nm-2) <!2>1/2 (deg) 83 24.6% 24 24.5% 25 25.5% L -180738.25 108 Sh -52026.293 31 St -52026.000 31 Table 1 Formation Ef and binding energies Eb of the three model systems as defined in the main text, and normalized to the to the number of Si surface atoms, of C buffer atoms and to the number of Si-C bonds, also reported. The average surface density of the buffer atoms, NC/S and the average corrugation level of the buffer, as measured by the standard deviation from zero of the out of plane dihedral are also reported <!2>1/2. 3.4 Electronic properties The electronic Density of States (DoS) for all models are reported in Fig 4. Once aligned to the Fermi level, they show the same global structure, displaying a low DoS area ~0.7 eV large, delimited by two rapidly increasing wings. This is an electronic gap of size comparable of that reported in ref [20], although a n-doping is present being the Fermi level located above the gap. A zoom in of the DoS (lower plot Fig 4) reveals some differences between the three models. Some residual DoS is present in the gap, revealing some more or less pronounced or separated "peaks" in the three models. The local DoS integrated in the gap range is reported as insets in Fig 4 (iso-density surfaces, in red), and clearly reveals that in-gap states are localized over the crests in the three cases. At a first sight, these states appear to be the $ orbitals of the protruding C atoms. For the St case, where two separate peaks are clearly visible, we integrated over them separately (intervals indicated in yellow and purple in the plot). The inset on the right shows in corresponding colors the separated local DoS, and reveals that the lower energy in-gap peak (yellow) belong to states located on the more irregular sites, while the states belonging to higher energy in-gap peak (magenta) follow the more regular hexagonal pattern, which explains why in the Sh case only one peak (the "regular" one) is present, being the symmetry higher. Fig 4. Total electronic DoS the L and S models. In the upper plot, the total DoS in the full energy range is reported for the three models (color coding reported) aligned to the Fermi Level. The shaded area includes the filled states. In the lower plot a zoom around the Fermi level is reported (same color coding for the lines). The three central insets report in red an iso-surface representation of the Local DoS integrated within the "gap" range [-0.7;-1.4] (the models are indicated and the insets are profiled with the same color coding of the lines); in the side inset profiled in green, the DoS of the St model integrated over the two separated in-gap peaks in the intervals indicated in yellow and purple in the plot are reported separately (in corresponding colors). We systematically performed this analysis in other four energy intervals, namely the left wing (LW) and the right wing (RW) of the gap up to the Fermi level, and the the "LUMO" state, including a the lower unoccupied states up to 1eV (boundaries of the intervals indicated by vertical lines in Fig 4, lower plot). The local DoS of these intervals (and of the gap) are reported in Fig 5 in different representations aimed at showing the vertical localization of the states: The first two representations ("STM" and "vol slice over") display the electronic density on top of the buffer layer, and could be compared with STM  images in fixed current and fixed height mode, respectively. The third one ("vol slice under") is obtained plotting the density in a plane located between the buffer and the substrate, and displays the inner part of the charge density. This analysis confirms that the in-gap states (second column from the left) are basically localized in the crests area: the superficial images show density localized on the protruding C atoms of the crests, while the inner image shows density localized under the C atoms intruding, i.e. those covalently bound to the substrate. Overall, then, the in-gap states represent bonding orbitals either between the crests atoms or between the intruding crests atoms and the substrate. The in-gap states do not display charge density on the "benzene" units. Conversely, the distribution of the LW states is localized on the benzene units, or under them, as shown by the "slice under" images. Besides those clearly visible rings, the inner images show also density localized under the C sites interstitial between rings, which are bonded to the substrate. Also the superficial image shows density on the benzene rings, visible as small circles within the larger hexagonal shapes delimited by crests, besides some residual density of the $ orbitals over the protruding atoms. Overall, these states appear to have bonding character, either between the out of crest atoms, or between them and the substrate. The charge density of the states of the RW below the Fermi level has a strong inner component under the "tiles", which appear however to be different, being localized over Si atoms located in the hollow positions, and therefore these states do not participate to the covalent bonding to the substrate, but are rather "dangling Si bonds". The superficial charge density is localized mainly on the crests, but appears to be a sum of localized pz orbitals rather than a $ system. These two aspects are similar and enhanced in the "LUMO" states: the superficial states are protruding pz orbitals and the inner states are dangling bonds under the "benzene rings" (although with a different symmetry with respect to the RW states). Overall, then, the states above the gap appear to have an increasing anti-bonding character. 4. Discussion and conclusions In this work we have comparatively analyzed results from three different model systems for the buffer carbon layer on the Si-rich surface of SiC: the "standard one" (L) with exactly 30 deg rotation between the SiC and buffer lattices, 13!13 / 6"3!6"3 supercells and a superficial C buffer atoms density of ~38.3 C atom/nm2, and two alternative ones (Sh and St with 7!7 / "31!"31 supercells), with a slightly different rotation between the two lattices (~29 deg), slightly larger superficial C density (~38.7 C/nm2). L is fully compatible with the (quasi) 6!6/SiC supercell, while S models are compatible only approximately. The three models share common general structural features: a system of crests of alternating protruding and intruding C atoms forming an honeycomb pattern and separating hexagonal "tiles" of ~1nm side. Within the "tiles" carbon organizes in "benzene-like" rings, separated by intruding C sites. These and the intruding sites in the crests are covalently bound to Si atoms of the substrate. The main differences between the three models are in the form of the "tiles": these are regular hexagons in the Sh model, distorted hexagons with partially broken sides in St, while in the L model three different types of hexagons, one regular and two distorted are present. The comparison with the atomic microscopy data does not completely solve the ambiguity between the models: AFM and STM images clearly show the honeycomb system of crests of size compatible with all of them. Small irregularities of the tiles seem to appear in some images!$,#", while in others they are not visible!. In general, however, crests appear smeared and regularized with respect to L model. While both the effects could be attributed to the already mentioned thermal fluctuations (neglected in our calculations), the regularization is an indication in favor of Sh model. Some features of the fine structure we observe, e.g. the "benzene-like rings", are barely reported in the literature. However, we observe that these need very high resolution and specific voltage conditions to be revealed. In fact, structures that resemble those of Fig 5, with modulation of contrast on the crests and within the tiles similar to benzene rings were observed in ref [#'], where a systematic study at different voltages and temperature was performed. In summary, the theory-experiment comparison of structural data would indicate that none of the models should be excluded.  Fig 5. Local DoSs integrated over different energy intervals (columns: LW= left wing [-2.3;-1.4]; gap [-1.4; - 0.7]; RW= right wing [-0.7;0]; "LUMO" [0;+1]) for the three different models (from top to bottom: Sh, St and L, each represented in three different modes: STM = "fixed current" mode, obtained coloring an iso-density surface (~10-4 a.u.) according to the height; "vol slice over" obtained coloring a plane placed over the layer according to the local value of the density, and "vol slice under", same but with the plane placed between the buffer -- reported in red in wireframe representation -- and the substrate (the location of the two planes is represented in the forth row, first cell). Also represented are an iso-density surface of the "gap" local DoS (in red, first row, first cell) and of the LUMO state (in red, 7th row, first cell). Other indications come from the analysis of stability. The comparison between L and S models indicates that the formation energy per unit surface is smaller for the S models (the smallest for Sh), while the same evaluated per C atom of the BL is smaller for L. This is not a contradiction, however, because the superficial concentration of C atoms of the buffer is larger in S models. It might seem not intuitive that a more crowded (and consequently more corrugated) conformation is more stable than a more "diluted" (and slightly flatter). However, one must consider that the buffer is not a completely flat sp2 graphene, nor even a completely sp3 hybridized graphane-like sheet, but a combination of the two, whose optimal corrugation/contraction level is not easy to evaluate a priori. Therefore it is not straightforward to estimate which is the optimal superficial density of C atoms in the buffer. In addition, Sh conformation might be structurally favored by its high symmetry, bringing a particularly favorable binding pattern. This conclusion is also compatible with a recent general theoretical analysis showing that conformations with relative rotations of lattices with low commensuration might help releasing the strain27. Our results for the electronic structure return similar DoS for the three models, displaying a gap approximately 0.7 eV wide, with localized in-gap states and n-doping. We remark that, while a similar n- doping level is reported in the theoretical calculations (e.g. !$), this is not fully confirmed by experiments, the bulk polarization in the theoretical calculations ##,!$. However, the doping level is not likely to affect which seem rather to indicate the Fermi level located in proximity of the gap or within it. This may be attributed to different causes: extrinsic doping induction from the experimental setup or poor treatment of the features (shape, localization, extension) of the electronic states, which in fact, appear to be fully compatible with the structural analysis. The basically non-dispersive in-gap states appear to descend from the $ orbital system of the crests atoms and are fully localized on them, constituting the binding system between crests atoms or among them and the substrate. The specific space distribution of those state of course depends on the model (being the crests different in the three models) and in particular the symmetry breaking occurring in Sh seems to split the in-gap state in two distinct ones, with separate spatial distribution, while this splitting is not observed in the more symmetric Sh model. Furthermore, just below the gap, we observe a more conventional $ binding system, localized in the "benzene like" rings within the "tiles", though with some residual density on the crests. On the other side, above the gap, we revealed a set of dangling Si bonds localized underneath the buffer layer, and a set of pz orbitals mainly localized over the crests and in proximity of them. These states might have relevance in the intrinsic doping of the graphene monolayer on BL, when present, which was attributed to donor surface states either of the buffer or of the SiC surface#!. The experimental observation of this variety of mentioned benzene rings#', localized pz orbitals or dangling bonds#", often organized in triangular or states requires a systematic study STM analysis of the BL at different voltages, performed only in a few studies, which in fact, reveal the presence of structures we observe in our analysis, such as the already hexagonal symmetries, compatible with our models. In those images both regular and irregular hexagonal "tiles" are observed, as far as areas with broken crests typical of St model. In summary, our work does not completely solve the problem of which is the exact symmetry and conformation of the BL, conversely suggesting that different symmetries might coexist and possibly inter- convert at sufficiently high temperature. The symmetry of L gives a more extended spatial commensurability, a relative rotation that seems globally more compatible apparently exact 30 deg rotation of the two lattices. However S models seems to be locally more stable. The preference for one or the other conformation should be investigated accounting also for the kinetics of the buffer formation, which is currently under investigation and matter of a forthcoming paper 28. From the modeling perspective, S models are much less expensive, and certainly a good approximation of the real sample. The peculiar structural and electronic properties of the BL suggest interesting applications. The corrugation and the considerable variability of electronic structure between crests and "tiles" suggest that this system is particularly reactive29: states localized in the different energy regions (gap, and its sides) display well separate spatial localization and very different (sometimes opposite) "chemical" character. The correlation between these two features suggests that different chemical species (e.g. electrophilic, nucleophile, or dienophile) might select different spatial areas, following the moiré superlattice, offering a  unique potentiality of obtaining a nano-patterned chemical functionalization. Acknowledgments We thank Dr. Stefan Heun, Prof. Paolo Giannozzi, Dr Luca Bellucci, Dr. Camilla Coletti, Dr. Yuya Murata, and Dr. Vittorio Pellegrini for useful discussions. We gratefully acknowledge financial support by EU-H2020, Graphene-Core1 (agreement No 696656) and Core2 (agreement No 725219), by CINECA awards IsB11_flexogra (2015), IsC36_ElMaGRe (2015), IsC44_QFSGvac (2016), IsC44_ReIMCGr (2016) and PRACE "Tier0" award Pra13_2016143310 (2016). We acknowledge CINECA staff for technical support. Author contributions TC performed calculations, produced data and performed part of the analyses. VT finalized the analyses, produced the figures and wrote the manuscript. Additional information Competing interest: the authors declare no competing interests. References (((((((((((((((((((((((((((((((((((((((((((((((((((((((( 1 Norimatsu, W.; Kusunoki, M. Epitaxial Graphene on SiC{0001}: Advances and Perspectives. Phys. Chem. Chem. Phys. 16, 3501 (2014) 2 Goler, S.; Coletti, C.; Piazza, V.; Pingue, P.; Colangelo, F.; Pellegrini, V.; Emtsev, K. V.; Forti, S.; Starke, S.; et al. Revealing the Atomic Structure of the Buffer Layer between SiC (0001) and Epitaxial Graphene. Carbon 51, 249%254 (2013) 3 Riedl, C.; Coletti, C.; Starke, U. Structural and Electronic Properties of Epitaxial Graphene on SiC(0001): a Review of Growth, Characterization, Transfer Doping and Hydrogen Intercalation. J. Phys. D: Appl. Phys. 43, 374009 (2010) 4 Fiori S, Murata Y, Veronesi S, Rossi A, Coletti C, and Heun S Li-intercalated graphene on SiC(0001): An STM study Phys. Rev. B 96, 125429 (2017) 5 Riedl C., Coletti C., Iwasaki T., Zakharov A. A., and Starke U. Quasi-Free-Standing Epitaxial Graphene on SiC Obtained by Hydrogen Intercalation Phys. Rev. Lett. 103, 246804 (2009) 6 Mallet, P.; Varchon, F.; Naud, C.; Magaud, L.; Berger, C.; Veuillen, J. Y. Electron States of Mono- and Bilayer Graphene on Si Probed by Scanning-Tunneling Microscopy. Phys. Rev. B:. 76, 041403 (2007) 7 Telychko, M.; Berger, J.; Majzik, Z.; Jelínek, P.; !vec, M. Graphene on SiC(0001) Inspected by Synamic Atomic Force Microscopy at Room Temperature. Beilstein J. Nanotechnol. 6, 901%906 (2015) 8 Cavallucci T and Tozzini V Multistable Rippling of Graphene on SiC: A Density Functional Theory Study J. Phys. Chem. C, 120, 7670%7677 (2016) 9 Murata Y, Mashoff T, Takamura M, Tanabe S, Hibino H, Beltram F, and Heun S, Correlation between morphology and transport properties of quasi-free-standing monolayer graphene App Phys Lett 105, 221604 (2014) 10 Murata Y, Cavallucci T, Tozzini V, Pavli&ek N, Gross L, Meyer G, Takamura M, Beltram F, Hibino H, Heun S. Atomic and Electronic Structure of Si Dangling Bonds in Quasi-Free-Standing Monolayer Graphene Nano Res 11 , 864 -- 873 (2018) 11 Cavallucci T, Murata Y, Takamura M, Hibino H, Heun S, Tozzini V, Unraveling localized states in quasi free standing monolayer graphene by means of Density Functional Theory Carbon 130, 466-474 (2018) 12 Ristein J., Mammadov S., Seyller Th. Origin of Doping in Quasi-Free-Standing Graphene on Silicon Carbide Phys Rev Lett, 108, 246104 (2012) 13 Riedl C "Epitaxial Graphene on Silicon Carbide Surfaces: Growth, Characterization, Doping and Hydrogen intercalation" PhD Thesis (2010), Fridrich Alexander Universität Erlangen, Nurberg 14 Hu T. W., Ma F., Ma D. Y., Yang D., Liu X. T., Xu K. W., Chu P.K. Evidence of atomically resolved 6x6 buffer layer with long-range order and short-range disorder during formation of graphene on 6H-SiC by thermal decomposition App Phys Lett, 102, 171910 (2013) 15 Kim, S.; Ihm, J.; Choi, H. J.; Son, Y.-W. Origin of Anomalous Electronic Structures of Epitaxial Graphene on Silicon Carbide. Phys. Rev. Lett. 100, 176802, (2008) 16 Varchon, F.; Mallet, P.; Veuillen, J.-Y.; Magaud, L. Ripples in Epitaxial Graphene on the Si-terminated SiC(0001) Surface. Phys. Rev. B: Condens. Matter Mater. Phys. 77, 235412 (2008) 17 Sforzini J., Nemec L., Denig T., Stadtmüller B., Lee T.-L., Kumpf C., Soubatch S., Starke U., Rinke P., Blum V., Bocquet F. C., and Tautz F. S. Approaching Truly Freestanding Graphene: The Structure of Hydrogen-Intercalated Graphene on 6H- SiC(0001) Phys Rev Lett 114, 106804 (2015) 18 Deretzis I. La Magna A. Role of covalent and metallic intercalation on the electronic properties of epitaxial graphene on SiC(0001) Phys Rev B 84, 235426 (2011) 19 Cavallucci T, Tozzini V Multistable Rippling of Graphene on SiC: A Density Functional Theory Study J of Phys Chemi C 120 7670-7677 (2015) 20 Nair M N., Palacio I, Celis A, Zobelli A, Gloter A, Kubsky S, Turmaud J-P, Conrad M, Berger C, de Heer W, Conrad E H, Taleb-Ibrahimi A, Tejeda A Band Gap Opening Induced by the Structural Periodicity in Epitaxial Graphene Buffer Layer Nano ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( Lett., 17, 2681%2689 (2017) 21 Rappe, A. M.; Rabe, K. M.; Kaxiras, E.; Joannopoulos, J. D. Optimized Pseudopotentials. Phys. Rev. B: 41, 1227 (1990) 22 Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 77, 3865%3868. (1996) 23 Grimme, S. Semiempirical GGA-type Density Functional Constructed with a Long-Range Dispersion Correction. J. Comput. Chem. 27, 1787 (2006) 24 Billeter, S. R.; Curioni, A.; Andreoni, W. Efficient Linear Scaling Geometry Optimization and Transition-State Search for Direct Wavefunction Optimization Schemes in Density Functional Theory Using a Plane-Wave Basis. Comput. Mater. Sci. 27, 437. (2003) 25 Monkhorst, H. J.; Pack, J. D. Special Point for Brillouin Zone Integration. Phys. Rev. B 13, 5188. (1976) 26 Giannozzi, P. QUANTUM ESPRESSO: a Modular and Open- Source Software Project for Quantum Simulations of Materials. J. Phys.: Cond. Matt 21, 395502 (2009) 27 Sclauzero G., Pasquarello A Low-strain interface models for epitaxial graphene on SiC(0001) Diam & Rel Mat 23 178 -- 183. (2012) 28 Bellucci L, Cavallucci T, Tozzini V From buffer layer to graphene on Silicon Carbide: a multi-scale simulation, in preparation 29 Cavallucci T, Kakhiani K, Farchioni R Tozzini V Morphing graphene-based systems for applications: perspectives from simulations, GraphITA. Carbon Nanostructures., 87-111 (2017) Intrinsic structural and electronic properties of the Buffer Layer on Silicon Carbide unraveled by Density Functional Theory Tommaso Cavallucci and Valentina Tozzini NEST- Scuola Normale Superiore and Istituto Nanoscienze, Cnr, Piazza San Silvestro 12, 56127 Pisa, Italy Supporting information S.1. Distributions of bonds As mentioned in the main text, the covalent bonds distribution can be evaluated in different ways (i) by evaluation of the z coordinate (ii) by selecting the Si-C minimum distance for each C atom in the BL and (iii) evaluating the charge density value in a point midway between each C atom in the BL and the nearest Si atom (see Fig S.1 b). Fig S.1 (a) compares the methods (ii) and (iii) showing the distribution of Si-C minimal distances and the values of charge density in the midway point. The two sets of distribution (three for each method, for the three models) both display three separate populations, corresponding to the bound atoms (small Si-C distance, high density), the unbound atoms (intermediate distance and density) and crests atoms (large distance, small density). The populations selected with the two methods coincide and are represented in colors in Fig S.2, (a-c). Fig S.1 (a) Distributions of the minimum distances between C atoms of the BL and Si atoms of the substrate (top) and of the charge density value evaluated in the middle points of the Si-C distance from a C of the buffer and the nearest Si (the latter shown with inverted abscissa, to better compare with the fomer). (b) Total electronic charge density evaluated on a vertical plane cutting some Si-C bonds (indicated) and on an horizontal plane located between the BL and the substrate. The comparison of methods (ii)-(iii) and (i) is reported in Fig S.2, (d)-(e). It can be seen that the population of bound and unbound atoms are not very well separable on the basis of the z coordinate, because their tail superimpose. The bonds spatial distribution is visible also in Fig S.3 as they appear as bright spots when the charge density is plotted on a plane between the BL and the substrate. In Fig S.3, also visible (less bright) is the aromatic ! system of the "benzene" rings (corresponding to "unbound" atoms in dark grey in Fig S.2). As explained in the text, the rings are separated by bound atoms.  Fig S.2 Space distribution of bound (yellow), unbound (dark grey) and crests atoms (light grey), in L (a), Sh (b) and St (c) models, as selected with the method (ii). (d-f) report the height distribution of the three classes of atoms, colored as in (a-c). (b) (c) (a) Fig S.3 Charge density distribution plotted on a plane located between the BL and the substrate, as in Fig S.1 (b) (upper), for the three models (a)=L, (b)=Sh, (c)=St. S.2. Binding energy An alternative to the formation energy Ef to evaluate the relative stability of the models is the binding energy, e.g. the energy of the system evaluated with respect to the graphene sheet separated from the substrate !!!!!"#! !!"#!!!" being Eopt, Esub and Egr the energy of the systems and of their substrate and graphene isolated components respectively evaluated in the same cell. Eb accounts both for the chemical energy included in the covalent bonds between the buffer and the substrate and of their structural readjustment due to binding. This quantity (and its value per unit surface, per Si atom, per C atom and per bond is reported in Table S.1. For S models the estimated dissociation energy of the buffer to the separated graphene is ~0.7eV (little larger for Sh) per Si-C covalent bond, which is little less than what expected for an average value for a covalent bond. Conversely the value for L model is considerably larger, indicating that the sheet is more difficult to detach in the L model than in S. We remark, however, that this is just a very rough evaluation of the binding energy, especially because the reference system is not the flat graphene, but a laterally compressed graphene sheet. Therefore the reference systems in the two cases are different being the lateral compression different. Nb Eb (eV) L 83 -199.667 Sh -16.870 24 St -16.577 25 Table S.1 Binding energies Eb of the three model systems as defined in the main text, and normalized to the to the number of Si surface atoms, of C buffer atoms and to the number of Si-C bonds, also reported. The average surface density of the buffer atoms, NC/S is also reported. Eb/NSi (eV) -1.849 -0.544 -0.535 NC/S (nm-2) 38.306 38.654 38.654 Eb/NC (eV) -0.5907 -0.1721 -0.1692 NSi 108 31 31 NC 338 98 98 Nb /NC 24.6% 24.5% 25.5% Eb/Nb (eV) -2.406 -0.703 -0.663 

1903.06645

1

1903

Engineering Effects of Vacuum Fluctuations on Two-dimensional Semiconductors

The resonance energy and the transition rate of atoms, molecules and solids were understood as their intrinsic properties in classical electromagnetism. With the development of quantum electrodynamics, it is realized that these quantities are linked to the coupling of the transition dipole and the quantum vacuum. Such effects can be greatly amplified in macroscopic many-body systems from virtual photon exchange between dipoles, but are often masked by inhomogeneity and pure dephasing, especially in solids. Here, we observe an exceptionally large renormalization of exciton resonance and radiative decay rate in transition metal dichalcogenides monolayers due to interactions with the vacuum in both absorption and emission spectroscopy. Tuning the vacuum energy density near the monolayer, we demonstrate control of cooperative Lamb shift, radiative decay, and valley polarization as well as control of the charged exciton emission. Our work establishes a simple and robust experimental system for vacuum engineering of cooperative matter-light interactions.

cond-mat.mes-hall

cond-mat

1 Engineering Effects of Vacuum Fluctuations on Two-dimensional Semiconductors Authors: Jason Horng1*, Yu-Hsun Chou1,2 , Tsu-chi Chang2, Chu-Yuan Hsu2, Tien-chang Lu2, Hui Deng1* Affiliations: 1Department of Physics, University of Michigan, Ann Arbor, Michigan 48109-1040, United States. 2Department of Photonics, College of Electrical and Computer Engineering, National Chiao Tung University, Hsinchu 300, Taiwan. *Correspondence to: [email protected], [email protected] Abstract: The resonance energy and the transition rate of atoms, molecules and solids were understood as their intrinsic properties in classical electromagnetism. With the development of quantum electrodynamics, it is realized that these quantities are linked to the coupling of the transition dipole and the quantum vacuum. Such effects can be greatly amplified in macroscopic many-body systems from virtual photon exchange between dipoles, but are often masked by inhomogeneity and pure dephasing, especially in solids. Here, we observe an exceptionally large renormalization of exciton resonance and radiative decay rate in transition metal dichalcogenides monolayers due to interactions with the vacuum in both absorption and emission spectroscopy. Tuning the vacuum energy density near the monolayer, we demonstrate control of cooperative Lamb shift, radiative decay, and valley polarization as well as control of the charged exciton emission. Our work establishes a simple and robust experimental system for vacuum engineering of cooperative matter-light interactions. 2 The Lamb shift of the atomic transition frequency arises from the emission and re-absorption of virtual photons by single atoms.(1) Its discovery ushered the development of quantum electrodynamics and led to the surprising realization that the vacuum is not empty. The quantum fluctuations in vacuum couples with the transition dipole of matter and modifies its resonance energy and transition rate(1 -- 8). The effects of the vacuum coupling become enhanced collectively in certain many-body systems by coherent exchange of virtual photons among the dipoles, leading to, for example, the cooperative Lamb shift(5, 9, 10) and superradiance(6, 10 -- 12). However, strong dephasing and inhomogeneity in many-body systems have rendered these effects extremely difficult to observe experimentally. The cooperative Lamb shift in the optical domain has only been reported in a few experiments with cold atoms or ions(9, 13 -- 17). In solids, it has only been observed in nuclei excited by synchrotron x-rays and in superconducting microwave circuits(18, 19). Recently, high reflectivity was measured from a mere single layer of transition metal dichalcogenides crystal(TMDC)(20, 21), owing to the large exciton-photon coupling strength, near radiative-limited linewidth and two-dimensional(2D) translational symmetry. Based on this, theoretical work suggests monolayer TMDCs may provide an easy-to-access 2D many-body system for observing and utilizing the effects of vacuum fluctuations(22, 23). In this work, using a mirror to control the dipole transition of excitons in a high-quality, monolayer TMDC, we demonstrate the cooperative Lamb shift of the excitons accompanied by modified exciton radiative decay rate. This observation suggests that the TMDC monolayer can be used as a sensitive probe of the vacuum modes(7, 24, 25). We also demonstrate control of the charged exciton and valley polarization in the TMDCs with the tunable vacuum environment. In contrast to the strong coupling and Purcell effect, where a cavity is used to modify the resonances, in this work we have a fully open system coupled to a continuous spectrum of modes. Figure 1A shows a schematic of our system. A monolayer TMDC is placed at a distance L from a mirror made of a distributed Bragg reflector(DBR) (Fig. 1A). The mirror imposes a boundary condition on the electromagnetic field in the space, creating a standing-wave pattern E(r) = 2E0sin(kr)with an electric field node at the mirror plane. Here E0 and k are the electric field amplitude and the wavenumber of the incident planewave, and r is the distance from the mirror. This mode structure applies to not only classical fields but also the vacuum fluctuations. The structure of the vacuum fluctuations can be measured through its effect on a dipole emitter, such as an exciton in solids. Given a transition dipole, the radiative decay of the excited state to the ground state is proportional to the strength (spectral density) of electromagnetic fluctuations near the resonance frequency ω that are present in the environment. Therefore, measuring the lifetime of the excitation probes the local strength of vacuum fluctuations at the transition frequency. In effect, the monolayer can act as a local vacuum field analyzer. The effects of vacuum fluctuations on the exciton resonance in a 2D material can be modeled by the following Hamiltonian in the rotating wave approximation: H= ℏω0b†b+∫dkℏωkak†ak−ig∫dksin(kL)(akb†+ak†b) (1) where b†,ak†, b, ak are creation and annihilation operators for an exciton and a photon, respectively, ℏω0, ℏωk are the corresponding energies of the exciton and the photon, g is the 3 dipole coupling constant between the exciton and photon field, L is the distance between the monolayer and the mirror. The factor sin(kL) represents the spatial mode structure of the electric field in front of a mirror. With the time-dependent Schrodinger equation, we solve for the exciton wavefunction X(t)and obtain the exciton resonance energy and radiative decay rate. As shown in the supplementary, the solution for 𝐿≪2𝜋𝑐/𝛾(𝛾 is the radiative decay rate for a free-standing monolayer) can be written as: X(t)=X(0)exp [−iE0t/ℏ]exp [−γ2t] (2) where E0=E0−ℏγ2sin(2kL) (3a) γ=2γcos2(kL) (3b) The physical meaning of E0 and γis the renormalized exciton energy and exciton radiative rate, respectively, modified by the vacuum field. The renormalized exciton radiative rate γ equals 2γ a at the anti-node(kL = (n+0.5)π) and zero at the node(kL = nπ) of the electric field. This is due to modification of exciton-photon coupling through local electric field. At the anti-node, the local field 2E0 enhances both the absorption and the emission rate, similar to the Purcell effect in a cavity configuration. At the node, local electric field is suppressed, therefore radiative decay of the exciton excitation is suppressed. The renormalized exciton energy E0 is shifted from 𝐸0 due to coupling with the vacuum field. This energy shift in a collective excitation system has been named as "cooperative Lamb shift". It shares the same origin as the Lamb shift but can be as large as the radiative linewidth ℏγ, due to the cooperative enhancement. In a radiative-limited sample, the renormalization on the resonance energy and the radiative decay rate can be directly observed through spectra of the exciton in frequency domain. An alternative approach is to treat the renormalization as the result of interaction between the monolayer dipole moment, induced by vacuum fluctuations, and its image dipole(Fig. 1B). In our case, the image dipole has a π phase difference compared with the original dipole due to the boundary condition set by the mirror. The renormalization of the radiative decay rate can be understood as follows: when the two dipole sheets are separate by nλ(the node condition), the radiated fields generated by them destructively interfere, leading to suppression in emission and apolariton of infinite lifetime(23). When their distance is (n+0.5)λ(the anti-node condition), the radiated fields constructively interfere leading to enhanced emission rate and short exciton lifetime. The cooperative Lamb shift can be understood through the dipole-dipole interaction: the emitted field from the image dipole at the original dipole has a phase shift of exp(i(π/2+2kL)), where the π/2 comes from the phase relation between dipole and its radiation and 2kL is the phase accumulation during the propagation. Due to the dipole interaction H=-μ·E, the energy is lower if L=[n/2, (n/2+0.25)]λ and the energy is higher if L=[(n/2+0.25), (n/2+0.5)]λ leading the shift of exciton energy. Observation of the cooperative Lamb shift and corresponding modulation of the radiative decay rate requires a homogeneous ensemble of emitters with nearly radiative-limited line 4 broadening, which is challenging to realize in conventional semiconductors. In this work, using a monolayer TMDC, we observe the renormalization of both exciton energy E0 and linewidth broadening ℏγ due to coupling with a radiation vacuum. To measure the renormalization of the exciton mode, we change the distance L between the monolayer and the mirror to modulate the local vacuum fluctuation at the 2D exciton position. We place an hBN-encapsulated MoSe2 monolayer on a sapphire substrate in front of DBR mirror, whose position is controlled by a piezo-electric stage as illustrated in Fig. 1A. This setup allows in situ spectroscopy measurement of the same piece of MoSe2 while we tune the local vacuum field. First, we measure the reflection contrast spectra of the monolayer as it is moved through the standing wave of the electric field profile, using a weak femtosecond laser with bandwidth covering the exciton absorption peak. As shown in Figure2A, the absorption dip of the MoSe2A-exciton at1660meV is strongly modulated, following the period of the standing wave profile. Figure 2B shows several reflection contrast spectra (horizontal linecuts of Fig. 2A). The absorption dips are fit very well by Lorentzian functions(gray dashed curves), indicating minimal inhomogeneous broadening in the sample. The absorption depth is tuned from as low as 4% to as high as 99% at node and anti-node positions of the field, respectively. In the node region, we can hide the monolayer from the classical probe field even though it sits fully exposed in an open space. In the anti-node region, we can enhance the absorption to achieve the critical coupling condition where all photon energy is dumped into exciton energy, giving nearly 100% absorption. We summarize in Fig. 2C the modulation of the absorption depth, linewidth and the resonance energy of the exciton as a function of mirror distance. We use the absorption depth to determine whether the monolayer is located at a node or anti-node(vertical gray dashed lines in Fig. 2C indicates the anti-nodal positions).In excellent agreement with predictions by Eq. (3), both the linewidth and energy of the exciton resonance show periodic modulations with a 𝜋/2 phase shift relative to each other and a modulation amplitude different by about a factor of 2. The linewidth of the exciton changes by 2 meV, from 5.5meV at an anti-node to 3.5meV at a node (middle panel), corresponding to twice the non-renormalized radiative linewidth ℏγ. Fitting the linewidth modulation with Eq. (3a)after including a constant offset𝛾′(blue dashed curve), we obtain ℏ𝛾=1.15±0.11 𝑚𝑒𝑉 and ℏ𝛾′=3.51±0.16𝑚𝑒𝑉. The offset of ℏ𝛾′accounts for contributions from other broadening mechanisms, including inhomogeneous, non-radiative and pure dephasing broadening. Theℏγ agrees with the radiative linewidth measured from linear reflection(20, 21) and 2D spectroscopy(26, 27). Similar linewidth modulations due to modified dipole-vacuum coupling has been observed in atomic systems(13, 14) and superconducting qubits(7, 19) in absorption spectra, but has not been demonstrated in a solid state system due to the small radiative linewidth compared with the total linewidth in typical semiconductor materials. Cooperative Lamb shift of the exciton resonance due to exciton-vacuum coupling (lower panel of Fig. 2C) follows the same period of the linewidth modulation but 𝜋/2displaced in phase. The shift is zero at both the nodes and anti-nodes of the field, as predicted by Eq. (3b). Fitting the modulation of the exciton energy with Eq. (3b), we obtain the magnitude of the modulation ℏ𝛾=1.14±0.18𝑚𝑒𝑉, consistent with the result from the linewidth modulation. 5 Such cooperative Lamb shift has been predicted theoretically(5), but demonstrated only in atomic and superconducting qubit systems recently(9,13,18,19).Note that the observed linewidth and cooperative Lamb shift modification is on the order of 1meV(~250THz), which is much larger compared with atomic and superconducting qubits systems(tens of MHz or smaller),because of the extraordinary oscillator strength of the collectively coupled excitons. Compared to absorption, incoherent photoluminescence (PL)stems from the coupling of incoherent exciton polarization to vacuum fluctuations, which cannot be described semi-classically. It is, therefore, interesting to test if the same renormalization effect appears in the emission spectrum of the exciton. Figure 3A shows a few examples of the PL spectra with varying L under the excitation of a continuous wave laser at 532nm. The features at 1662meV and 1632meV are due to exciton and trion emission, respectively. The linewidth of exciton emission measured at anti-nodes is clearly narrower than that measured at nodes(figure 3B), indicating that the same radiative decay rate renormalization is present in PL spectra. We summarize in Fig 3Cmodulation of the PL intensity, linewidth and resonance energy of the exciton peak. Reflection spectra are taken together to identify the node positions for exciton wavelength(750nm), which are labeled as vertical gray dashed lines in figure 3C. The PL intensity is complicated by the factor that the absorption of the 532nm pump laser is also modulated by the monolayer-mirror distance L. As a result, we observe PL intensity minima when the monolayer is located at the nodes of the field of either 532nm or 750nm, indicated by green arrows and gray dashed lines, respectively. The suppression of PL at the nodes of 750nm light arises from the optical interaction between the monolayer and its image dipole. To estimate the suppression/enhancement effect, we take the ratio of PL intensity at a node(step 16) versus an anti-node(step27) and compare the ratio to when no mirror is present. (Both points are around anti-node of the 532nm light.)In our measurements, the ratio can be as low as 5% and as high as 175%, showing control of PL quantum efficiency through radiative decay rate modulation. The renormalization effect on the exciton decay rate can be observe in the frequency domain more directly. The third panel in figure 3B show the PL linewidth as a function of monolayer-mirror distance. The PL linewidth follows only the mode profile of the 750nm standing wave and not that of the 532nm excitation laser. It changes from about 4.5meV at the anti-nodes to about 2.2meV at the nodes. Correspondingly, a cooperative Lamb shift of 1.1meV is measured in PL(lower panel of Fig. 3B). Both results are in agreement with absorption measurements. Therefore we demonstrate the same exciton renormalization due to vacuum field in the emission domain, which is difficult to observe in atomic and superconducting qubit systems. The dramatic effects of the vacuum on the TMDC excitons observed above demonstrate the possibility to control optical properties of TMDCs by vacuum engineering. We give two other examples, where we use the same tunable mirror approach to tune the charged exciton and valley polarization via the vacuum-matter coupling. Charged exciton, or trions, are pronounced in TMDCs due to the strong Coulomb interaction in a two-dimensional film. Since the exciton and trion wavelengths are well separated, we can selectively enhance either the exciton or the trion emission by tuning the vacuum field strength at the respective wavelength. Figure 4Ashows two PL spectra from the 6 same MoSe2 monolayer, but at different mirror locations. The emission spectra show very different exciton/trion intensity ratio despite measured from the same position on the monolayer and at a fixed charge doping. This result shows that the ability to shape the vacuum-matter coupling allows us to control the interference effect for both exciton and trion emission. We show in Figure 4B the mirror-monolayer distance dependence of the PL intensities of exciton, trion and their ratio. The suppression of exciton emission is observed at the node for 750nm light indicated by the dashed gray lines while the suppression of trion emission is observed at the node for 780nm light indicated by the orange arrows. The ratio of exciton and trion emission intensity changes over two orders of magnitude, from 0.02 to 2.48. This simple technique can be utilized to various 2D materials applications to enhance or suppress the transition of interest. Another important property of TMDC materials is the valley degree of freedom. However, the valley polarization of the excitons is often complicated by several competing decay and exchange mechanisms. Considering the two dominant mechanisms -- the radiative decay and inter-valley exchange interaction -- we can formulate the valley polarization, quantified by the degree of circular polarization (DOCP) as DOCP=I+−I−I++I−=γγ+2η (4) where γ and η are the radiative decay rate and intervalley exchange rate, respectively(28). For this study, we use an encapsulated WSe2 sample and 633nm laser with σ+ polarization for excitation. The exciton PL spectra around 1.74eV with anti-node, node and no mirror condition are shown in Fig. 4C. The measured DOCP of exciton as a function of mirror distance(Fig. 4D) shows oscillation behavior in accord with the reflection contrast modulation. The tuning of vacuum fluctuation through mirror distance allows us to change the radiative decay rate γ of the WSe2 exciton, resulting in enhancement (suppression) of DOCP at anti-nodes (nodes). Given that the measured DOCP without mirror present is 37%(blue dashed line in Fig. 4D), the modulation should be from 0% to 54% based on Eq. (4) when the radiative decay rate is tuned from 0 to 2γ. However, we observe a modulation from 25% to 43% experimentally. The discrepancy might come from other depolarization mechanisms or effective radiative lifetime from dark excitons in the WSe2 monolayer and require further studies. Nonetheless, the modulation via simply a mirror is already comparable with other reports using microcavities(28 -- 30). In conclusion, we observe the renormalization of exciton resonance energy and radiative linewidth in TMDC monolayers due to coupling to vacuum fields. This effect has only been observed in atomic and superconducting qubit systems before. The tightly bound exciton in TMDC monolayers leads to an exceptionally large radiative linewidth, enabling pronounced effects of vacuum engineering, manifested as a large cooperative Lamb shift, strong modulation of the radiative linewidth, and control over trion and valley degrees of freedom. Our study shows intriguing collective physics of vacuum effects on excitonic many-body systems and could pave the way for future quantum optics research with 2D materials. Note: Upon the completion of this manuscript, we became aware of preprints of related works by You Zhou, et. al.(31) and Christopher Rogers, et. al.(32). 7 References and Notes: 1. W. E. Lamb, R. C. Retherford, Fine Structure of the Hydrogen Atom by a Microwave Method. Phys. Rev. 72, 241 -- 243 (1947). 2. Dirac Paul Adrien Maurice, Bohr Niels Henrik David, The quantum theory of the emission and absorption of radiation. Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character. 114, 243 -- 265 (1927). 3. T. Fujisawa et al., Spontaneous Emission Spectrum in Double Quantum Dot Devices. Science. 282, 932 (1998). 4. J. Eschner, C. Raab, F. Schmidt-Kaler, R. Blatt, Light interference from single atoms and their mirror images. Nature. 413, 495 (2001). 5. R. Friedberg, S. R. Hartmann, J. T. Manassah, Frequency shifts in emission and absorption by resonant systems ot two-level atoms. Physics Reports. 7, 101 -- 179 (1973). 6. F. \mboxç\else ç\fiois Dubin et al., Photon Correlation versus Interference of Single-Atom Fluorescence in a Half-Cavity. Phys. Rev. Lett. 98, 183003 (2007). 7. I.-C. Hoi et al., Probing the quantum vacuum with an artificial atom in front of a mirror. Nature Physics. 11, 1045 (2015). 8. Kazuki Koshino and Yasunobu Nakamura, Control of the radiative level shift and linewidth of a superconducting artificial atom through a variable boundary condition. New Journal of Physics. 14, 043005 (2012). 9. J. Keaveney et al., Cooperative Lamb Shift in an Atomic Vapor Layer of Nanometer Thickness. Phys. Rev. Lett. 108, 173601 (2012). 10. U. Dorner, P. Zoller, Laser-driven atoms in half-cavities. Phys. Rev. A. 66, 023816 (2002). 11. M. Gross, S. Haroche, Superradiance: An essay on the theory of collective spontaneous emission. Physics Reports. 93, 301 -- 396 (1982). 12. R. J. Bettles, S. A. Gardiner, C. S. Adams, Enhanced Optical Cross Section via Collective Coupling of Atomic Dipoles in a 2D Array. Phys. Rev. Lett. 116, 103602 (2016). 13. L. Corman et al., Transmission of near-resonant light through a dense slab of cold atoms. Phys. Rev. A. 96, 053629 (2017). 14. J. Javanainen, J. Ruostekoski, Y. Li, S.-M. Yoo, Shifts of a Resonance Line in a Dense Atomic Sample. Phys. Rev. Lett. 112, 113603 (2014). 15. E. Shahmoon, D. S. Wild, M. D. Lukin, S. F. Yelin, Cooperative Resonances in Light Scattering from Two-Dimensional Atomic Arrays. Phys. Rev. Lett. 118, 113601 (2017). 8 16. Z. Meir, O. Schwartz, E. Shahmoon, D. Oron, R. Ozeri, Cooperative Lamb Shift in a Mesoscopic Atomic Array. Phys. Rev. Lett. 113, 193002 (2014). 17. T. Peyrot et al., Collective Lamb Shift of a Nanoscale Atomic Vapor Layer within a Sapphire Cavity. Phys. Rev. Lett. 120, 243401 (2018). 18. R. Röhlsberger, K. Schlage, B. Sahoo, S. Couet, R. Rüffer, Collective Lamb Shift in Single-Photon Superradiance. Science. 328, 1248 (2010). 19. A. F. van Loo et al., Photon-Mediated Interactions Between Distant Artificial Atoms. Science. 342, 1494 (2013). 20. P. Back, S. Zeytinoglu, A. Ijaz, M. Kroner, A. Imamoıfmmode \breveg\else ğ\filu, Realization of an Electrically Tunable Narrow-Bandwidth Atomically Thin Mirror Using Monolayer $\mathrmMoSe_2$. Phys. Rev. Lett. 120, 037401 (2018). 21. G. Scuri et al., Large Excitonic Reflectivity of Monolayer $\mathrmMoSe_2$ Encapsulated in Hexagonal Boron Nitride. Phys. Rev. Lett. 120, 037402 (2018). 22. S. Zeytinoıfmmode \checkg\else ǧ\filu, C. Roth, S. Huber, A. ıfmmode \dotI\else \.I\fimamoıfmmode \breveg\else ğ\filu, Atomically thin semiconductors as nonlinear mirrors. Phys. Rev. A. 96, 031801 (2017). 23. D. S. Wild, E. Shahmoon, S. F. Yelin, M. D. Lukin, Quantum Nonlinear Optics in Atomically Thin Materials. Phys. Rev. Lett. 121, 123606 (2018). 24. C. Riek et al., Direct sampling of electric-field vacuum fluctuations. Science. 350, 420 (2015). 25. J. Kim, D. Yang, S. Oh, K. An, Coherent single-atom superradiance. Science. 359, 662 (2018). 26. T. Jakubczyk et al., Radiatively Limited Dephasing and Exciton Dynamics in MoSe2 Monolayers Revealed with Four-Wave Mixing Microscopy. Nano Lett. 16, 5333 -- 5339 (2016). 27. G. Moody et al., Intrinsic homogeneous linewidth and broadening mechanisms of excitons in monolayer transition metal dichalcogenides. Nature Communications. 6, 8315 (2015). 28. Y.-J. Chen, J. D. Cain, T. K. Stanev, V. P. Dravid, N. P. Stern, Valley-polarized exciton -- polaritons in a monolayer semiconductor. Nature Photonics. 11, 431 (2017). 29. S. Dufferwiel et al., Valley-addressable polaritons in atomically thin semiconductors. Nature Photonics. 11, 497 (2017). 30. S. Dufferwiel et al., Valley coherent exciton-polaritons in a monolayer semiconductor. Nature Communications. 9, 4797 (2018). 9 31. You Zhou et al., Controlling excitons in an atomically thin membrane with a mirror. arXiv:1901.08500 (2019). 32. Christopher Rogers et al., Coherent Control of Two-Dimensional Excitons. arXiv:1902.05036 (2019). Acknowledgments: We gratefully thank Professor Mack Kira for fruitful discussions. Funding: This work was supported by the Multidisciplinary University Research Initiative(MURI) program by the Army Research Office(ARO) under Grant # W911NF-17-1-0312. Y.-H. C. acknowledges the support from Graduate Student Study Abroad Program(GSSAP) from Ministry of Science and Technology in Taiwan. This work was partially supported by the Minister of Science and Technology (Taiwan) under contracts Nos. 106-2917-I-564-021. Author contributions: J.H. and H.D. conceived the research. J.H. designed the experiments, carried out optical measurements, analyzed the data and performed theoretical analysis. T.-C. C., C.-Y. H. and T.-C. L. fabricated DBR mirrors. Y.-H. C. fabricated the TMDC samples assisted by J.H.. J.H. and H.D. wrote the manuscript with inputs from all authors. Competing interests: Authors declare no competing interests. Data and materials availability: All data is available in the main text or the supplementary materials. 10 Figures Fig. 1. Modifying vacuum fluctuations at a two-dimensional semiconductor in front of a mirror. (A) A monolayer MoSe2 is placed in front of a mirror with a tunable distance L. Depending on the mirror distance L, the monolayer samples different vacuum fluctuation due to the standing wave imposed by the mirror boundary condition. Altering the vacuum-monolayer coupling leads to renormalization of exciton resonance energy and radiative decay rate. (B) Another approach to understand the system is to consider the exciton in the MoSe2 monolayer interacting with its mirror image through dipole-dipole interactions. Due to the macroscopic dipole moment from two-dimensional excitons, the renormalization effect can be significant. 11 Fig. 2. Effects of vacuum fluctuations on the exciton transition measured via absorption spectroscopy. (A) Measured reflection contrast of a MoSe2 monolayer in front of a distributed-Bragg-reflector as a function of photon energy and monolayer-mirror distance L. The absorption dip around 1660meV corresponds to the A-exciton resonance. (B) Several spectra from (A) showing the shift and broadening of the exciton absorption when the monolayer is moved from anode (step 7) to an anti-node(step 17) of the field. (C) Mirror-position dependence of the depth (top panel), linewidth (middle panel) and resonance energy (bottom panel) of the A-exciton absorption dip. The anti-node positions are identified by the maximum absorption depth (~99%), while the node positions are identified by the minimum absorption depth (~0.04%) and marked by the dashed lines. The modulation of the vacuum fluctuations lead to modulations of both the linewidth and the cooperative Lamb shift, which are fit but sinusoidal functions with a 𝜋/2 relative phase shift (blue and red dashed curves, respectively). 12 Fig. 3. Effects of vacuum fluctuations on the photoluminescence(PL) properties of a MoSe2 monolayer. (A) Measured PL spectra of a MoSe2 monolayer in front of a mirror as the monolayer is moved from an anti-node (black line) to a node (blue line) of the modified vacuum field. The emission peaks around 1660 and 1630 meV correspond to the A-exciton and trion resonances, respectively. (B) Normalized PL spectra at an anti-node and a node, showing different linewidths. (C) Mirror-position dependence of the intensity, linewidth and resonance energy of the A-exciton PL, showing modulations following the modified vacuum fluctuations. The PL resonance energy also shows the cooperative Lamb shift. The vertical dashed lines mark the nodes of the vacuum field identified from absorption spectra. The green arrows indicate where the absorption of the 532 nm excitation laser is suppressed. 13 Fig. 4. Controlling the emission properties of 2D materials via vacuum fluctuations. (A) Two MoSe2 emission spectra measured at the same position on the monolayer with a fixed doping. (B)The exciton and trion emission intensities (top) and their ratio (bottom) as a function of the monolayer-mirror distance, showing enhancement and suppression of the exciton relative to the trion emission with varying distances. (C) Helicity-resolved PL spectra of monolayer WSe2 at the field anti-node (left) and node (right) in front of a mirror, and without a mirror (middle). (D) Degree of circular polarization(DOCP) vs. the mirror position. It changes from 25% to 40%, showing the effect of vacuum coupling on the valley dynamics of TMDCs. The blue dashed line indicate the DOCP when mirror is no present. 

1511.05247

1

1511

Do micromagnets expose spin qubits to charge and Johnson noise?

An ideal quantum dot spin qubit architecture requires a local magnetic field for one-qubit rotations. Such an inhomogeneous magnetic field, which could be implemented via a micromagnet, couples the qubit subspace with background charge fluctuations causing dephasing of spin qubits. In addition, a micromagnet generates magnetic field evanescent-wave Johnson noise. We derive an effective Hamiltonian for the combined effect of a slanting magnetic field and charge noise on a single-spin qubit and estimate the free induction decay dephasing times T2* for Si and GaAs. The effect of the micromagnet on Si qubits is comparable in size to that of spin-orbit coupling at an applied field of B=1T, whilst dephasing in GaAs is expected to be dominated by spin-orbit coupling. Tailoring the magnetic field gradient can efficiently reduce T2* in Si. In contrast, the Johnson noise generated by a micromagnet will only be important for highly coherent spin qubits.

cond-mat.mes-hall

cond-mat

a Do micromagnets expose spin qubits to charge and Johnson noise? Allen Kha,1 Robert Joynt,2, 1 and Dimitrie Culcer1 1)School of Physics, The University of New South Wales, Sydney NSW 2052, Australia 2)Department of Physics, University of Wisconsin, Madison, Wisconsin 53706, USA (Dated: 20 January 2020) An ideal quantum dot spin qubit architecture requires a local magnetic field for one-qubit rotations. Such an inhomogeneous magnetic field, which could be implemented via a micromagnet, couples the qubit subspace with background charge fluctuations causing dephasing of spin qubits. In addition, a micromagnet generates magnetic field evanescent-wave Johnson noise. We derive an effective Hamiltonian for the combined effect of a slanting magnetic field and charge noise on a single-spin qubit and estimate the free induction decay dephasing times T ∗ 2 for Si and GaAs. The effect of the micromagnet on Si qubits is comparable in size to that of spin-orbit coupling at an applied field of B = 1T, whilst dephasing in GaAs is expected to be dominated by spin-orbit coupling. Tailoring the magnetic field gradient can efficiently reduce T ∗ 2 in Si. In contrast, the Johnson noise generated by a micromagnet will only be important for highly coherent spin qubits. Solid state spin systems, in which coherence times of up to a few seconds have been measured,1 -- 3 are promising candidates for scalable quantum computer architectures,4,5with silicon ideal for hosting spin-based qubits.6 -- 11 Addressing individual qubits is vital, yet us- ing electron spin resonance requires bulky on-chip coils that dissipate heat close to the electron.12,13 However, electric fields can be generated locally with small, low voltage electrodes, and electrical spin rotations14,15 can be accomplished by the modulation of a quantum dot electric field in a slanting static magnetic field.15 -- 18 Quantum dot spin qubits are typically located near semiconductor interfaces where defects are present.19 -- 23 The resulting fluctuations in the local electric field are a well known source of dephasing in charge qubits,24 -- 29 as well as relaxation30 and dephasing of spin qubits31 when spin-orbit coupling is present. Here we show that the micromagnet couples spin qubits to charge fluctuations and causes dephasing even when spin-orbit coupling is absent. In addition, a ferromagnet contains currents and spins that fluctuate due to both thermal and quantum effects. This generates random magnetic fields nearby. Thus a micromagnet in the vicinity of spin qubits can cause spin dephasing and relaxation. This effect is sim- ilar to the relaxation caused by evanescent-wave John- son noise32 recently observed in NV centers in diamond close to metallic surfaces.33 In the case of a micromag- net, however, we must consider the dissipative magnetic, not electrical, response of the noise source. An analysis of this kind has recently been done in Ref. 34 for one type of micromagnet design. Although we treat here a different design, our analytical results are in qualitative agreement with the numerical results of Ref. 34. In this paper we study two effects: qubit dephasing in the presence of (A) the combined effects of charge noise and an inhomogeneous magnetic field, and (B) Johnson-type magnetic field noise generated by a micro- magnet. We model the first effect as random telegraph noise (RTN) and 1/f noise35 -- 40 together with an inhomo- geneous magnetic field. We compare dephasing times T ∗ 2 for identical dot designs in Si and GaAs. For the second effect we derive the appropriate formulas that govern the strength of the fluctuating magnetic fields in the vicinity of the micromagnet and their effect on the qubit. For the parameters appropriate to a representative device archi- tecture this effect is not large. It will be important as coherence times approach the range of 1 − 10 s. Charge noise combined with an inhomogeneous mag- netic field. We focus on a single-spin qubit implemented in a symmetric gate defined quantum dot, located at the flat interface of a semiconductor heterostructure (Si/SiGe or GaAs/AlGaAs). The two-dimensional electron gas (2DEG) lies in the xy-plane. The gate confinement is assumed harmonic V (x, y) = 1 2 m∗ω2(x2 + y2), where m∗ is the in-plane effective mass, ω =  m∗a2 the oscillator frequency and a the dot radius. The Hamiltonian for the electron kinetic energy and confinement in the xy-plane H0 = − 2 2m∗(cid:18) ∂2 ∂x2 + ∂2 ∂y2(cid:19) + 2 2m∗a4 (x2 + y2). (1) For the ground state D0(x, y) = (1/a√π) e− x2+y2 energy ε0 = 2 cited state D±(x, y) = (1/a2√π) (x ± iy)e− x2+y2 2a2 with 2m∗a2 . For the twofold degenerate first ex- and ε1 = 3ε0. We initially model charge noise by a single charge trap located in the xy-plane at a distance x = xd from the dot center. The noise Hamiltonian HN (t) is a random function of time and represents a fluctuat- ing Coulomb interaction between the electron in the dot In the absence of screening the time- and one at xd. , independent Coulomb potential VC (x, y) = 4πε0εrrd where ε0 is the permittivity of free space, εr the dielectric 2a2 e2 constant and rd =p(x − xd)2 + y2 the distance between the defect and the dot center. The non-zero matrix ele- ments in HN are v0 = hD0VCD0i, v1 = hD±VCD±i, α = hD0VCD±i and β = hD±VCD∓i. Electrons in the 2DEG screen the defect potential.41 We use the same notation to denote the matrix elements v0, v1, α, β but replace VC → Uscr = R d2q/(2π)2e−iq·rUscr(q), where M z B0 Sx bsl Substrate Semiconductor x FIG. 1. Sketch of the implementation of two ferromagnetic strips with uniform magnetisation M due to the external mag- netic field B0 in the x -direction. The resulting out-of-plane z-component of the field from the strips has a slanting form with gradient bsl. The spin of the electron is initialised k x. Uscr(q) is the Fourier transform of the potential42 and we neglect contributions from momenta q ≥ 2kF . lable homogeneous part B0 and a slanting field,15 -- 17 We take a total magnetic field composed of a control- B = (B0 + bslz)x + (bslx)z, (2) where the magnetic field gradient takes the value15,16 bsl = 0.8 Tµm−1. The magnetic field gradient is cre- ated by two ferromagnetic strips integrated on top of the QD, magnetized uniformly by the in-plane magnetic field B0. This structure results in a stray magnetic field with an out-of-plane component that varies linearly with po- sition with a gradient bsl as shown in Figure 1. The Zeeman Hamiltonian HZ = 1 2 gµB σ · B, where σ is the vector of Pauli spin matrices. We have six basis states {D0,↑, D0,↓, D+,↑, D+,↓, D−,↑, D−,↓}, with ↑i, ↓i the σx eigenstates corresponding to eigenvalues 1, −1 respec- tively. The qubit subspace is {D0,↑, D0,↓}. The total Hamiltonian H = H0 + HN + HZ is given in the Supplement.41 The magnetic field couples the qubit subspace to spin anti-aligned orbital excited states through the gradient bsl. We project H onto the qubit subspace by eliminating orbital excited states.43 The re- sulting effective spin Hamiltonian Hef f = HQbt + 1 σ · 2 V(t) expresses the perturbations of noise and the mag- netic field as an effective fluctuating magnetic field in the qubit subspace (for which we use σ) Hef f =(cid:18)ε+ 0 0 0 ε− 0(cid:19) + 8η2δvεZ (δε)3 σz − 4αηδv (δε)2 σx (3) 2 gµBhD0bslxD+i = 1 where η = 1 2 gµBhD−bslxD0i, δε = ε0 − ε1, δv(t) = v0(t) − v1(t) and εZ = hD0,↑HZD0,↑i. The first term corresponds to a pseu- dospin 1/2 in an effective magnetic field assumed to be controllable, while the second arises from noise. The term ∝ σx is behind spin relaxation30 and Rabi oscillations,44 but its contribution to dephasing is very small as long as 2 it is much smaller than δε, which is true for this work. Dephasing here arises from the σz term. In Si particular attention must be paid to the val- ley splitting, which is the magnitude of the valley-orbit coupling ∆v.45 In this work we focus on the case δε > ∆v ≫ εZ, though we also expect our results to hold when ∆v ≪ εZ ≪ δε (the critical condition is δε > ∆v). The inhomogeneous magnetic field will give a small matrix element coupling states from different valleys. For an in- teraction to couple valley states significantly, it must be sharp in real space31. The z-dependence of the magnetic field is not sharp enough to couple valleys, and the ma- trix element is further diminished by the small value of the Bohr magneton. The small intervalley matrix ele- ments only matter when ∆v ≈ εZ , otherwise εZ does not In this case it is known that affect the valley physics. a relaxation hotspot (a peak in T1) exists,46 which sug- gests that T1 limits T ∗ 2 , and the valley degree of freedom must be taken into account explicitly, yet by adjusting the magnetic field one can tune away from this point. The formal treatment of dephasing is summarised in the Supplement.41 We divide the noise spectrum into two parts: (i) random telegraph fluctuators which are close to the qubit and whose effect may be resolved individu- ally in a noise measurement and (ii) a background 1/f spectrum. We first focus on a single nearby source of RTN. To facilitate comparison with the spin-orbit cou- pling case, we consider fluctuators with shorter switching times than a cut-off of τ = 1 µs. In this case V 2 ≪ (  τ )2 and the initial spin decays as ∝ e−t/T ∗ V 2τ 22 , 2 , with (4) = 2 (cid:19)RT N (cid:18) 1 T ∗ where for the slanting magnetic field V = 8η2δv(t)εZ the background 1/f noise we find41 δε3 . For 2 (cid:19)1/f ≈r γN kBT (cid:18) 1 22 (cid:18) 8η2εZ δε3 (cid:19) , T ∗ (5) where γN (units of energy) is derived from experiment. We consider sample QDs in Si/SiGe and GaAs struc- tures and set the fluctuator switching time τ = 1 µs, defect position xd = 40 nm, Fermi wave vector kF = 108 m−1 and Zeeman energy εZ = 60 µeV. This does not affect η as it is ∝ bsl and not ∝ B0.47 We first cal- culate T ∗ 2 for a fixed quantum dot confinement energy δε = 0.5meV in the two materials, and then for fixed dot radius a = 20 nm. For Si g = 2, m∗ = 0.2me and εr = 12.5 (Si/SiGe),48,49 for GaAs g = 0.41, m∗ = 0.067me and εr = 12.9 (GaAs/AlGaAs),50 where me is the bare electron mass. For 1/f noise we assume S(ω) scales lin- early with temperature51,52 so we extract the factor ∆ = γN kBT , representing the strength of the noise from Refs. 53 and 28 for Si/SiGe and GaAs respectively and scale it to T = 100 mK, which gives ∆Si/SiGe = 8.85× 10−7 meV2 and ∆GaAs = 3.79 × 10−3 meV2. The results are presented in Tables I and II, which are the main results of this paper, and since Eqs. S5 and S7 only give estimates for T ∗ 2 we plot the time evolution of the spin Sx(t) in Figures 2 and 3. We also compare dephasing times with those due to spin-orbit coupling as calculated in Ref. 31. In Si/SiGe the dephasing times for spin-orbit and the slanting magnetic field are essen- tially the same, while in GaAs spin-orbit has a far more destructive effect on qubit coherence compared to the magnetic field. Although the numbers in Ref. 31 are estimates, the noise profile assumed was the same as in this work. The much weaker effect on GaAs is due to the smaller g-factor. The Overhauser field in GaAs QDs is several orders of magnitude larger than in Si QDs,54 and relevant energy scales have the same order of magni- tude as for spin-orbit interactions,31 so without consider- ing feedback mechanisms55 or echo sequences56 designed to increase dephasing times, the contribution from the Overhauser field is the same as due to spin-orbit. The contribution of the nuclear spin bath to qubit decoher- ence for a Si QD can be drastically reduced by using isotopically purified 28Si samples.54,57 We also find that T ∗ 2 is heavily dependent on the 2 )1/f ∝ δε4 and QD radius and confinement energy. (T ∗ 2 )RT N ∝ δε6, so by doubling the confinement energy, (T ∗ or equivalently reducing the dot radius by a factor of √2, the dephasing time can be increased by an order of mag- nitude. Dephasing times can also be increased by reduc- ing the noise spectrum by going to lower temperatures since S(ω) ∝ T , or reducing sources of charge noise. The latter has recently been achieved by developing quantum dots in an undoped Si/SiGe wafer,58 with results indi- cating that doped materials produce more charge noise sources via the 2DEG and interface trapping sites. More- over, the Rabi oscillation term is ∝ η. The dephasing rate for RTN is ∝ η4, while that due to 1/f noise is approxi- mately ∝ η2. Reducing η therefore reduces the 1/f noise dephasing rate faster than the Rabi oscillation gate time (considerably faster for RTN). Inhomogeneous magnetic fields are also an essential in- gredient of singlet-triplet qubits.59 -- 63 There, however, de- phasing will be dominated by direct coupling to charge noise via exchange.38 The slanting magnetic field yields a term ∝ σx, which gives relaxation, but not dephasing. Evanescent-wave Johnson noise from a micromagnet. For an electron spin qubit in a fluctuating magnetic field 1 T1 = µ2 B 2 Xi=x,y hBi (r) Bi (r)iω0 , where r is the position of the qubit and ω0 is the Zeeman splitting caused by the applied field B0 taken here to be k z. Only the transverse components of B contribute If a micromagnet is placed at the origin of the to T1. coordinates, then, with r = (x1, x2, x3),41 hBiBjiω0 = V Im α (ω0) coth(cid:18) ω0 2kBT(cid:19) 3xixj + δijr2 r8 (6) Considering only finite frequencies, the magnetic polar- 3 TABLE I. Sample T ∗ 2 for a quantum dot with a = 20 nm, defect position xd = 40 nm, τ = 1 ms for RTN, T = 100 mK, εZ ≈ 60 µeV, S(ω) for 1/f noise estimated from Refs. 28 and 53, and εr = 12.5, 12.9 for Si/SiGe, GaAs respectively. For spin-orbit we used Rashba coefficients from Ref. 31. δε RTN + B RTN+SO 1/f + B 1/f + SO Si/SiGe 1 meV 30 ms GaAs 3 meV 1900 s 1 ms 25 ns 130 µs 7 s 30 µs 25 µs a RTN + B RTN+SO 1/f + B 1/f + SO Si/SiGe 30 nm 520 µs GaAs 50 nm 40 ms GaAs† 50 nm 610 µs 310 µs 500 ps 500 ps 10 µs 7 ms 840 µs 7 µs 770 ns 770 ns TABLE II. Sample T ∗ 2 for a quantum dot with confinement energy δε = 0.5 meV, and all parameters except a, xd identical to Table I; we fix xd/a = 2. †We set η to be constant, so for GaAs, bsl is ∼ 3 times larger than for Si to account for the smaller g-factor (not 5 times larger as η ∝ a also.) izability α (ω) is defined by m (ω) = V α (ω) h (ω), where m (ω) is the total magnetic moment of the micromag- net particle, V is the volume of the magnet, and h is a spatially uniform applied magnetic field. We need to estimate the dissipative (imaginary) part of α. In general α is a matrix, but we are mainly in- terested in order-of-magnitude estimates, so we ignore anisotropies and factors of order unity. This is justified below. The dynamics of the magnet are dominated by ferromagnetic resonance. In principle, this is dangerous for qubit decoherence in Si (g ≈ 2), since the fundamen- tal resonance frequency gµBB0/ for the precessions of the qubit spin and the bare resonance frequency ω0 of the Co magnet (g-factor also very close to 2) are close. Let M0 be the permanent magnetization and let this be k B0, since this is the geometry chosen in all experi- ments to date. Following the established theory of ferro- magnetic resonance,64 we decompose the magnetization and field into a large time-independent term and a small time-dependent term. Thus we have M (t) = M0 + m (t) and H = B0 + h (t). The equation of motion neglecting spin-orbit coupling and shape anisotropy is: dM dt = dm dt ≈ −γ (M0 × h + m × B0) . The first term is the change in the macroscopic M that is produced by h, and the second term is the precession of m in the external field. Changing to the frequency domain at a fixed driving frequency ω leads to −iωm = γ(M0h − B0m) × z = (ωM h − ω0m) × z where ω is the driving frequency: h (t) = he−iωt, ω0 = γB0 is the "bare" Larmor frequency associated with the DC applied field B0 and ωM is the frequency as- sociated with M0 : ωM = γM0; γ = gµB/. Solv- ing these equations we find m = χh + iGh × z, with FIG. 2. Time evolution of the spin in Si/SiGe and GaAs with dot radius a = 20nm, cut-off ω0 = 1s and all other dot parameters identical. Sx
t

Sx0 1.0 0.8 0.6 0.4 0.2 Sx
t

Sx0 1.0 0.8 0.6 0.4 0.2 10 20 30 40 50 60 70 80 90 100 t
Μs
RTN
Si
1
f
Si
RTN
GaAs
1
f
GaAs
1 2 3 4 5 6 7 8 9 10 t
Μs
RTN
Si
1
f
Si
RTN
GaAs
1
f
GaAs
FIG. 3. Time evolution of the spin in Si/SiGe and GaAs with fixed confinement energy δε = 0.5meV, cut-off ω0 = 1s and xd/a = 2, with all other dot parameters identical. 0 − ω2(cid:1). χ (ω) = ωM ω0/(cid:0)ω2 0 − ω2(cid:1) and G (ω) = ωM ω/(cid:0)ω2 Here χ acts as an ordinary (diagonal) susceptibility while the G term gives the effect of precession about the mag- netization. We will not compute the effects of spin-orbit coupling and shape anisotropy in detail, since they de- pend on such unknowns as the microcrystallinity of the magnet. We only note that the resonance frequency ω0 of the magnet is strongly shifted by the restoring forces due to these effects. Hence the resonance frequencies of the magnet and qubit, though of the same order of magnitude, in general do not coincide, and ω0 should be replaced by the physical resonance frequency ωres 6= ω0. This physical picture is in good agreement with the noise spectra in Ref. 34, in which nearly all the weight is shifted upwards from the bare resonance frequency. So far all the response is real and there is no dissi- pation. The form of the damping term depends on the mechanism. However, at the phenomenological level of 4 the present treatment, a simple Bloch-Bloembergen form dM dt = −γM × H −(cid:18) M − M0 z T2 (cid:19), is sufficient, which leads to the final result Imχ (ω) = ωM ωres 2 2ωT −1 2 (cid:1)2 res − ω2 + T −2 (cid:0)ω2 , + 4T −4 2 and similarly for Im G. There are too many unknowns to attempt a very accurate estimate of χ and G but we may use values of parameters from bulk Co65 and Co wires66,67 to make an order-of-magnitude estimate. It is simplest to estimate the effect of the noise in the experiment of Ref. 17, in which the separation r from the magnet to the qubit satisfies r ≫ L, where L is the largest linear dimension of the magnet. This is not the case in the set-ups in Refs. 14 -- 16, and 18, which has larger magnets closer to the qubit. For an analysis of this type of device, see Ref. 34. Taking B0 = 20 mT from Ref. 17 as the applied field, we find ωM = 28 GHz using the formulas above and from experiment T2(magnet) ≈ 10−9 s.65 ωres has the same order of magnitude as ω0 = 4 GHz. Since all three quantities that enter Imχ are roughly of the same order, Imχ and Imα can be taken to be of or- der unity, and indeed α cannot much exceed unity since it is proportional to the fraction of energy absorbed from the time-dependent magnetic field h (t). Focusing on the design in Ref. 59 (somewhat different from Fig. 1), we substitute r = 1.8 µm and find T1 ≈ T ∗ 2 ≈ 10 s. (Since the Johnson noise is white, the longitudinal and trans- verse decoherence times do not usually differ by a factor of 2.) This value for T ∗ 2 , is far longer than the measured 2 ≈ 200 ns,17 implying that the micromagnet noise is T ∗ not contributing to decoherence in this experiment. Ref. 34 also found T1 & 10 s, which suggests that when r ∼ L there is some cancellation in the noise fields. In conclusion, we have studied the contribution to dephasing of an applied inhomogeneous magnetic field and charge noise on a single-spin qubit and found it is an effective source of qubit decoherence particularly for Si/SiGe devices. Our results imply that when implement- ing slanting magnetic field architectures for spin control, noise sources need to be considered and reduced to im- prove coherence times. By contrast, the Johnson noise from the micromagnet is probably not a significant source of decoherence in the current generation of experiments. It may become important as decoherence times become longer, of the order of seconds. We are grateful to L. Vandersypen for discussions, Andrea Morello and M. Pioro-Ladri`ere for bringing to our attention existing micromagnet designs, and to M. A. Eriksson for discussion of experimental devices. We would also like to acknowledge support from the Gordon Godfrey bequest. This research was partially supported by the US Army Research Office (W911NF-12-0607). Appendix A: Total Hamiltonian S0xe−t/T ∗ 2 , with 5 (S5) . The total Hamiltonian is H = H0+HN +HZ, which, in the basis {D0,↑, D0,↓, D+,↑, D+,↓, D−,↑, D−,↓}, is written as H = , (S2) ε+ 0 0 α η α η 0 ε− η α η α 0 α η ε+ 1 0 β 0 η α 0 ε− 1 0 β α η β 0 ε+ 1 0 η α 0 β 0 ε− 1     0 = ε0 + v0 ± εZ and ε± where ε± 1 = ε1 + v1 ± εZ are the Zeeman-split orbital levels including the charge noise and magnetic field contributions, and the mag- netic field matrix elements εZ = hD0,↑HZD0,↑i and η = 1 In the Schrieffer-Wolff transformation we keep first or- der terms in δv(t) and εz, and neglected terms propor- tional to the identity matrix and non-fluctuating terms (terms not involving δv(t)). 2 gµBhD0BzD+i = 1 2 gµBhD−BzD0i. Appendix B: Formal treatment of dephasing The qubit is described by a spin density matrix ρ(t) = 1 σ · S(t) which satisfies the quantum Liouville equation 2 dρ/dt + (i/)[H, ρ] = 0. The fluctuating z-component of the effective Hamiltonian causes dephasing, which for RTN is V(t) = V (t)σz(−1)N (t), where N (t) = 0, 1 is a Poisson random variable with fluctuator switching time τ . We work in a rotating reference frame which takes into account the effect of the laboratory effective magnetic field, assumed to be spatially homogeneous, in which the z-component of the spin projection is conserved so we are studying pure dephasing. To determine the full time evolution of the density matrix with initial conditions ρ(0) = 1 V (t′)dt′, with 2 h(t) = h(t). The time evolution of the ith spin compo- nent Si(t) = tr[σiρ(t)] is R t σ · S0, we define h(t) ≡ 1 0 Si(t) = S0i cos h(t) − ǫijkhk(t) sin h(t) + hi(t)[h(t) · S0][1 − cos h(t)]. (S3) If Sx(0) = S0x x then Sx(t) ≈ S0x cos h(t), and taking the average over the realisations of cos h(t),31,68 -- 70 hcos h(t)i = e−t/τ" sinh t Ξτ + cosh Ξt#, (S4) where Ξ = p(1/τ )2 − V 2/2. We consider fluctuators with switching times τ < 1 µs, in which the condition τ )2 is satisfied and we may expand Eq. (S4) in V 2 ≪ (  . The initial spin decays exponentially as Sx(t) = V 2τ 2 2 (cid:18) 1 2 (cid:19)RTN T ∗ = V 2τ 22 , where for the slanting magnetic field V = 8η2δv(t)εz δε3 For 1/f noise the main contribution is concentrated at low frequencies, so it primarily affects qubit dephas- ing in both semiconductor35 -- 38 and superconductor39,40 It is typically Gaussian in semiconductors51 devices. and can be described by the correlation function S(t) = hV (0)V (t)i. The Fourier transform of this is the noise spectral density which has the form S(ω) = γN kB T , where γN (units of energy) is obtained from experi- ment. Hence, for V(t) in the qubit subspace, we have S(ω). We write Sx(t) = S0xe−χ(t), ω δε3 (cid:17)2 SV (ω) ≈ (cid:16) 8η2εz where71 χ(t) = 2γN kBT 2 δε3 (cid:19)2 (cid:18) 8η2εz Z ∞ ω0 dω(cid:18) sin2 ωt/2 ω3 (cid:19) , (S6) for a low-frequency cut-off ω0 typically chosen as the in- verse of the measurement time. We assume ω0t ≪ 1 and 2 (cid:17)2 we can approximate χ(t) ≈(cid:16) t 2 (cid:19)1/f ≈r γN kBT 22 (cid:18) 8η2εz δε3 (cid:19) . (cid:18) 1 ω0t , with ln 1 (S7) T ∗ T ∗ Appendix C: Johnson Noise The detailed derivation of Eq. (6) will be given else- where. However, its qualitative form is not difficult to understand by analogy with an interaction of the van der Waals type.72 Let the ferromagnetic particle be located at the origin. A momentary fluctuation of the magnetic dipole of the qubit produces a corresponding fluctuation of the magnetic field ∼ 1/r3 at the ferromagnet. This induces a magnetic polarization µ ∼ α/r3 of the magnet which in turn causes a field B ∼ µ/r3 ∼ α/r6 at the qubit. The temperature dependence is specified by the fluctuation-dissipation theorem, which also requires that it is only the dissipative part of α that contributes. 1A. M. Tyryshkin, T. Shinichi, J. J. L. Morton, H. Riemann, N. V. Abrosimov, P. Becker, H. Pohl, T. Schenkel, M. L. W. Thewalt, K. M. Itoh, and S. A. Lyon, Nature Materials 11, 143 (2012). 2S. Amasha, K. MacLean, I. P. Radu, D. M. Zumbuhl, M. A. Kastner, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 100, 046803 (2008). 3N. Bar-Gill, L. Pham, A. Jarmola, D. Budker, and R. Walsworth, Nature 453, 1043 -- 1049 (2008). 4D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). 5G. Burkard, D. Loss, and D. P. DiVincenzo, Phys. Rev. B 59, 2070 (1999). 6P. Huang and X. Hu, Phys. Rev. B 90, 235315 (2014). 7K. M. Itoh and H. Watanabe, arXiv:1410.3922 (to be published in MRS Communications). 6 8F. A. Zwanenburg, A. S. Dzurak, A. Morello, M. Y. Simmons, L. C. L. Hollenberg, G. Klimeck, S. Rogge, S. N. Coppersmith, and M. A. Eriksson, Rev. Mod. Phys. 85, 961 -- 1019 (2013). 9H. W. Liu, T. Fujisawa, Y. Ono, H. Inokawa, A. Fujiwara, K. Takashina, and Y. Hirayama, Phys. Rev. B 77, 073310 (2008). 10M. Prada, R. H. Blick, and R. Joynt, Phys. Rev. B 77, 115438 (2008). 40Y. Makhlin, G. Schon, and A. Shnirman, Rev. Mod. Phys. 73, 357 -- 400 (2001). 41See supplemental material at [URL will be inserted by AIP] for the total Hamiltonian and details on the formal treatment of dephasing and Johnson noise. 42J. H. Davies, The physics of low-dimensional semiconductors: an introduction (Cambridge University Press, 1998). 11J. K. Gamble, M. Friesen, S. N. Coppersmith, and X. Hu, Phys. 43R. Winkler, Spin-orbit coupling effects in two-dimensional elec- Rev. B 86, 035302 (2012). 12F. H. L. Koppens, C. Buizert, K. J. Tielrooij, I. T. Vink, K. C. Nowack, T. Meunier, L. P. Kouwenhoven, and L. M. K. Vander- sypen, Nature 442, 766 -- 771 (2006). 13E. Kawakami, P. Scarlino, L. R. Schreiber, J. R. Prance, D. E. Savage, M. G. Lagally, M. A. Eriksson, and L. M. K. Vander- sypen, Appl. Phys. Lett. 103, 132410 (2013). 14K. C. Nowack, F. H. L. Koppens, Y. V. Nazarov, and L. M. K. Vandersypen, Science 318, 1430 (2007). 15M. Pioro-Ladri`ere, T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Nature Phys. 4, 776 -- 779 (2008). 16M. Pioro-Ladri`ere, Y. Tokura, T. Obata, T. Kubo, and S. Tarucha, Appl. Phys. Lett. 90, 024105 (2007). 17Y. Tokura, W. G. van der Wiel, T. Obata, and S. Tarucha, Phys. Rev. Lett. 96, 047202 (2006). 18E. Kawakami, P. Scarlino, D. R. Ward, F. R. Braakman, D. E. Savage, M. G. Lagally, M. Friesen, S. N. Coppersmith, M. A. Eriksson, and L. M. K. Vandersypen, Nature Nanotech. 9, 666 -- 670 (2014). 19A. van der Ziel, Proceedings of the IEEE 58, 1178 -- 1206 (1970). 20S. W. Jung, T. Fujisawa, Y. Hirayama, and Y. H. Jeong, Appl. Phys. Lett. 85, 768 -- 770 (2004). 21K. Hitachi, T. Ota, and K. Muraki, Appl. Phys. Lett. 102, 192104 (2013). 22C. R. Helms and E. H. Poindexter, Rep. Prog. Phys. 57, 791 -- 852 (1994). 23E. Machlin, "Chapter {VII} - Defects and Properties," in Ma- terials Science in Microelectronics {II}, edited by E. Machlin (Elsevier Science Ltd, Oxford, 2006) second edition ed., pp. 215 -- 250. 24E. Paladino, Y. M. Galperin, G. Falci, and B. L. Altshuler, Rev. tron and hole systems (Springer, 2003). 44E. I. Rashba, Phys. Rev. B 78, 195302 (2008). 45D. Culcer, L. Cywi´nski, Q. Li, X. Hu, and S. Das Sarma, Phys. Rev. B 82, 155312 (2010). 46C. H. Yang, A. Rossi, R. Ruskov, N. S. Lai, F. A. Mohiyaddin, S. Lee, C. Tahan, G. Klimeck, A. Morello, and A. S. Dzurak, Nat. Comm. 4, 2069 (2013). 47With these assumptions the variations in V(t) arise from η, δv and δε. 48A. Gold, "Transport properties of silicon/silicon germanium (Si/SiGe) nanostructures at low temperatures," in Silicon Ger- manium (SiGe) Nanostructures, Woodhead Publishing Series in Electronic and Optical Materials, edited by Y. Shiraki and N. Us- ami (Woodhead Publishing, 2011) pp. 361 -- 398. 49N. Mori, "Electronic band structures of silicon -- germanium (SiGe) alloys," in Silicon-Germanium (SiGe) Nanostructures, Woodhead Publishing Series in Electronic and Optical Materi- als, edited by Y. Shiraki and N. Usami (Woodhead Publishing, 2011) pp. 26 -- 42. 50S. Adachi, "Optical properties of AlGaAs: reststrahlen re- gion (discussion)," in Properties of aluminium gallium arsenide, edited by S. Adachi (INSPEC, the Institution of Electrical Engi- neers, 1993) pp. 89 -- 94. 51S. Kogan, Electronic noise and fluctuations in solids (Cambridge University Press, New York, 2008). 52P. Dutta and P. M. Horn, Rev. Mod. Phys. 53, 497 -- 516 (1981). 53K. Takeda, T. Obata, Y. Fukuoka, W. M. Akhtar, J. Kamioka, T. Kodera, S. Oda, and S. Tarucha, Appl. Phys. Lett. 102, 123113 (2013). 54L. V. C. Assali, H. M. Petrilli, R. B. Capaz, B. Koiller, X. Hu, and S. Das Sarma, Phys. Rev. B 83, 165301 (2011). 55H. Bluhm, S. Foletti, D. Mahalu, V. Umansky, and A. Yacoby, Mod. Phys. 86, 361 -- 418 (2014). Phys. Rev. Lett. 105, 216803 (2010). 25Y. Nakamura, Y. A. Pashkin, T. Yamamoto, and J. S. Tsai, 56H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Mahalu, V. Uman- Phys. Rev. Lett. 88, 047901 (2002). 26O. E. Dial, M. D. Shulman, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Phys. Rev. Lett. 110, 146804 (2013). 27E. Dupont-Ferrier, B. Roche, B. Voisin, X. Jehl, R. Wacquez, M. Vinet, M. Sanquer, and S. D. Franceschi, Phys. Rev. Lett. 110, 136802 (2013). 28K. D. Petersson, J. R. Petta, H. Lu, and A. C. Gossard, Phys. Rev. Lett. 105, 246804 (2010). 29T. Hayashi, T. Fujisawa, H. D. Cheong, Y. H. Jeong, and Y. Hi- rayama, Phys. Rev. Lett. 91, 226804 (2003). 30P. Huang and X. Hu, Phys. Rev. B 89, 195302 (2014). 31A. Bermeister, D. Keith, and D. Culcer, Appl. Phys. Lett. 105, 192102 (2014). sky, and A. Yacoby, Nature Phys. 7, 109 (2011). 57J. Muhonen, J. Dehollain, A. Laucht, F. Hudson, T. Sekiguchi, K. Itoh, D. Jamieson, J. McCallum, A. Dzurak, and A. Morello, Nature Nanotechnology 9, 986991 (2014). 58T. Obata, K. Takeda, J. Kamioka, T. Kodera, W. M. Akhtar, K. Sawano, S. Oda, Y. Shiraki, and S. Tarucha, JPS Conference Proceedings 1, 012030 (2014). 59X. Wu, D. R. Ward, J. R. Prance, D. Kim, J. K. Gamble, R. T. Mohr, Z. Shi, D. E. Savage, M. G. Lagally, M. Friesen, S. N. Coppersmith, and M. A. Eriksson, Proc. Natl. Acad. Sci. 111, 11938 -- 11942 (2014). 60C. H. Wong, M. A. Eriksson, S. N. Coppersmith, and M. Friesen, Phys. Rev. B 92, 045403 (2015). 32L. S. Langsjoen, A. Poudel, M. G. Vavilov, and R. Joynt, Phys. 61R. Thalineau, S. R. Valentin, A. D. Wieck, C. Bauerle, and Rev. B 89, 115401 (2014). 33S. Kolkowitz, A. Safira, A. A. High, R. C. Devlin, S. Choi, Q. P. Unterreithmeier, D. Patterson, A. S. Zibrov, V. E. Manucharyan, H. Park, and M. D. Lukin, Science 347, 1129 -- 1132 (2015). 34R. Neumann and L. R. Schreiber, J. Appl. Phys. 117, 193903 (2015). 35I. Martin and Y. M. Galperin, Phys. Rev. B 73, 180201 (2006). 36E. Paladino, Y. M. Galperin, G. Falci, and B. L. Altshuler, "1/f implications for solid-state quantum information," Rev. noise: Mod. Phys. 86, 361 (2014). 37L. Chirolli and G. Burkard, Adv. Phys. 57, 225 (2008). 38D. Culcer, X. Hu, and S. Das Sarma, Appl. Phys. Lett. 95, 073102 (2009). 39J. Clarke and F. K. Wilhelm, Nature 453, 1031 -- 1042 (2008). T. Meunier, Phys. Rev. B 90, 075436 (2014). 62M. D. Shulman, O. E. Dial, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Science 336, 202 -- 205 (2012). 63B. M. Maune, M. G. Borselli, B. Huang, T. D. Ladd, P. W. Deelman, K. S. Holabird, A. A. Kiselev, I. Alvarado-Rodriguez, R. S. Ross, A. E. Schmitz, M. Sokolich, C. A. Watson, M. F. Gyure, and A. T. Hunter, Nature 481, 344 -- 347 (2012). 64S. M. Vonsovskii, Ferromagnetic Resonance (Pergamon Press, 1966). 65Z. Frait, Czech. Journ. Phys. 11, 360 (1961). 66R. Ferr´e, K. Ounadjela, J. M. George, L. Piraux, and S. Dubois, Phys. Rev. B 56, 14066 -- 14075 (1997). 67M. Respaud, M. Goiran, J. M. Broto, F. H. Yang, T. Ould- Ely, C. Amiens, and B. Chaudret, Phys. Rev. B 59, 3934 -- 3937 (1999). 68R. de Sousa and S. Das Sarma, Phys. Rev. B 68, 115322 (2003). 69D. Culcer, X. Hu, and S. D. Sarma, Appl. Phys. Lett. 95, 073102 (2009). 70D. Culcer and N. M. Zimmerman, Appl. Phys. Lett. 102, 232108 (2013). 71R. de Sousa, "Electron spin as a spectrometer of nuclear-spin noise and other fluctuations," in Electron Spin Resonance and Related Phenomena in Low-Dimensional Structures, Topics in Applied Physics, Vol. 115, edited by M. Fanciulli (Springer Berlin Heidelberg, 2009) pp. 183 -- 220. 72L. Pitaevskii and E. Lifshitz, Statistical Physics, Part 2: Theory of the Condensed State (Butterworth-Heinemann, 1980). 7 

1601.01889

2

1601

Structured Back Gates for High-Mobility Two-Dimensional Electron Systems Using Oxygen Ion Implantation

We present a new approach of back gate patterning that is compatible with the requirements of highest mobility molecular beam epitaxy. Contrary to common back gating techniques, our method is simple, reliable and can be scaled up for entire wafers. The back gate structures are defined by local oxygen implantation into a silicon doped GaAs epilayer, which suppresses the conductance without affecting the surface quality.

cond-mat.mes-hall

cond-mat

Structured Back Gates for High-Mobility Two-Dimensional Electron Systems Using Oxygen Ion Implantation M. Berl,1, a) L. Tiemann,1 W. Dietsche,1 H. Karl,2 and W. Wegscheider1 1)Solid State Physics Laboratory, ETH Zurich, 8093 Zurich, Switzerland 2)Lehrstuhl fur Experimentalphysik IV, Universitat Augsburg, 86159 Augsburg, Germany (Dated: 2 July 2021) We present a reliable method to obtain patterned back gates compatible with high mobility molecular beam epitaxy (MBE) via local oxygen ion implantation that sup- presses the conductivity of an 80 nm thick silicon doped GaAs epilayer. Our technique was optimized to circumvent several constraints of other gating and implantation methods. The ion-implanted surface remains atomically flat which allows unper- turbed epitaxial overgrowth. We demonstrate the practical application of this gating technique by using magneto-transport spectroscopy on a two-dimensional electron system (2DES) with a mobility exceeding 20× 106 Vs/cm2. The back gate was spa- tially separated from the Ohmic contacts of the 2DES, thus minimizing the proba- bility for electrical shorts or leakage and permitting simple contacting schemes. 6 1 0 2 n a J 2 1 ] l l a h - s e m . t a m - d n o c [ 2 v 9 8 8 1 0 . 1 0 6 1 : v i X r a a)Electronic mail: [email protected] 1 Reliable electrostatic gating is a key element in most experiments in semiconductor physics to tune the charge carrier density of a two-dimensional electron (2DES) or hole (2DHS) system, which is intrinsically defined by doping. For sophisticated mesoscopic sys- tems or nanostructures, gates need to be patterned. While structured top gates can be obtained with relative ease during the sample fabrication, patterning back gates is demand- ing and each available method has certain technological limitations. Structured back gates can be obtained by thinning down the substrate, followed by the evaporation of metal gates on the backside1,2. The fragility of these thinned samples, how- ever, make their handling difficult. In addition, as a result of the large gate separation, substantial gate voltages have to be applied. Several alternative technologies have therefore been exploited to pattern back gates on the epitaxial side of the wafer. For overgrown pre- structured back gates, the gates are wet-chemically etched out of a highly doped layer on the epitaxial side of the substrate3 -- 6, prior to the growth of the heterostructure. Since high mobility epitaxy critically depends on the surface roughness, the etched surface of these systems will not favor high mobility 2DESs. A different approach is the implantation of oxygen ions to locally suppress the conductivity of a doped epilayer on an otherwise insulat- ing wafer. It is a commonly used industrial technique and has also been applied in research scale molecular beam epitaxy7,8. However, along with focused ion beam implantation9, all these methods can result in an amorphization of the surface and are not tailored to highest mobility samples required in fundamental research. Here we report on an optimized technique using oxygen ion implantation which circum- vents the limitations of other available back gating technologies. Our method allows to struc- ture entire wafers for high mobility heterostructure growth by using standard optical photo resist as a shadow mask for the impinging oxygen ions. Atomic force microscopy (AFM) con- firms that the substrate surface remains atomically flat after the implantation which is the prerequisite for high mobility molecular beam epitaxy. We use standard magneto-transport measurements to demonstrate the suitability and reliability of this gating method and the compatibility with high mobility MBE. The fabrication begins with a metal organic chemical vapor deposition (MOCVD) of a 80 nm thick homogeneous silicon doped GaAs epilayer grown on a semi-insulating two inch GaAs wafer (Fig. 1a). The doping layer is covered by a 30 nm thick undoped GaAs caping layer to ensure that the peak value of the subsequent implantation will be located within 2 FIG. 1. a)-f) Sample processing scheme; a) MOCVD growth of an 80 nm Si doped GaAs epilayer (green layer) and a 30 nm thick undoped GaAs cap grown on a (001) semi-insulating GaAs wafer b) Patterning with photo resist c) Oxygen ion implantation d) Epiready cleaning of the surface after photo resist removal e) MBE overgrowth including a quantum well confined 2DES (green layer) f) Wet chemical mesa etching and removal of the 2DES over the back gate contact areas g) Implan- tation depth and doping profile within the GaAs substrate. The total implantation (red squares) is the sum of two implantation distributions with 1.6× 1013 Ions/cm2 at 20 keV (light grey shaded area) and 1.0× 1014 Ions/cm2 at 50 keV (dark grey shaded), which were determined from TRIM simulations11. The doping layer (30 nm-110 nm) with a doping concentration of 7.5× 1017 cm−3 is indicated by the green bar. the doping layer. The shift of the peak further into the substrate also reduces the implan- tation damage to the surface. We experimented with doping concentrations ranging from 3× 1016 cm−3 to 3× 1018 cm−3 and found 7.5× 1017 cm−3 to be the most reliable choice for the subsequent ion implantation. If the doping concentration is too high, it is not possible to suppress the conductivity by oxygen implantation10. If the doping concentration is too low, the resistivity of the back gate and its Ohmic contacts will be too high. After the MOCVD growth, the wafer is covered with photo resist (Microchemicals GmbH AZ1518) and patterned by standard optical photo lithography (Fig. 1b). The photo resist is suffi- ciently thick (∼ 2 µm) to work as a selective absorber, i.e., shadow mask, for the impinging oxygen ions so that only the exposed parts will become implanted with ions (Fig. 1c). The reliable operation of our gating technique crucially depends on the interplay between doping concentration and ion implantation profile which is fundamentally given by the dose and 3 the energy of implantation. To ensure that most of the ion incorporation is targeted at the 80 nm thick doping layer, we have chosen two implantation doses of different energies after determining the optimal implantation distribution with a TRIM simulation11 (Fig. 1g). We tested several pairs of doses and found 1.6× 1013 Ions/cm2 at 20 keV and 1.0× 1014 Ions/cm2 at 50 keV to be the optimal dose which allows for a stable suppression of conductivity while minimizing the impact of implantation to the surface crystal structure. The atomically smoothness of the implanted surface was confirmed with AFM measurements. The measured root mean square roughness is smaller than 0.5 nm, which is comparable to the roughness of non-implanted surfaces. Following implantation, the photo resist has to be carefully and completely removed from the wafer with acetone and plasma ashing. The wafer is then undergoing an epi- ready treatment (Fig. 1d), which consists of a surface cleaning with H2SO4 acid and a surface passivation by controlled surface oxidation on a hotplate (3 minutes at 300◦C). The epi-ready process is necessary to prepare the surface for the MBE overgrowth. For the heterostructure to be overgrown we choose comparable conditions as used for standard high mobility structures, which includes an arsenic-free bake-out of the wafer at around 450◦C and a subsequent high temperature treatment (630◦C for about 5 hours) during the heterostructure growth (Fig. 1e). After this high temperature treatment we observed changes in the resistivity of the implanted layers. These changes, which are known to arise from the annealing of the implantation induced defects12,13, however, are negligible and do not affect the gate operation. We have grown several GaAs/AlGaAs high mobility heterostructures on back gate pat- terned wafers and now present the results from two of these MBE overgrown heterostruc- tures, labeled as HS1 & HS2. HS1 is doubled-sided delta-doped, whereas HS2 is single-sided delta-doped on the surface-facing side of the quantum well to avoid screening effects be- tween 2DES and gate. Both heterostructures have a 27 nm wide GaAs quantum well buried 230 nm below the sample surface. The setback between doping layers and 2DES is 80 nm and the distance between back gate and 2DES is ∼ 1 µm. The Al concentration was kept low (∼ 24%) around the 2DES to obtain high mobilities but was increased towards the substrate in several steps up to ∼88% to prevent gate leakages. Standard lock-in magneto-transport measurements on square van der Pauw samples and soldered indium contacts were used to determine whether the insulating and conductive parts 4 FIG. 2. Magneto transport measurements at T≈ 1 K on van der Pauw samples with the same high mobility heterostructure (HS1) grown on different types of substrates; b) Substrate is silicon doped but not oxygen implanted; c) Substrate is silicon doped and oxygen implanted; d) Substrate is neither silicon doped nor oxygen implanted. (*) Values were calculated from Hall measurements at low magnetic field. of the sample are working properly after the high mobility heterostructure overgrowth. At low temperatures and high (perpendicular) magnetic fields, the density of states of the 2DES condenses into a dispersionless bands of quantized Landau levels. As a result of this Landau quantization, the longitudinal resistance displays Shubnikov -- de Haas oscillations whereas the transversal Hall resistance displays plateaus at integer values of the filling factor ν ∝ n B ; known as the integer quantum Hall effect14. Figure 2a shows magneto-transport measure- ments on a van der Pauw sample which was grown on a conductive silicon doped substrate without implantation. In this sample the contacts penetrate through the heterostructure including the conductive substrate. The quantum Hall effects exhibited by the 2DES are therefore superimposed to the parasitic parallel conductance originating from the silicon doped substrate. However, when the implantation sufficiently suppresses the substrate con- ductance, only the pristine quantum Hall effects (Fig. 2b) of the 2DES are observed. Charge carrier mobilities of 20 Mio. cm2/Vs could be measured, demonstrating unambiguously the suitability of implanted substrates for the delicate high mobility heterostructure overgrowth. 5 FIG. 3. 2DES mobility and carrier density as a function of the back gate voltage on a Hall bar structure using HS2 (T≈ 25 mK). Inset: Picture of an indium (black spots) contacted Hall bar mesa with the implanted substrate areas highlighted in red and the non-implanted conductive substrate parts highlighted in blue. To study the impact of the implantation on high mobility heterostructure growth, we have performed measurements on an identical heterostructure grown on a semi-insulating GaAs wafer (Fig. 2c) which was neither doped nor ion-implanted. The magneto-transport mea- surements clearly show that both the signatures of the quantum Hall effects and mobilities and carrier densities are nearly identical to that from the implanted substrate. High mobility heterostructures are the testbed for fragile exotic quantum effects. A promi- nent example are the fractional quantum Hall effects15,16 that arise due to electron-electron interactions at non-integer values of the filling factor. To demonstrate the utilization and ad- vantages of an implantation patterned back gate, we present low temperature measurements on standard Hall bar structures of high mobility heterostructures (HS1 & HS2). As depicted in the inset of Fig. 3, the inner Hall bar region of our Hall bar structure is located above the back gate, i.e., the region which was not ion-implanted, whereas the region underneath the contact arms was rendered non-conductive via oxygen ion bombard- ment. The exposed contacts (including the contact to the back gate) can therefore directly be contacted without risking an electrical short between back gate and 2DES. The only 6 FIG. 4. Magneto transport measurements at T≈ 25 mK on a Hall bar structure using HS1 with gate voltages applied from -2.5 V to +2 V. As a guide to the eye the Landau level filling factor 2 is labeled in each measurement. For reference, some fractional quantum Hall states were labeled as well. lithographic step which is necessary before making indium solder contacts is a wet-chemical etching process to define the Hall bar mesa and to remove the 2DES above the back gate contact areas (Fig. 1f). For HS2 the charge carrier density of the 2DES can be tuned over a wide range (Fig. 3) without provoking significant leakage current from the gate (Ileakage (cid:46) 30 pA). Over almost the entire range the carrier density is responding linearly to the gate volt- age. With increasing density the mobility could be raised from unbiased ∼ 6 Mio cm2/Vs up to ∼ 18 Mio. cm2/Vs by applying a gate voltage of about +3.5V. For HS1 low temperature magneto-transport measurements on the high mobility het- erostructure are shown in Fig.4 for three exemplary gate voltages. The existence of a variety of fractional quantum states are indicative of the very high sample quality. Coincident with density changes, the observed fractional quantum Hall states are becoming more and less distinct. The highest mobility (∼ 18 Mio. cm2/Vs) was achieved by increasing the charge carrier density to ∼ 3.7× 1011 cm−2 via the back gate. In summary, we demonstrated that the conductivity of a 80 nm thick silicon doped GaAs- substrate can be suppressed by oxygen implantation. No negative influences to the sample 7 surface, the heterostructure growth or the overall quality of the high mobility 2DES were observed. The functionality of a patterned ion-implantation to define back gate structures was demonstrated with magneto-transport measurements. The method we presented here is very reliable and gate-able structures can be produced with relative ease. We believe that this method will also be suitable for a variety of nanostructures, which currently rely on top gating schemes, as well as for an improved concept for double layer 2DES/2DHS systems4,5,9. The authors acknowledge financial support by the Swiss National Science Foundation (SNF) and the NCCR QSIT (National Competence Center in Research - Quantum Science and Technology. We thank Emilio Gini (ETH-FIRST cleanroom staff) for providing us with n-doped MOCVD substrates. REFERENCES 1J.P. Eisenstein, L.N. Pfeiffer, and K.W. West, Appl. Phys. Lett. 57, 2324 (1990). 2M.V. Weckwerth, J.A. Simmons, N.E. Harff, M.E. Sherwin, M.A. Blount, W.E. Baca, and H.C. Chui, Superlattice Microst. 20, 561 (1996). 3M.P. Grimshaw, D.A. Ritchie, J.H. Burroughs, and G.A.C. Jones, J. Vac. Sci. Technol. B 12(2), 1290 (1994). 4H. Rubel, A. Fischer, W. Dietsche, K. von Klitzing, and K. Eberl, Mater. Sci. Eng. B 51, 207 (1998). 5L. Tiemann, W. Dietsche. M. Hauser, and K. von Klitzing, New J. Phys. 10, 045018 (2008). 6X. Huang, W. Dietsche, M. Hauser, and K. von Klitzing, Phys. Rev. Lett. 109, 156802 (2012). 7Y. Ota, M. Yanagihara, and M. Inada, J. Appl. Phys. 64, 926 (1988). 8C.L. Chen, L. J. Mahoney, S.D. Calawa, K.M. Molvar, P.A. Maki, R.H. Mathews, J.P. Sage, and T.C.L.G. Sollner, Appl. Phys. Lett. 74, 4058 (1999). 9K.M. Brown, E.H. Linfield, D.A. Ritchie, G.A.C. Jones, M.P.Grimshaw, and M. Pepper, Appl. Phys. Lett. 64, 1827 (1994). 10S.J. Pearton, Ion implantation for isolation of III-V semiconductors, Mater. Sci. Rep. 4 (1990). 11J.F. Ziegler, and J.P. Biersack, Monte Carlo simulation for the transport of ions in matter, SRIM-2013. 8 12S.J. Pearton, J.M. Poate, F. Sette, J.M. Gibson, D.C. Jacobson and J.S. Williams, Nucl. Instr. Meth. Phys. Res. B19/20, 369 (1987). 13H. Muessig, C. Woelk, H. Brugger, and A. Forchel, Inst. Phys. Conf. Ser. No 136, 529 (1993). 14K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). 15D.C. Tsui, H.L. Stormer, A.C. Gossard Phys. Rev. Lett. 48, 1559 (1982). 16R.B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983). 9 

1708.01740

1

1708

Geometrical phase shift in Friedel oscillations

This work addresses the problem of elastic scattering through a localized impurity in a one-dimensional crystal with sublattice freedom degrees. The impurity yields long-range interferences in the local density of states known as Friedel oscillations. Here, we show that the internal degrees of freedom of Bloch waves are responsible for a geometrical phase shift in Friedel oscillations. The Fourier transform of the energy-resolved interference pattern reveals a topological property of this phase shift, which is intrinsically related to the Bloch band structure topology in the absence of impurity. Therefore, Friedel oscillations in the local density of states can be regarded as a probe of wave topological properties in a broad class of classical and quantum systems, such as acoustic and photonic crystals, ultracold atomic gases in optical lattices, and electronic compounds.

cond-mat.mes-hall

cond-mat

Geometrical phase shift in Friedel oscillations Univ Lyon, Ens de Lyon, Univ Claude Bernard, CNRS, Laboratoire de Physique, F-69342 Lyon, France C. Dutreix and P. Delplace This work addresses the problem of elastic scattering through a localized impurity in a one-dimensional crystal with sublattice freedom degrees. The impurity yields long-range interferences in the local density of states known as Friedel oscillations. Here, we show that the internal degrees of freedom of Bloch waves are responsible for a geometrical phase shift in Friedel oscillations. The Fourier transform of the energy-resolved interference pattern reveals a topological property of this phase shift, which is intrinsically related to the Bloch band structure topology in the absence of impurity. Therefore, Friedel oscillations in the local density of states can be regarded as a probe of wave topological properties in a broad class of classical and quantum systems, such as acoustic and photonic crystals, ultracold atomic gases in optical lattices, and electronic compounds. Electric screening in metals arises as a collective response of the conduction electrons to the Coulomb potential of a charged impurity. A long wavelength description of the prob- lem captures the exponential screening of the impurity by the surrounding electrons, in agreement with the classical picture that depicts electrons as point charges. In quantum mechan- ics, however, particles are described by wavefunctions and may interfere. In the 1950s, J. Friedel actually understood that a charged impurity additionally yields a long-range inter- ference pattern in the electronic density [1]. It consists of al- gebraically decaying 2kF-wavevector oscillations, and results from Fermi surface nesting associated to twice the Fermi mo- mentum kF. Thus, these oscillations reported by Friedel for charges in metals rely on wave features, and they have subse- quently been revisited in other contexts, such as magnetic in- teractions [2–5] and noninteracting electrons [6–8]. In partic- ular, Friedel oscillations have been observed in nonrelativistic electron gases via scanning tunnelling microscopy (STM); an experimental technique that images the local density of states (LDOS), i.e., the electronic density with atomic-scale and en- ergy resolutions [9–11]. These experiments have confirmed that backscattering was the most efficient process involved in the elastic scattering through short-range impurities. For non- interacting electrons in a one-dimensional crystal, backscat- tering is indeed responsible for 2k0-wavevector Friedel oscil- lations that behave as δρ(m, ω) = V(ω) cos(2k0m) . (1) Here δρ denotes the correction to the LDOS induced by a lo- calized impurity, m labels the distance to the impurity in units of the Bravais lattice vector, V(ω) is some real function, and wavevector 2k0 refers to the elastic backscattering between the time-reversed states −k0 and k0 at energy ω ( = 1). Such backscattering wavevectors are illustrated in Fig. 1 by the dou- ble arrows. The reader may find the details of the derivation of Eq. (1) in Supplemental Material (SM) [12]. In this letter, we would like to highlight the effects of inter- nal degrees of freedom on the long-range interference pattern induced by a localized impurity. These degrees of freedom could for example consist of spins up and down, or charge- conjugated particles such as electrons and holes. Here, how- ever, we rather consider they result from a sublattice structure, FIG. 1. (Color online) Top: monatomic (left) and diatomic (right) pattern crystals with a localized impurity V. Bottom: Bloch spec- tra of the monatomic (left) and diatomic (right) pattern crystals in the first Brillouin zone. The horizontal double arrows depict elastic scattering processes between time-reversed states −k0 and k0. as illustrated in the right-hand column of Fig. 1. This one- dimensional dimerized structure naturally appears in some or- ganic compounds [13], and may be viewed as resulting from the Peierls instability of a metallic crystal like the one depicted in the left-hand column of Fig. 1. It has also been realized in acoustics, photonics and ultracold atomic gases [14–17]. The system is obviously invariant under time and space inversions. For the sake of simplicity we restrict the discussion to two sublattices, namely, A and B, within a nearest-neighbor tight- binding approximation. Nonetheless, this is not detrimental to the relevance of the results presented below, and the reader may find a generalization to an arbitrary number of freedom degrees in SM [12]. The Bloch band structure is then charac- terized by a 2 × 2 Hermitian matrix which, in the sublattice i=1 di(k) σi, where d1(k) = t1(α + cos k) and d2(k) = t1 sin k. The parameters αt1 and t1 respectively denote the intra- and inter-dimer hop- ping amplitudes whose ratio is α, while matrix σi is the ith Pauli matrix. Since there is only one energy scale in this de- scription, energy will be given in units of t1, if not otherwise specified. This prototypical model is sometimes referred to as SSH model, with reference to the work of Su, Schrieffer, and Heeger about the formation of topological solitons in poly- acetylene CnHn [13]. Note that there is no term scaling with basis {A, B}, generically reads H(k) = (cid:80)2 -π-k0k0πEnergyelastic scattering-π-k0k0πEnergyelastic scattering1(k) + d2 σ3 under time and space inversions. Furthermore, we disre- gard any contribution scaling with the identity matrix in the Bloch Hamiltonian matrix, such as next-nearest-neighbor pro- cesses. This would neither change the Bloch wavefunctions, nor the effects they are responsible for in the elastic scatter- ing. The band structure is then entirely characterized by the set of the dispersion relation E±(k) = ±[d2 2(k)]1/2 and the Bloch eigenstates (cid:104)±, k = (1, ±eiθk). The Bloch eigen- states are not gauge invariant, obviously, since eiφk±, k(cid:105) is also eigenstate for any arbitrary φk. But function θk, which defines the phase shift between the two internal degrees of freedom of Bloch spinors, is not affected by such a gauge change. How- ever, this phase shift turns out to be ill-defined, because there exists an ambiguity when defining the diatomic unit cell of the translationally invariant crystal. Indeed, this phase shift can be defined either as θk, or as θk − k. Of course, the con- vention we chose to describe the system will never affect the observable we will be focussing on, namely, the LDOS [12]. From now on, we consider a localized impurity on sublat- tice A, which is simulated by V δm,0. The impurity breaks the translational invariance of the crystal and defines a natural origin. The elastic scattering experienced by Bloch wavefunc- tions on the defect is described within a T-matrix approach. This consists of a perturbation theory in the impurity poten- tial V. As shown in Fig 2, for such a localized potential the scattering diagrams define a geometric series and all the or- ders can be summed up exactly, regardless of the magnitude of the impurity potential. After introducing the retarded bare Green's matrix as G(0)(k, ω) = [ω+i
−H(k)]−1, where
−1 > 0 describes a finite quasiparticle lifetime, the T-matrix reads T(ω) = t(ω) . (2) (cid:33) (cid:32) 1 0 0 0 (cid:82) Here t(ω) = V [1 − V AA(k, ω)]−1 does not depend on wavevector k, due to the integration that runs over the Bril- louin zone. The correction to the LDOS in the presence of the impurity is finally obtained as dk G(0) (cid:104) (cid:105) G(0)(m, ω) T(ω) G(0)(−m, ω) δρ(m, ω) = − 1 π Im Tr (3) in the limit
→ 0+. According to Fig. 1, the dispersion re- lation of the valence band looks the same as the one of the monatomic crystal, and so do the elastic scattering wavevec- tors between two time-reversed states −k0 and k0. Therefore, if one restricts the elastic scattering problem to a spectral anal- FIG. 2. Diagrammatic representation of the T-matrix within a pertur- bation theory in the impurity potential V. The oriented lines between two elastic scattering processes denote the bare Green's functions in momentum space. 2 FIG. 3. (Color online) Energy-resolved interference pattern induced in the LDOS of sublattices A (top) and B (bottom) by a localized impurity. The latter, which is simulated by a potential of V = 1, belongs to sublattice A and unit cell m = 0. Its position is marked by the vertical white dashed line. Energy is given in units of hopping t1. ysis, one would naively expect the localized impurity to in- duce Friedel oscillations that behave as in Eq. (1) on both sub- lattices, A and B. The interference patterns obtained from the numerical evaluation of Eq. (3) for V = 1 and α = 0.9 are shown in Fig. 3. These patterns are resolved in energy, and the latter is restricted to positive values because we arbitrarily focus on the conduction band (1 − α ≤ ω ≤ 1 + α). Whereas the interferences on sublattice A are symmetric with respect to the impurity, they turn out to be asymmetric on sublattice B at low energies. This asymmetry is clearly in disagreement with the behavior of the Friedel oscillations introduced in Eq. (1), and it cannot simply be understood from spectral features. In order to get some insight into this asymmetric inter- ference pattern, we perform the T-matrix approach analyti- cally [12]. The LDOS corrections it leads to for the two sub- lattices read δρA(m, ω) = V(ω) cos(2k0m) δρB(m, ω) = V(ω) cos(2k0m + ϕk0) , (4) where m labels the diatomic unit cells from the impurity, k0 is a pole of G(0)(k, ω) defined by ω2 = E2±(k0) for −π ≤ k0 ≤ 0. TG(0)G(0)VVVG(0)VVV=+++...-1010m 02ωδρA(m,ω)-11-1010m 02ωδρB(m,ω)-113 Thus, if the spectral analysis illustrated in Fig. 1 is sufficient to understand the 2k0-wavevector oscillations as resulting from elastic backscattering between time-reversed states k0 and −k0, it does not explain the existence of the phase shift in the Friedel oscillations on the pristine sublattice, namely, sublat- tice B, which yields the asymmetric interferences shown in Fig. 3. It has to be stressed that this phase shift does not arise from the condition that the scattering wavefunctions have to satisfy at the boundary with the impurity, as initially intro- duced by Friedel for finite size defects [1]. In the case of a localized impurity, indeed, there is no Friedel phase shift, as already shown in Eq. (1). The phase shift involved on sublat- tice B actually arises from the internal degrees of freedom of the Bloch wavefunctions involved in the elastic scattering. It explicitly reads = θk0 − θ−k0 ϕk0 = 2θk0 , (5) where the last equality may be understood as resulting from time reversal symmetry. It is directly related to θk0, that is, the phase shift between the two components of the Bloch spinor of state k0. Importantly, the LDOS corrections on sublattices A and B are observables; they are for instance accessible in atomic-scale-resolved STM experiments. So they do not de- pend on the ambiguity there is in the definition of the diatomic unit cell. This issue is explicitly fixed in SM [12]. The Fourier analysis of interference patterns is often very instructive too. For example, it has demonstrated the ability of STM to probe the Fermi iso-energy contours in nonrelativis- tic electron gases [10, 11], as well as the absence of backscat- tering of the massless relativistic charge carriers in graphene, and at the surfaces of three-dimensional topological insulators [18–20]. Here, the Fourier transform of the Friedel oscilla- tions introduced in Eq. (4) leads to Dirac combs δ(q − 2k0 + n2π) δρA(q, ω) = V(ω) δ(q − 2k0 + n2π) e−iϕk0 , n=−∞ δρB(q, ω) = V(ω) (6) where δ denotes the Dirac delta function, and −π ≤ k0 ≤ +π. Thus, the Fourier transform is a 4π-periodic function of q [12]. The modulus of the Fourier transform is the same for both sub- lattices, A and B. Its intensity is maximum for the backscat- tering wavevectors q = 2k0 [2π]. This can be understood from the spectral features of the band structure in the absence of the impurity, as shown by the double arrows in Fig. 1. This behavior is in agreement with the top panel of the left-hand column in Fig. 4, which represents the Fourier transform of the interference pattern of sublattice B, previously depicted in Fig. 3. It results from the numerical evaluation of the Fourier transform of the LDOS correction +∞(cid:88) +∞(cid:88) n=−∞ (cid:90) (cid:90) i δρ(q, ω) = 2π − i 2π Tr Tr (cid:104) (cid:104) dk dk G(0)(k + q, ω) T(ω) G(0)(k, ω) G(0)(k, ω) T(ω) G(0)(k + q, ω) (7) (cid:105) (cid:105)∗ FIG. 4. (Color online) Modulus (top) and argument (bottom) of the Fourier transform of the energy-resolved interference pattern on sub- lattice B for α < 1 (left) and α > 1 (right). The white dashed lines mark the wavevectors q = ±2k0 at energy ω. for the impurity potential V = 1 and α = 0.9. The high intensity areas agree with the white dashed lines that mark the Dirac delta functions when the backscattering wavevec- tors q = 2k0 [2π] described by Eq. (6). Note that only the Fourier transform for sublattice B is shown in Fig. 4, since the modulus of the Fourier transform for sublattice A is identical. The argument of the Fourier transform along the maximum- intensity lines reveals the phase shift involved in the Friedel oscillations on sublattice B, according to Eq. (6). It is repre- sented in the second plot of the left-hand column in Fig. 4. Note that only one white dashed line is shown in the figure. It defines a periodic line in (q, ω)-space that we denote by C2k0. This line corresponds to the wavevectors q = 2k0 depicted by the blue double arrow in Fig. 1. The Umklapp scattering pro- cesses associated to the red double arrow would lead to an- other white dashed line from which we would learn the same information, so it is not shown. Path C2k0 is a closed path, along which we can define the following mapping: ϕ : q ∈ S 1 = [−2π, 2π] (cid:55)→ ϕq/2 ∈ S 1 = [−π, π] (8) This mapping supports a topological characterization that in- dexes equivalence classes referred to as homotopy classes. They form some groups that are examples of topological in- variants. As far as we are concerned, the topology of the map- ping ϕ : S 1 (cid:55)→ S 1 is characterized by the first homotopy group of spheres, namely, π1(S 1) = Z. This topological invariant is an integer that counts the number of times that ϕq/2 winds around the circle S 1 = [−π, π], when scattering wavevector q runs once along S 1 = [−2π, 2π]. This is nothing but the -2π2πq 02ω01α<1-2π2πq 02ω01α>1-2π2πq02ω-ππα<1-2π2πq02ω-ππα>1(cid:90) π −π dk 2π W = 2 (cid:72) (cid:73) winding number given by W = 1 2iπ dq ∂q ln C2k0 (cid:34) (cid:35) = (cid:90) π −π dq 2π δρA(q, ω) δρB(q, ω) ∂q ϕq . (9) From the second plot of the left-hand column in Fig. 4, we in- fer that W = 2. When changing the ratio between intra- and inter-dimer hoppings to α > 1, one gets the right-hand column of Fig. 4. It has been obtained for parameters α = 10/9 and t1 = 0.9, so that the energy spectrum as well as the modulus of the Fourier transform remain unchanged. The phase of the Fourier transform along the maximum intensity closed path C2k0 no longer winds and W = 0. Therefore, the presence of a localized impurity, which breaks the translational invariance, induces interference patterns that support nonequivalent topo- logical characterizations, depending on the global properties of the geometrical phase shift involved in Friedel oscillations. = 2θk0 under time-reversal symmetry. So the topological properties of the interference-pattern Fourier transform are intrinsically related to the Bloch band structure topology, which is itself characterized by the winding number W: Moreover, the geometrical phase shift satisfies ϕk0 ∂k θk = 2W . Topological invariant W is then obviously connected to the Zak phase of the Bloch wavefunctions in the absence of im- BZ dk(cid:104)±, k∂k±, k(cid:105), via the following purity, namely, γ = i relation: W = 2γ/π. The Zak phase is analogous to the Berry phase for one-dimensional systems [21] where the the role of the periodic external parameter is played by the quasi- momentum running over the 1D Brillouin zone. It refers here to the gauge-invariant geometrical phase picked up by the Bloch wavefunctions along the Brillouin zone. This is a crucial quantity involved in many fields of physics, among which charge pumping, electric polarization, orbital mag- netism, symmetry-protected topological order and edge states [22–27]. It must be stressed that the value of such a phase depends on the choice of origin in the Fourier transform for sublattice B, or equivalently on the ambiguity there is in the definition of the unit cell. Indeed, since the definition of θk depends on the unit cell convention we choose, so does the winding number W it leads to. Despite this ambiguity, the dif- ference of winding numbers W(α > 1) − W(α < 1) does not depend on this choice and is a well-defined quantity. This is actually a well-known issue, and it explains for example why the polarization in a crystal, which depends on the Zak phase of Bloch wavefunction, is only defined modulo a quantum of polarization [23, 24]. A fortiori, only W(α > 1) − W(α < 1) is well-defined and enables the distinction between the Four- rier transform of two topologically nonequivalent interference patterns, as the ones of Fig. 4. In summary, we have addressed the scattering problem of a localized impurity in a one-dimesional crystal for Bloch waves that possess internal degrees of freedom. While the im- purity obviously yields Friedel oscillations in the LDOS asso- ciated to backscattering wavevector q = 2k0, we have shown 4 that the internal freedom degrees of Bloch waves are respon- sible for a geometrical phase shift. The latter does not relate to the nature of the impurity and is then intrinsically different from the so-called Friedel phase shift. The Fourier transform of the energy-resolved interference pattern has revealed the momentum dependence of this geometrical phase shift, whose global properties enable us to discriminate two topologically nonequivalent interference patterns. Remarkably, these topo- logical features are intrinsically connected to the Zak phase under-space and time-inversion symmetries, which character- izes the Bloch band structure topology in the absence of impu- rity. Measurements of the Zak phase have already been real- ized, for example through Bloch oscillations [14] and in non- Hermitian systems with losses [16, 28], but these prescriptions remain rather unsuitable for electronic compounds. Since the LDOS is a physical observable that is accessible in acoustic and photonic crystals, ultracold atomic gases, and electronic materials via STM, the interference pattern induced by a local- ized defect offers a priori a joint route to image the band struc- ture topology. Besides, this consists of a bulk measurement since it probes Bloch bands properties. Above, we have in- deed focused on elastic scattering occurring in the conduction band of a one-dimensional insulator. A strong impurity po- tential V (cid:29) 1 would be responsible for a symmetry-protected zero-energy edge state exponentially localized on one or the other side of the defect, depending on the Zak phase of the Bloch wavefunctions. Because the prescription we propose to probe the Zak phase is resolved in energy, it should then be possible to probe both the topological properties of Bloch bands, as well as the existence of zero-energy edge states they lead to within the band gap. This would offer a unique op- portunity to observe the bulk-edge correspondence in a single experiment. The authors are very grateful to D. Carpentier for stimu- lating discussions. This work was supported by the French Agence Nationale de la Recherche (ANR) under grant Topo- Dyn (ANR-14-ACHN-0031). [1] J. Friedel, Phil. Mag. 43, 153 (1952). [2] M. A. Ruderman and C. Kittel, Phys. Rev. 96, 99 (1954). [3] T. Kasuya, Progress of theoretical physics 16, 45 (1956). [4] K. Yosida, Phys. Rev. 106, 893 (1957). [5] A. Fert and P. M. Levy, Phys. Rev. Lett. 44, 1538 (1980). [6] S. K. Adhikari, American Journal of Physics 54, 362 (1986). [7] V. E. Barlette, M. M. Leite, and S. K. Adhikari, Eur. J. Phys. [8] C. Dutreix and M. Katsnelson, Phys. Rev. B 93, 035413 (2016). [9] M. F. Crommie, C. P. Lutz, and D. M. Eigler, Nature 363, 524 21, 435 (2000). (1993). [10] P. T. Sprunger, L. Petersen, E. W. Plummer, E. Laegsgaard, and F. Besenbacher, Science 275, 1764 (1997). [11] L. Petersen, P. T. Sprunger, P. Hofmann, E. Laegsgaard, B. G. Briner, M. Doering, H.-P. Rust, A. M. Bradshaw, F. Besen- bacher, and E. W. Plummer, Phys. Rev. B 57, R6858 (1998). [12] C. Dutreix and P. Delplace, Supplemental Material. 5 [13] W. P. Su, J. R. Schrieffer, and A. J. Heeger, Phys. Rev. Lett. 42, 1698 (1979). [14] M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kita- gawa, E. Demler, and I. Bloch, Nature Physics 9, 795 (2013). [15] M. Xiao, G. Ma, Z. Yang, P. Sheng, Z. Zhang, and C. T. Chan, Nature Physics 11, 240 (2015). [16] J. M. Zeuner, M. C. Rechtsman, Y. Plotnik, Y. Lumer, S. Nolte, M. S. Rudner, M. Segev, and A. Szameit, Phys. Rev. Lett. 115, 040402 (2015). [17] C. Poli, M. Bellec, U. Kuhl, F. Mortessagne, and H. Schome- rus, Nature Communications 6 (2015). [18] I. Brihuega, P. Mallet, C. Bena, S. Bose, C. Michaelis, L. Vitali, F. Varchon, L. Magaud, K. Kern, and J. Y. Veuillen, Phys. Rev. Lett. 101, 206802 (2008). [19] P. Roushan, J. Seo, C. V. Parker, Y. S. Hor, D. Hsieh, D. Qian, A. Richardella, M. Z. Hasan, R. J. Cava, and A. Yazdani, Na- ture 460, 1106 (2009). [20] Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. Hor, R. Cava, et al., Nature Physics 5, 398 (2009). [21] J. Zak, Phys. Rev. Lett. 62, 2747 (1989). [22] D. Thouless, Phys. Rev. B 27, 6083 (1983). [23] R. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993). [24] R. Resta, Rev. Mod. Phys. 66, 899 (1994). [25] A. Raoux, M. Morigi, J.-N. Fuchs, F. Pi´echon, and G. Mon- tambaux, Phys. Rev. Lett. 112, 026402 (2014). [26] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. Ludwig, Phys. [27] P. Delplace, D. Ullmo, and G. Montambaux, Phys. Rev. B 84, Rev. B 78, 195125 (2008). 195452 (2011). [28] M. Rudner and L. Levitov, Phys. Rev. Lett. 102, 065703 (2009). SUPPLEMENTAL MATERIAL FOR "GEOMETRICAL PHASE SHIFT IN FRIEDEL OSCILLATIONS" Friedel oscillations for N = 2 This SM section details the derivations of the Friedel oscillations induced in the local density of states, as introduced in Eq. (1) and Eq. (4) in the main text. 6 Retarded bare Green's matrix Within a two-band nearest-neighbor tight-binding description, the off-diagonal component of the bare Green's matrix may be written as (cid:90) +π (cid:90) +π (cid:73) −π −π C dk 2π dk 2π dz 2iπ (α + eik) eikm ω2 − E2±(k) (α + eik) eikm ω2 − α + eik2 (α + eik) zm P(z) , G(0) BA(m, ω) = = = (10) (11) (12) where it is assumed that m ≥ 0, z = eik, P(z) = −α z2 + (ω2 − 1− α2) z− α, C denotes the unit circle, and energy is given in units of hopping amplitude t1. As we are interested in probing the bulk energy bands, which implies (1 − α)2 ≤ ω2 ≤ (1 + α)2, there are two complex roots for polynomial P, namely, (cid:112)(cid:2)(1 + α)2 − ω2(cid:3)(cid:2)ω2 − (1 − α)2(cid:3) ω2 − 1 − α2 ± i z± = (13) These roots satisfy z± = 1. Nevertheless, we have to consider a finite quasiparticle lifetime, which is achieved by adding a small imaginary part i
to the frequency ω. Moreover, we assume
> 0 to work with the retarded Green function. Thus, z+ > 1 and z− is the only pole that remains inside the unit circle. It is simply denoted z0 = eik0 from now on. Since z0 has a negative imaginary part, this implies −π < k0 < 0. Therefore, in the limit
→ 0+ the bare Green's function is given by 2α . BA(m, ω) = − i G(0) 2vk0 eik0m+iθk0 (14) = −α sin(k0) ω−1. For simplicity we focus where we have used that θk0 on the conduction bands (ω ≥ 0) from now on. Besides, we have to be careful with the sign of m which has been assumed to be positive so far. When m < 0, then we can use the variable change z = e−ik, which straightforwardly leads to = Arg[α + eik0], and ω2 = α + eik02, so that vk0 dk k0 = dω The other off-diagonal component of the bare Green's matrix is obtained in the same way. This results in BA(±m, ω) = − i G(0) 2vk0 eik0m+iθ±k0 . AB(±m, ω) = − i G(0) 2vk0 eik0m−iθ±k0 . (15) (16) When doing the substitution α + eik ↔ ω in the numerator of Eq. (10), we immediately obtain the real-space representation of the diagonal component of the bare Green's matrix, that is AA(±m, ω) = G(0) G(0) BB(±m, ω) = − i 2vk0 eik0m . (17) Note finally that this latter expression does not depend on wether there is one or several atoms per unit cell, so that the LDOS correction δρA(m, ω) it leads to will be the same for the monatomic and diatomic pattern crystals. T-matrix appraoch (cid:33) (cid:32) 1 0 0 0 , T(ω) = t(ω) 7 (18) The T-matrix satisfies (cid:82) (cid:112) where t(ω) = V [1 − V zone. From the real-space expression of the bare Green's functions introduced above, this can be rewritten as AA(k, ω)]−1 does not depend on wavevector k, due to the integration that runs over the Brillouin dk G(0) t(ω) = V 1 − VG(0) AA(0, ω) = −i 2vk0V (2vk0)2 + V2 (cid:112) eiτk0 , (19) where τk0 imaginary part of the following matrix product: G(0)(m, ω)T(ω)G(0)(−m, ω). It finally behaves as (2vk0)2 + V2]. The variation of local density of states on both sublattices is then obtained as the = Arg[(V + i2vk0)/ δρA(m, ω) = V(ω) cos(2k0m ± τk0) δρB(m, ω) = V(ω) cos(2k0m ± τk0 + ϕk0) . where the sign "±" refers to the positive and negative values of m, ϕk0 V = θk0 − θ−k0 and we have defined (cid:112) 1 2vk0 V(ω) = . (2vk0)2 + V2 (20) (21) As shown in Fig. 5 for α = 0.9 and V = 1, these analytical expressions of Friedel oscillations are in agreement with the numerical evaluation of the LDOS correction. They describe interferences that are symmetric with respect to the impurity site on sublattice A, and asymmetric on sublattice B due to the phase shift ϕk0. Importantly, the T-matrix, which does not depend on space coordinates, has two contributions. On the one hand, it modulates (2vk0)2 + V2]. the amplitudes of Friedel oscillations. On the other hand, it is responsible for the phase shift τk0 = Arg[(V+i2vk0)/ (cid:112) FIG. 5. (Color online) Energy-resolved interference patterns in the LDOS on sublattice A (top) and sublattice B (bottom) obtained numerically (left) and analytically (right). Only the scattering occurring in the conduction band is shown, i.e., 1 − α ≤ ω ≤ 1 + α where α = 0.9. The vertical white dashed line marks the site of the impurity that belongs to sublattice A and unit cell m = 0. The impurity potential is simulated by V = 1. Energy is given in units of the hopping amplitude t1. -1010m 02ω-11-1010m 02ω-11-1010m 02ω-11-1010m 02ω-11(cid:112) 8 This phase shift is obviously bounded (i.e. can be chosen as monovalued), since (V + i2vk0)/ (2vk0)2 + V2 cannot wind around the complex plane origin when varying k0. As we are interested in the global geometrical properties of the phase shifts in Friedel oscillations, it is sufficient to consider that the LDOS correction is given by δρA(m, ω) = cos(2k0m) 1 2vk0 1 2vk0 δρB(m, ω) = cos(2k0m + ϕk0) (22) in real space. These expressions actually correspond to the limits of strong impurity potential (V (cid:29) 1) or low energies (k0 ∼ ±π) in which τk0 vanishes. They already describe the 2k0-wavevector Friedel oscillations in the LDOS, as well as the global geometrical properties of the phase shift ϕk0, which are the two universal behaviors that the main text aims to describe. Therefore, the expressions provided in Eq. (22) are the ones discussed in the main text. Note that they are discussed with respect to numerical evaluations of the LDOS correction for V = 1 in the main text, which does not fall into the limits V (cid:29) 1 nor k0 ∼ ±π. Nonetheless, they show a very good agreement with each other, which confirms that the expressions in Eq. (22) already describes universal behaviors of Friedel oscillations. Ambiguity in the unit cell definition Here, we would like to point out an ambiguity in the definition of the unit cell. Fig. 6 depicts two copies of a dimerized crystal in the presence of a localized impurity. The latter fixes a natural origin, so that the impurity site unambiguously belongs to the unit cell m = 0 of sublattice A. In order to define the diatomic pattern, we then have to chose one the two nearest-neighbor sites as belonging to the unit cell m = 0. This choice leads to an ambiguity in the definition of the unit cell, hence the two configurations illustrated in the figure. Of course, the Friedel oscillations as introduced in Eq. (22) appear through an observable, namely, the LDOS, and they cannot depend on this choice of unit cell. This can also be understood explicitly. The top and bottom configurations in Fig. 6 respectively correspond to the following Bloch Hamiltonian matrices: (cid:32) (cid:33) (cid:32) H(1)(k) = 0 α + eik α + e−ik 0 and H(2)(k) = 0 e−ik(α + eik) eik(α + e−ik) 0 . (23) (cid:33) After reproducing the derivations of the bare Green's functions introduced above, we end up with the following expressions for the Friedel oscillations on sublattice B: (cid:104) (cid:105) (cid:105) B (±m, ω) ∝ cos(2k0m ± 2θ(1) δρ(1) k0 e−ik0(α + eik0) and θ(2) k0 = Arg (cid:104) ) where θ(1) k0 θ(2) = θ(1) k0 k0 α + eik0 = Arg − k0, so that the LDOS correction on sublattice B can be rewritten as and B (±m, ω) ∝ cos(2k0m ± 2θ(2) δρ(2) k0 ) , (24) . The two geometrical phase shifts are then related to one another via B (±m, ω) ∝ cos(2k0m ± 2θ(1) δρ(1) k0 ) and B (±m, ω) ∝ cos(2k0(m ∓ 1) ± 2θ(1) δρ(2) k0 ) . (25) FIG. 6. (Color online) Illustration of the two possible definitions of the unit cell. The site of the impurity (grey square) unambiguously belongs to the site A (blue disk) of the unit cell m = 0. Then, there are two nearest-neighbor sites B (red disks) and we have to pick one of them to obtain the diatomic pattern of the unit cell, which leads to the two configurations depicted in the figure. As it can be seen from these expressions, the Friedel oscillations in the LDOS are the same on every physical site and does not depend on the choice we made to label the unit cell. For example, the site B of unit cell m = 0 in the top configuration of Fig. 6, corresponds to the site B of unit cell m = 1 in the bottom configuration of Fig. 6. The LDOS correction associated to this site is unambiguously given by 9 (26) Besides, the site B of unit cell m = −1 in the top configuration of Fig. 6, corresponds to the site B of unit cell m = 0 in the bottom configuration of Fig. 6. The LDOS correction associated to this site is unambiguously given by ) . δρ(1) B (0, ω) = δρ(2) k0 B (+1, ω) ∝ cos(2θ(1) B (−1, ω) = δρ(2) δρ(1) B (0, ω) ∝ cos(2θ(2) k0 ) . (27) Consequently, the interference pattern we describe, namely the Friedel oscillations in the observable LDOS, does not depend on the way we define the unit cell. Fourier transform of Friedel oscillations The Friedel oscillations on sublattice B described by the general expression in Eq. (20) satisfies that we have introduced a normalized summation over n, so that it satisfies(cid:80) where −π ≤ k0 ≤ 0, and we have used the fact that z0 = eik0 lies inside the unit circle, before considering the limit
→ 0. Note = −τk0, this can be rewritten as n = 1. Since τ−k0 + n2π) e−iϕk0 , (28) where now −π ≤ k0 ≤ π and δρB(q, ω) is a 4π-periodic function of q. For sublattice A, this straightforwardly leads to (29) (30) m>0 (cid:88) (cid:88) (cid:88) (cid:88) m<0 m>0 n=−∞ m>0 δρB(q, ω) =V(ω) +V(ω) =V(ω) +V(ω) =V(ω) +V(ω) =V(ω) +V(ω) =V(ω) n=−∞ n=−∞ n=−∞ +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) +∞(cid:88) n=−∞ n=−∞ n=−∞ n=−∞ n=−∞ ei(2k0+q+n2π)m eiϕk0 eiτk0 + e−i(2k0−q+n2π)m e−iϕk0 e−iτk0 ei(2k0+q+n2π)m eiϕk0 e−iτk0 + e−i(2k0−q+n2π)m e−iϕk0 eiτk0 ei(2k0+q+n2π)m eiϕk0 eiτk0 + e−i(2k0−q+n2π)m e−iϕk0 e−iτk0 e−i(2k0+q+n2π)m eiϕk0 e−iτk0 + e+i(2k0−q+n2π)m e−iϕk0 eiτk0 1 1 − e−i(2k0−q+n2π) e−iϕk0 e−iτk0 1 − ei(2k0−q+n2π) e−iϕk0 eiτk0 1 − ei(2k0−q+n2π) 1 1 1 − ei(2k0+q+n2π) eiϕk0 eiτk0 + 1 − e−i(2k0+q+n2π) eiϕk0 e−iτk0 + 1 1 − e−i(2k0+q+n2π) 1 − cos(2k0 + q + n2π) 1 − ei(2k0+q+n2π) 1 − cos(2k0 + q + n2π) cos(τk0) − cos(2k0 + q − τk0 1 − cos(2k0 + q + n2π) eiϕk0 eiτk0 + eiϕk0 e−iτk0 + + n2π) e−iϕk0 e−iτk0 +n2π 1 − cos(2k0 − q + n2π) 1 − e−i(2k0−q+n2π) 1 − cos(2k0 − q + n2π) e−iϕk0 eiτk0 cos(τk0) − cos(2k0 − q − τk0 1 − cos(2k0 − q + n2π) eiϕk0 + δρB(q, ω) =V(ω) cos(τk0) − cos(2k0 − q − τk0 1 − cos(2k0 − q + n2π) + n2π) e−iϕk0 , δρA(q, ω) =V(ω) cos(τk0) − cos(2k0 − q − τk0 1 − cos(2k0 − q + n2π) + n2π) . +∞(cid:88) n=−∞ +∞(cid:88) n=−∞ The Fourier transform of the Friedel oscillations introduced in Eq. (22) is given by 10 (31) +∞(cid:88) +∞(cid:88) n=−∞ where −π ≤ k0 ≤ 0. This subsequently reduces to δρB(q, ω) = V(ω) δρA(q, ω) = V(ω) n=−∞ [δ(q + 2k0 + n2π) + δ(q − 2k0 + n2π)] [δ(q + 2k0 + n2π) e+iϕk0 + δ(q − 2k0 + n2π) e−iϕk0 ] , δρA(q, ω) = V(ω) δ(q − 2k0 + n2π) +∞(cid:88) +∞(cid:88) n=−∞ (32) where now −π ≤ k0 ≤ π, so that −2π ≤ q ≤ 2π. Note once again that this behavior in reciprocal space agrees with the one described in Eq. (30). The modulus is maximum for backscattering wavevectors q = 2k0 [2π], while the argument yields to the phase shift ϕk0. The expression above is the one mentioned in the main text in Eq. (6). δρB(q, ω) = V(ω) n=−∞ δ(q − 2k0 + n2π) e−iϕk0 . Generalization to N internal degrees of freedom Band structure under Chiral symmetry So far we have focused on the interference pattern induced in the LDOS when there are two sublattice degrees of freedom. It shows that, in the presence of space- and time-inversion symmetries, the interference pattern Fourier transform exhibits topo- logical properties that are intrinsically related to the winding number characterizing the one-dimensional Bloch band structure. This relies on the space- and time-inversion symmetries, which prevents any mass term scaling with σ3 in the two-band descrip- tion. Since we disregard the processes scaling with the identity matrix σ0, for they do not change the Bloch eigenstates, nor the topological properties they involve, our description relied on a bipartite Hamiltonian that had chiral symmetry. The Bloch band structure it is associated to then belongs to a chiral symmetry class, and the topological features it may exhibit are characterized by a winding number. From now on, we consider a Bloch Hamiltonian matrix, namely H(k), that satisfies a chiral symmetry: Γ†H(k)Γ = −H(k), where the chiral operator Γ is unitary and squares to plus identity. The chiral symmetry requires the eigenstates of H(k), namely n(k)(cid:105), to come in pairs with opposite energies: H(k)n(k)(cid:105) = En(k)n(k)(cid:105) H(k) Γn(k)(cid:105) = −En(k) Γn(k)(cid:105) . (33) This obviously yields a particle-hole symmetric spectrum that we assumed to be gapped, i.e., En(k) (cid:44) 0 for all k and n. Thus, there is an even number of energy bands, and we generically denote it 2N, where N is a natural number. Besides, we assume without loss of generality that the eigenstates n(k)(cid:105) (respectively Γn(k)(cid:105)) are normalized and associated to the negative (respectively positive) energy bands sorted in the ascending (respectively descending) order according to index n ∈ [1··· N]. Second, the chiral-conjugate eigenspaces are orthogonal to each other: (cid:104)n(k)Γn(k)(cid:105) = 0 for every n. Since Γ2 = +1, the chiral operator has two orthogonal subspaces of normalized states that have opposite chiralities: An(k)(cid:105) = Bn(k)(cid:105) = √ n(k)(cid:105) + Γn(k)(cid:105) n(k)(cid:105) − Γn(k)(cid:105) 2 √ with ΓAn(k)(cid:105) = +An(k)(cid:105) , ΓBn(k)(cid:105) = −Bn(k)(cid:105) . (34) In particular, chiral operator Γ supports a diagonal matrix representation of the form Γ = IN ⊗ σ3, where IN denote the N × N identity matrix and σz is the third Pauli matrix. For such a representation, the anticommutation relation that defines the chiral symmetry requires the Bloch Hamiltonian matrix to be off-diagonal: with 2 (cid:32) (cid:33) H(k) = 0 F†(k) F(k) 0 . (35) 11 In the basis {A1(cid:105)···AN(cid:105),B1(cid:105)···BN(cid:105)}, the off-diagonal block F(k) becomes diagonal, provided all energy bands En(k) are non- degenerate. Otherwise it becomes trigonal, but it would not change the topological properties we aim to highlight here, since they relie on the assumption that there exists a gap around the zero-energy level. It reads  F(k) = E1(k) e−iθ1(k) 0 ... 0 0 ··· . . . . . . . . . . . . ··· 0 EN(k) e−iθN(k) 0 ... 0  . (cid:73) ∇k ln dk 2iπ BZ N(cid:89) n=1 Det F†(k) Det F†(k) , N(cid:88) i n=1 (cid:73)  . eiθn(k) = exp θn(k) Det F†(k) Det F†(k) = N(cid:88) n=1 W = Wn where Wn = ∇kθn(k) . dk 2π BZ (36) (37) (38) (39) In one dimension, the winding number that characterizes the first homotopy group of spheres of the Bloch band structure, namely Π1(S 1) = Z, can be defined as where W = The expression of the winding number finally reduces to It clearly appears through this expression that the winding number may become ill-defined if the assumption En(k) (cid:44) 0 is relaxed. Real-space representation of the bare Green's functions The retarded bare Green's function can then be introduced as G(0)(k, ω) =[ω − H(k)]−1 n=1 2N(cid:88) N(cid:88) N(cid:88) n=1 n=1 = = = n(k)(cid:105)(cid:104)n(k) (cid:34)n(k)(cid:105)(cid:104)n(k) ω − En(k) (cid:34) ω − En(k) ω ω2 − E2 n(k) (cid:35) Γn(k)(cid:105)(cid:104)n(k)Γ† ω + En(k) + (An(cid:105)(cid:104)An + Bn(cid:105)(cid:104)Bn) + En(k) ω2 − E2 n(k) eiθn(k) An(cid:105)(cid:104)Bn + e−iθn(k) Bn(cid:105)(cid:104)An(cid:17)(cid:35) (cid:16) . (40) In the expression above, it has been used that An(k)(cid:105) = eiθA n (k) is the phase shift between the freedom degrees of opposite chiralities for the nth energy band. Note that the chiral symmetry requires the Green's function to satisfy the following relation: Γ†G(0)(k, ω)Γ = −G(0)(k,−ω). The matrix representation of the Green's function consists of uncoupled 2 × 2 blocks. Therefore, we can use the derivations already done in the previous sections of this Supplemental Material, and the Green's function can be written as n (k) Bn(cid:105), so that θn(k) = θA n (k) An(cid:105) and Bn(k)(cid:105) = eiθB n (k) − θB (cid:104)An(cid:105)(cid:104)An + Bn(cid:105)(cid:104)Bn + eiθn(k0) An(cid:105)(cid:104)Bn + e−iθn(k0) Bn(cid:105)(cid:104)An(cid:105) , (41) G(0)(m, ω) =ei2k0m N(cid:88) n=1 δ(ω ± En(k0)) 2ivn(k0) where m > 0, time-reversal symmetry requires the spectrum to satisfy En(k) = En(−k), vn(k0) = ∂kEn(k0), and k0 is assumed to be a pole satisfying ω = ±En(k0) and eik0 = 1. Of course, this pole depends on the band index n, but we omit it for more clarity. Here, a few remarks are in order about the Green's function expression above. • The poles satisfying eikα < 1 do not yield any long-range contribution, since they lead to exponential decays with the distance to the impurity, so they are disregarded. 12 • The poles satisfying eikα = 1 are assumed to be simple poles. They could be poles of higher orders, but it is always possible to make them simple pole by adiabatically varying the model parameters (e.g. hopping amplitudes), as long as the zero-energy gap does not close. This procedure does not change the winding number of the Bloch band structure W, nor the topological property of the geometrical phase shift we aim to highlight in the Friedel oscillations. • There may be several simple poles satisfying eikα = 1. Nevertheless, we can still consider continuously deforming the spectrum by tuning adiabatically the model parameters without closing the zero-energy gap, in such a way that each energy band becomes a monotonic function of the momentum for 0 ≤ k ≤ π. This would bring us to a spectrum similar to the one discussed in the main text with only one simple pole k0. Friedel oscillations in the LDOS Let us consider a localized potential in real space such that it leads to the following representation of the T-matrix in momentum space: N(cid:88) T(ω) = tn(ω)An(cid:105)(cid:104)An . where n=1 tn(ω) = −i (cid:112) 2vn(k0)Vn (2vn(k0))2 + V2 n eiτn(k0) , (cid:112) (42) (43) (44) where Vn is the potential experienced by the eigenstates of the nth band, and τn(k0) = Arg[(Vn + i2vn(k0))/ (2vn(k0))2 + V2 n ]. The T-matrix does not depend on k because the impurity is localized in real space. This leads to the following LDOS correction δρA(m, ω) = δρA,n(m, ω) = N(cid:88) N(cid:88) n=1 cos(2k0m + τn(k0)) δ(cid:0)ω + En(k0)(cid:1) cos(2k0m + τn(k0) + ϕn(k0)) δ(cid:0)ω + En(k0)(cid:1) , N(cid:88) N(cid:88) n=1 n=1 δρB(m, ω) = δρB,n(m, ω) = bands (i.e., there is no term with δ(cid:0)ω − En(k0)(cid:1)), and n=1 where ϕn(k0) = θn(k0) − θn(−k0) = 2θn(k0) under time-reversal symmetry. Besides, we have only considered the conduction Vn(ω) = Vn 2vn(k0) (cid:112) 1 (2vn(k0))2 + V2 n . The energy-resolved interference pattern in momentum space behaves similarly to cos(τk0) − cos(2k0 − q − τk0 1 − cos(2k0 − q + p2π) cos(τk0) − cos(2k0 − q − τk0 1 − cos(2k0 − q + p2π) δρA,n(q, ω) = δρB,n(q, ω) = δρB(q, ω) = δρA(q, ω) = Vn(ω) Vn(ω) p=−∞ N(cid:88) N(cid:88) n=1 N(cid:88) N(cid:88) n=1 +∞(cid:88) +∞(cid:88) p=−∞ n=1 n=1 + p2π) + p2π) δ(cid:0)ω + En(k0)(cid:1) e−iϕn(k0) δ(cid:0)ω + En(k0)(cid:1) . (45) (46) When moving along the q-periodic paths Cn defined by the constraint q = 2k0 within each energy band En, we finally access: (cid:73) N(cid:88) n=1 Cn (cid:34) δρA,n(q, ω) (cid:35) δρB,n(q, ω) = (cid:73) N(cid:88) n=1 Cn dq 2iπ ∂q ln (cid:73) N(cid:88) n=1 Cn N(cid:88) n=1 dq 2π ∂q ϕn(q/2) = 2 dq 2π ∂q θn(q) = 2 Wn = 2W . (47) 

1305.0314

2

1305

Power spectrum of electronic heat current fluctuations

We analyze the fluctuations of an electronic thermal current across an idealized molecular junction. The focus here will be on the spectral features of the resulting heat fluctuations. By use of the Green functionmethod we derive an explicit expression for the frequency-dependent power spectral density of the emerging energy fluctuations. The complex expression simplifies considerably in the limit of zero frequency, yielding the noise intensity of the heat current. The spectral density for the electronic heat fluctuations still depends on the frequency in the zero-temperature limit, assuming different asymptotic behaviours in the low- and high-frequency regions. We further address subtleties and open problems from an experimental viewpoint for measurements of frequency-dependent power spectral densities.

cond-mat.mes-hall

cond-mat

physica status solidi Power spectrum of electronic heat current fluctuations Fei Zhan1,2, Sergey Denisov1, and Peter Hanggi*1,3 3 1 0 2 l u J 6 2 ] l l a h - s e m . t a m - d n o c [ 2 v 4 1 3 0 . 5 0 3 1 : v i X r a 1 Institut fur Physik, Universitat Augsburg, Universitatsstr. 1, D-86159 Augsburg, Germany 2 Centre for Engineered Quantum Systems, School of Mathematics and Physics, The University of Queensland, St Lucia QLD 4072, Australia 3 Department of Physics and Centre for Computational Science and Engineering, National University of Singapore, Singapore 117546, Singapore Received XXXX, revised XXXX, accepted XXXX Published online XXXX Key words: molecular junction, heat current fluctuation, power spectral density ∗ Corresponding author: e-mail: [email protected], Phone: +49-821-5983249, Fax: +49-821-5983222 We analyze the fluctuations of an electronic thermal cur- rent across an idealized molecular junction. The focus here will be on the spectral features of the resulting heat fluctuations. By use of the Green function method we de- rive an explicit expression for the frequency-dependent power spectral density of the emerging energy fluctua- tions. The complex expression simplifies considerably in the limit of zero frequency, yielding the noise inten- sity of the heat current. The spectral density for the elec- tronic heat fluctuations still depends on the frequency in the zero-temperature limit, assuming different asymp- totic behaviors in the low- and high-frequency regions. We further address subtleties and open problems from an experimental viewpoint for measurements of frequency- dependent power spectral densities . ε0 Gate TL TR (Color online) Sketch of a molecular junction setup used in the text. The average heat flow is generated by electrons moving from a hot electrode TL across the molecular junction towards a neighboring cold electrode TR. The inter-electrode electronic level ε0 can be tuned continuously. Copyright line will be provided by the publisher 1 Introduction The experimental activities over the last fifteen years in investigating electronic transport across molecular junctions [1,2,3] have triggered several waves of intense research in theory [4,5,6,7,8] and experiment [10,11,12,13]. Single molecule electronics is still con- sidered as a promising candidate for the substitution of silicon-based elements in the information processing tech- nology [2,3,10,11]. Likewise, molecular junctions have advantages in the context of energy-related applications. This is due to the potential of hybrid solid-state molecular structures which enable novel interface features. Moreover, the abundant selection of possible molecules and electrode materials allow to tailor specific properties. In particular, the topic of thermoelectric [8] and photovoltaic [9] conver- sion processes continue to prompt timely research in the field of molecular electronics. Apart from the standard current-voltage characteris- tics [1,12,13], it is also important to obtain insight into the fluctuations that accompany the corresponding trans- port processes. For example, by use of the full counting statistics [14,15,16,17] it is possible to extract information about the fluctuations of the electric current flowing across a molecular wire [18,19,20,21,22]. Copyright line will be provided by the publisher 2 First author et al.: Short title In the context of thermoelectric applications, the issue of energy transport through molecular junctions and the properties of the corresponding fluctuations acquire spe- cial importance. Thermal fluctuations may crucially impact the electronic transport features, and even affect the overall performance of the molecular junction. With the molecular systems operating on the nanoscale corresponding energy current fluctuations can become sizable. This may be so even in situations where the average energy current is van- ishing identically, as it is the case in thermal equilibrium with both interconnecting electrodes held at the same tem- perature. Moreover, the properties of nonequilibrium noise correlations, or likewise, its frequency-dependent spectral properties, are in no obvious manner related to the mean value of the energy flow itself. With this work we shall explore the fluctuations of the heat current caused by the transferring electrons. Our goal is to obtain analytical es- timates for the power spectral density (PSD) of the heat fluctuations, even at the expense that these may mainly ap- ply to idealized setups only. With such a restriction these analytical results may nevertheless be useful to appraise the role of heat current noise in more realistic molecular junctions. It is further of interest to have an estimate avail- able when devising molecular circuitry for more complex tasks. Energy transport across a molecular structure which links two electrodes is induced by a difference of the two electrode temperatures, see in Fig. 1. The physics of heat transfer generally involves both electrons and phonons and their mutual interaction [8,23,25,27,28,29,30,31,32,33, 34]. Therefore, the amount of energy flow carried across the wire should be addressed with care, with the need to distinguish between energy transfer mediated either by electrons or phonons, or a combination of both. If phonons are mainly at work this situation relates to the new field of phononics [35], a novel research area which may lead to new circuit elements, such as molecular thermal diodes, thermal transistors, thermal logic gates, to name but a few [35,36,37,38,39,40,41]. Then, the size of fluctuations in heat current does matter; this is so because those may well turn out to be deleterious to intended information process- ing tasks. Heat transport mediated by electrons relates at the same time to charge transfer: electrons moving from lead-to-lead carry not only charge but also energy [24,25,26]. How- ever, the amount of energy transferred by a single elec- tron, unlike to its charge, is not quantized [42]. In con- trast to those studies that examine the average heat flow, however, much less attention has been paid to the issue of fluctuations of the accompanying flow of energy. In prior work [43] the energy transport through a ballistic quan- tum wire has been considered in the Luttinger-liquid limit, by neglecting the discreteness of the wire's energy spec- trum. Likewise, with Refs. [44], the PSD of the heat current fluctuations has been derived within a scattering theory ap- proach, using the assumption that the electrons are trans- Copyright line will be provided by the publisher mitted (reflected) at the same rate, independently of their actual energies. The results of the last two papers, how- ever, are challenging because it has been shown therein that the noise characteristics of heat current at equilibrium ex- hibits a well-pronounced frequency dependence even at ab- solute zero-temperature. Therefore, this very zero temper- ature finding is in contradiction with the naive expectation as provided by the equilibrium fluctuation-dissipation the- orem (FDT). This found deviation from the FDT in those works is attributed loosely to the role of zero-point-energy fluctuations [44]. With this work we shall consider the electronic energy current that proceeds across a molecular wire composed of a single energy level with the two electrodes held at differ- ent temperatures. A preliminary short discussion of such electronic nonequilibrium heat noise has been presented by us with Ref. [45]. Here, we complement and extend this study and present further useful details on the theoret- ical derivation of the noise expression. Moreover, we dis- cuss the nonequilibrium heat noise of the corresponding heat current over much broader parameter regimes and fre- quency regimes away from the zero-frequency limit. With our setup we also corroborate the results obtained in the zero temperature limit for the power spectral density at fi- nite frequencies for a different setup in Ref. [44]. In addi- tion, we address several subtleties when it comes to the ex- plicit validation of our theoretical findings by experimental means. 2 Molecular junction setup In order to obtain analytical tractable expressions we shall neglect electron-phonon interactions and, as well, electron- electron interactions. Such a simplification can be justified for tailored situations that involve a very short wire only. Then, the Coulomb interaction via a double occupancy shifts the energy far above the Fermi level so that its role in thermal transport can be neglected. Likewise, the elec- tron dwell time is short as compared to the electron-phonon relaxation time scale. Note however, that in contrast to pre- vious works [44], we account here for the dependence of the transmission coefficient on its electron energies, and, within the Green function approach [6,34], derive an ex- plicit expression for the PSD of the heat current fluctua- tions, Sh(ω). In particular we demonstrate below that the net noise features of the heat current are quite distinct from their electronic counterpart. Our molecular junction setup is depicted with Fig. 1: It is described by a Hamiltonian H = Hwire + Hleads + Hcontacts . (1) It contains three different contributions, namely the wire Hamiltonian, the leads and the wire-lead coupling, respec- tively. We consider here the regime of coherent quantum transport whereby neglecting dissipation inside the wire. The wire is composed of a single orbital; i.e., Hwire = ε0d†d , (2) pss header will be provided by the publisher 3 at an energy ε0, with the fermionic creation and annihila- tion operators, d† and d. The energy level ε0 can be tuned by applying a gate voltage. Our idealized setup allows for explicit analytical calculations. Physically, it mimics a dou- ble barrier resonant tunneling structure GaAs/AlxGa1−x- structure of the type considered for electronic shot noise calculations in Ref. [46], herein truncated to a single Lan- dau level. As commonly implemented, the electrodes are modeled by reservoirs, composed of ideal electron gases, i.e., Hleads =Xℓq εℓqc† ℓqcℓq , (3) where the operator c† ℓq(cℓq) creates (annihilates) an elec- tron with momentum q in the ℓ =L (left) or ℓ =R (right) lead. We assume that the electron distributions in the leads are described by the grand canonical ensembles at the tem- peratures TL/R and with chemical potentials µL/R. Using such ideal electron reservoirs we obtain hc† ℓqcℓ′q′ i = δℓℓ′δqq′ fℓ(εℓq) , where fℓ(εℓq) =he(εℓq −µℓ)/kBTℓ + 1i−1 denotes the Fermi function. (4) (5) We impose a finite temperature difference ∆T = TL − TR and use identical chemical potentials, µL = µR = µ for the electrodes. When an electron tunnels out from a lead, the energy E is transferred into the wire which presents the heat transfer, δQ. Observing the value for the chemical potential, µ, it reads δQ = (E − µ). In the following we use that all the electron energies are measured from the chemical potential value µ, being set at µ = 0. The Hamiltonian which describes the tunneling events reads: Hcontacts =Xℓq Vℓqc† ℓqd + h.c. . (6) ΓL ΓR µ ε0 µ TL > TR Figure 1 (color online) Idealized setup of a molecular junction used in text: Two metal leads, each filled with an ideal electron gas, are connected by a single orbital ε0. The coupling strengths are determined by ΓL/R. The left lead is prepared at a higher temperature as compared to the oppo- site right lead, i.e. TL > TR. The chemical potential, µ, is the same for both leads so that no electric current due to a finite voltage bias is present. This part mediates the coupling between the wire and the electrodes. Here, the notation h.c. denotes Hermitian con- jugate. The quantity Vℓq is the tunneling matrix element, and the tunneling coupling is characterized in general by a spectral density, Γℓ(E) = 2πXq Vℓq2δ(E − εℓq) . (7) In the following, we shall use a wide-band limit of the electrode conduction bands, setting Γℓ(E) := Γℓ. 3 Power spectral density of electronic heat cur- rent fluctuations Working within the Heisenberg description of operators we present the detailed derivation of the electronic energy cur- rent induced by a finite temperature difference of the two leads and the PSD of the corresponding energy fluctua- tions. We limit the consideration to pure energy transfer that proceeds in absence of a finite voltage bias across the two leads and no particle concentration across the leads. Put differently, no cross-phenomena of energy transfer due to a charge current (i.e. no Joule heating) or due to a parti- cle concentration current (i.e. no Dufour effect) is at work. Therefore, because all other channels for the energy trans- port between the leads are then explicitly excluded form our consideration, we follow previous works, e.g. see in Refs. [24,44], and use throughout this study the term 'heat current' as synonym for energy current. The electronic thermal current then reads J h L(t) = P δQ(t) ∆t . (8) With our choice of chemical potentials µL = µR = 0, we find that the heat transfer operator is δQ(t) = EL, with the energy operator given by εLqc† LqcLq . EL =Xq (9) Its time derivative thus yields the operator for the heat flux, reading: J h L(t) = −Xq 2εLq ¯h Im[VLqc† Lq(t)d(t)] . (10) The heat current is positive valued when heat transport pro- ceeds from the hot left lead, i.e. TL > TR to the adjacent cold lead, see in Fig. 1. In deriving the above expression we have employed the Heisenberg representation for the lead electron operators. The average current is obtained by the L(t)i. Because there are no electron ensemble average hJ h sinks and sources in between the leads we have hJ h L(t)i = R(t)i. We henceforth focus on the quantities derived −hJ h with regard to the left lead; i.e., hJ h L(t)i := hJ h(t)i. Copyright line will be provided by the publisher 4 First author et al.: Short title ∆J h L(t) = J h L(t) − hJ h L(t)i . (12) ξℓ(t) = − The quantum correlation function of heat current fluc- tuations is described by the symmetrized auto-correlation function, i.e., Sh(t, t′) = L(t), ∆J h 1 2(cid:10)[∆J h L(t′)]+(cid:11) , (11) with respect to the operator of the heat current fluctuation The heat current noise is described with τ = t − t′ by the symmetrized quantum auto-correlation function Sh(τ ) = 1/2h[∆J h ℓ (τ ), ∆J h ℓ (0)]+i , (13) ℓ (s) − ℓ (s)i, where the anti-commutator [A, B]+ = AB + BA of the heat current fluctuation operator ∆J h hJ h ensures the Hermitian property. ℓ (s) = J h With this work we throughout consider the asymptotic long time limit t → ∞ when all transients are decayed. In this asymptotic limit the average heat current is stationary and the auto-correlation function of the heat current fluc- tuations becomes time-homogeneous; i.e. it is independent of initial preparation effects. It thus depends on the time difference τ = t − t′ only. The Fourier transform yields the power spectral density (PSD) Sh(ω) for the heat cur- rent noise, i.e., Sh(ω) = Sh(−ω) =Z ∞ −∞ dτ eiωτ Sh(τ ) ≥ 0 . (14) Sh(ω) is an even function in frequency and strictly semi- positive, in accordance with the Wiener-Khintchine theo- rem [47]. In the following we address positive values of the frequency, ω > 0, only. The annihilation operators of the electrode states sat- isfy the Heisenberg equations of motion; i.e., cℓq(t) = − i ¯h εℓqcℓq(t) − i ¯h Vℓqd(t) , (15) yielding the solution cℓq(t) =cℓq(t0)e−iεℓq (t−t0)/¯h − iVℓq ¯h Z t t0 dt′e−iεℓq(t−t′)/¯hd(t′) . (16) Here, the first term on the right hand side describes the dy- namics of the free electrons in the leads, while the second term accounts for the influence of the molecule. The Heisenberg equation of the molecular annihilation operator is given by d(t) = − i ¯h ε0d(t) − i ¯hXℓq V ∗ ℓqcℓq(t). (17) Copyright line will be provided by the publisher Upon inserting Eq. (16) into Eq. (17), we obtain d = i ¯h ε0d(t) − ΓL + ΓR 2¯h d(t) + ξL(t) + ξR(t), (18) where we have defined the noise operator i ¯hXq V ∗ ℓq exp(cid:20)− i ¯h εℓq(t − t0)(cid:21) cℓq(t0). (19) In addition, we have employed the definition (7) and used the wide-band limit. The noise quantity defined in Eq. (19) denotes operator- valued Gaussian noise, which is characterized by its mean and correlation properties, reading hξℓ(t)i = 0 hξ† ℓ′ (t′)ξℓ(t)i = δℓℓ′Z ∞ −∞ dε (20) 2π¯h2 e−iε(t−t′)/¯hΓℓ(ε)fℓ(ε). (21) This noise accounts for the influence of the states stem- ming from the electrodes l = L, R. Now the central problem is to solve the inhomogeneous differential equation (17). Once we obtain the solution of Eq. (17), we obtain also the solution for Eq. (16), the heat current (10) and also the power spectral density in Eq. (14). To obtain the solution of Eq. (18), we follow the Green function approach in Ref. [6] and start with solving the fol- lowing differential equation ( d dt + iε0 ¯h + ΓL + ΓR 2¯h )G(t − t′) = δ(t − t′) , (22) followed by the application of the convolution d(t) = R G(t − t′)(ξL(t′) + ξR(t′))dt′. The solution of Eq. (22) is thus given by: G(t) = θ(t)e−iε0t/¯h−(ΓL+ΓR)t/2¯h . (23) Then, the molecular operator in Eq. (18) assumes the form d(t) =Xℓq V ∗ ℓq exp[−iεℓq(t − t0)/¯h] εℓq − ε0 + i(ΓL + ΓR)/2 cℓq(t0) . (24) In what follows we address solely the asymptotic prop- erties which are reached with the initial time of prepara- tion t0 → −∞. This implies that average currents assume stationary values and correlation functions become time- homogeneous. With this expression and its Hermitian con- jugate, we obtain the occupation value of the molecular en- pss header will be provided by the publisher 5 8 4 0 2 1.5 1 0.5 ) W 3 1 − 0 1 ( h J ) z H / 2 W 4 3 − 0 1 ( h S (a) (b) TL 6= TR TL = TR TL 6= TR TL = TR −1 0 1 ε0 (meV) Figure 2 (color online) (a) Average electronic heat cur- rent J h and (b) the zero-frequency values of correspond- ing heat noise power Sh of the accompanying heat current fluctuations as a function of orbital energy for an iden- tical lead coupling strength Γ = 0.1 meV. The param- eters are: TL = 5.2 K, TR = 3.2 K (solid lines) and TL = TR = 4.2 K (dashed line). Figure in parts adapted from Ref. [45]. ergy level ε0 as nε0 = hd†(t)d(t)i Vℓq exp[iεℓq(t − t0)/¯h] [εℓq − ε0 − i(ΓL + ΓR)/2] = Xℓℓ′qq′ × V ∗ ℓ′q′ exp[−iεℓ′q′ (t − t0)/¯h] [εℓ′q′ − ε0 + i(ΓL + ΓR)/2] hc† ℓq(t0)cℓ′q′ (t0)i Vℓq2fℓ(εℓq) (εℓq − ε0)2 + (ΓL + ΓR)2/4 , (25) =Xℓq where we have employed the ensemble average, Eq. (4). We find that this occupation is determined by the Fermi function of the leads, weighted by the tunneling matrix elements Vℓq and the difference between lead states and the molecular energy level ε0, see in Eq. (25). This oc- cupation value is time-independent because there are no time-dependent external fields present. Upon substituting the result in Eq. (24) into Eq. (16), we find for the operators in the electrodes cℓq(t) = cℓq(t0)e−iεℓq (t−t0)/¯h VℓqV ∗ ℓ′q′ e−iεℓ′ q′ (t−t0)/¯h εℓ′q′ − ε0 + i(ΓL + ΓR)/2 cℓ′q′ (t0) +Xℓ′q′ × B[εℓ′q′ − εℓq] , where, B(E) = P(cid:18) 1 E(cid:19) − iπδ(E) , (26) (27) and P denotes the integral principal value. In going from Eq. (24) to Eq. (26) we have used Sokhotsky's formula which states that limǫ→0 1/(x + iǫ) = P(1/x) − iπδ(x), 0+ dx/x, see in Ref. [48]. where P(1/x) =R 0− −∞ dx/x +R ∞ Next we insert Eq. (24) and Eq. (26) into the heat cur- rent operator, Eq. (10), and by consequently taking the en- semble average, we obtain a Landauer-like formula for the heat current, reading [8,24,25,26,33,51]: hJ h(t)i := J h = 1 2π¯hZ dEET (E)[fL(E) − fR(E)] , (28) where the transmission coefficient T (E) = ΓLΓR/[(E − ε0)2 + Γ 2] , (29) is energy-dependent. The expression for the thermoelectric charge cur- rent [25] reads very similar to Eq. (28), except for its absence of the energy multiplier E in the integral on the rhs of Eq. (28). This seemingly small difference changes, however, the physics of transport through the wire, be- cause the multiplier inverts the symmetry of the integral. Namely, the thermolelectric current is an antisymmetric function of orbital energy and vanishes when the orbital energy level is aligned to the chemical potentials of the leads [45], while the heat current is a symmetric function and acquires a nonzero value at ε0 = 0, see in Fig. 2(a). 3.1 Main result and discussion Upon combining Eq. (14) and Eq. (10), we end up after a cumbersome evaluation with the nontrivial expression for the PSD of electronic heat current noise. Due to the com- plexity of this resulting expression the physics it inherits is not very illuminative. Nevertheless, we depict it here as Copyright line will be provided by the publisher 6 First author et al.: Short title ) z H / 2 W 5 3 − 0 1 ( h S 104 102 100 10−2 10−4 1 10 100 1000 ω (1011 Hz) Figure 3 Power spectral density of the heat current noise as a function of the frequency ω at temperatures TL = 6 K, TR = 2 K. The other parameters are ε0 = 0 and Γ = 0.1 meV. given in our preliminary report [45], reading: Sh(Ω = ¯hω; TL, TR) Ω =X± Z dE 2(cid:19)2 T (E)T (E ± Ω) 4π¯h("(cid:18)E ± [(E − ε0)2 + Γ 2] [(E ± Ω − ε0)2 + Γ 2] # L [E(E − ε0) − (E ± Ω)(E ± Ω − ε0)]2 Γ 2 + × fL(E)f L(E ± Ω) +(cid:18)E ± +(cid:20)(cid:18)E ± Ω 2(cid:19)2 2(cid:19)(cid:18)± Ω + E2R(E)T (E ± Ω) ∓ × fL(E)f R(E ± Ω) T (E)T (E ± Ω)fR(E)f R(E ± Ω) Ω 2(cid:19) Γ 2 LT (E ± Ω) (E − ε0)2 + Γ 2 1 2 EΩT (E)T (E ± Ω)(cid:21) +(cid:20)(E ± Ω)(cid:18)± + (E ± Ω)2 R(E ± Ω)T (E) Ω 2(cid:19) T (E)T (E ± Ω) 2(cid:19) Γ 2 (E − ε0)2 + Γ 2(cid:21) LT (E ± Ω) Ω +(cid:18)E ± Ω 2(cid:19)(cid:18)∓ × fR(E)f L(E ± Ω) , (30) wherein we abbreviated Ω ≡ ¯hω, f ≡ 1 − f , and R(E) ≡ 1 − T (E) denoting the reflection coefficient. Below we consider the case of symmetric coupling between the wire and the leads, ΓL = ΓR = Γ . We emphasize here that this heat PSD is a manifest nonequilibrium result where with a finite temperature bias the result accounts for 'heat'-shot noise and, simultane- ously nonequilibrium, Nyquist-like heat noise. Let us next discuss, via graphical means, some general features of the inherent complexity as depicted with Eq. (30). Copyright line will be provided by the publisher In Figure 3, we depict the dependence of the PSD of heat current fluctuations versus frequency ω at finite tem- perature bias, given by TL = 6 K, TR = 2 K. We deduce from the figure that this nonequilibrium PSD exhibits dif- ferent power laws in different frequency regions and grows with increasing frequency. Moreover, we find that the spectral density strength Γ of the wire-lead coupling can change the dependence of the heat fluctuation PSD on the parameters. In Figure 4 we depict the PSD as a function of the temperature difference ∆T over a wide regime of ∆T = 40 K, both in the case of weak and strong wire-lead couplings. With weak coupling, the PSD is smaller by one order of magnitude and only weakly (i.e. with a small slope) increases with ∆T , see Fig. 4(a). In contrast, the PSD increases very fast with ∆T when the coupling is very strong, see in Fig. 4 (b). Accord- ing to Eq. (29), the transmission coefficient becomes wider when Γ is larger, such that more electrons, whose energies deviate stronger from the chemical potential, are allowed to transport across the molecular junction. Therefore, the PSD becomes strongly enhanced and depends sensitively on ∆T . It is striking that both dependencies are near perfectly linear over the wide temperature regime of ∆T . Given the complex structure of the nonlinear nonequilibrium PSD detailed with the lengthy expression in (30) such extended linearity can hardly be expected a priori. The mechanism behind this distinctive feature is not evident and thus con- stitutes an interesting issue for further studies. 3.2 Issues relating to experimental validation It should also be mentioned here that the explicit verifi- cation of quantum mechanical power spectral densities is experimentally not at all straightforward. In clear contrast to the classical case, the symmetrized quantum correlation for heat in Eq. (13) presents no manifest quantum observ- able that can be measured directly, but rather it is a func- tional operator expression involving the time-evolution of the dynamics. This is so because the heat flux operators at different times do not commute. In fact, a quantum me- chanical evaluated PSD can be measured only indirectly via a single-time measurement of a tailored linear response function, via a corresponding, generally nonequilibrium quantum fluctuation-dissipation relation which connects this response function with a corresponding quantum me- chanical two-time correlation expression [47]. Put differ- ently, this tailored response function is then required to re- late precisely to our so calculated nonequilibrium quantum correlation of heat fluctuations in (13). This is so because a direct two-time quantum measurement of two observables at different times t would then impact (i.e. it will generally alter) the a priori theoretically determined quantum two- time correlation expression in (14); for further details and similar pitfalls see also in Refs. [49,50], where the problem of measuring quantum work poses the same challenge. pss header will be provided by the publisher (a) (b) ) z H / 2 W 5 3 − 0 1 ( h S 1.68 1.64 0 7800 7500 7200 40 0 7 40 ∆T (K) ∆T (K) Figure 4 Power spectral density of heat current noise at frequency ω = 2.16 × 1013 Hz (which is the Debye cut-off frequency of gold) as a function of temperature difference with (a) weak molecule-wire coupling Γ = 0.1 meV or (b) strong molecule-wire coupling Γ = 10 meV. The other employed parameters are TR = 300 K and ε0 = 0. The situation becomes more promising when we focus on the zero-frequency result of the PSD for heat noise: The variance < ∆Q2(t) > of the accumulated heat fluctuation over a time span t reads < ∆Q2(t) >=<(cid:16)Z t 0 ds∆J h(s)(cid:17)2 > (31) Using a long measurement time span t the time- dependent expectation value then relates to the zero fre- quency component of the PSD. This is the case upon noting that the symmetrized correlation is a symmetric function of its argument and assuming that the time-homogeneous auto-correlation of stationary heat fluctuations vanishes in sufficiently strong a manner for infinite time. Then, the integral in (31) can be extended to infinity, yielding lim t→∞ < ∆Q2(t) > /t =Z ∞ −∞ dτ Sh(τ ) = Sh(ω = 0). (32) The result for the zero-frequency limit therefore relates to a single time measurement of the manifest quantum observable ∆Q2(t). Still to measure accumulated 'heat' rather than 'heat-flux' presents a formidable challenge for the experimenter; the case with accumulated electric charge is a lot easier accessible. The detailed behavior of this zero-frequency nonequilibrium heat noise PSD will be studied next. 3.3 Zero frequency noise power The theoretical PSD of heat current noise at zero frequency ω = 0 simplifies considerably, assuming the appealing form Sh(ω = 0; TL, TR) = 1 2π¯hZ dEE2{T (E)(fL(E)[1 − fL(E)] + fR(E)[1 − fR(E)]) + T (E)[1 − T (E)][fL(E) − fR(E)]2} . (33) Here the last line refers to a heat-shot-noise contribu- tion while the first part corresponds to a nonequilibrium Nyquist-like heat noise contribution. Matters simplify con- siderably in thermal equilibrium where the shot noise con- tribution vanishes identically. Let us also briefly contrast this result with the zero- frequency PSD of the fluctuations displayed by the non- linear, accompanying thermoelectric current. The latter reads [4,6]: = Sel(ω = 0; TL, TR) e2 2π¯hZ dE{T (E)(fL(E)[1 − fL(E)] + fR(E)[1 − fR(E)]) + T (E)[1 − T (E)][fL(E) − fR(E)]2} , (34) Most importantly, the zero-frequency PSD for heat cur- rent in Eq. (33) differs by the energy factor E2 within the integrand. Although this distinction seemingly appears minor and may even be guessed beforehand without go- ing through the laborious task of doing a theoretical rig- orous derivation from which this limit derives from the frequency-dependent main result given in Eq. (30). It must be emphasized, however, that the two expressions lead to tangible differences. Particularly, note the different behav- ior of the electronic and heat noise PSDs versus the tunable energy level ε0 as depicted with Fig. 2 (b) and in Fig. 5. While the zero-frequency component of the electric PSD at ω = 0 exhibits a maximum at ε0 = 0, see in Fig. 2(c) in Ref. [45], its heat current PSD possesses instead a lo- cal minimum at this value, see Fig. 2(b). These two PSDs for charge current and heat current are compared in Fig. 5 over wide regimes of the electronic orbital energy ε0 and the lead-molecule strength Γ . These differences originate from the salient feature that the two transport mechanisms for charge and the en- Copyright line will be provided by the publisher 8 First author et al.: Short title 1.2 0.8 0.4 ) z H / 2 W 3 3 − 0 1 ( h S 0 1.5 ε 1 0.5 0(meV) −0.50 −1.5−1 0.4 0.2 1 0.8 0.6 ( m e V ) Γ 4 2 ) z H / 2 A 7 2 − 0 1 ( l e S 0 1.5 1 0.5 −0.50 ε0(meV) −1.5−1 0.4 0.2 1 0.8 ( m e V ) 0.6 Γ Figure 5 (color online) Power spectral density of the heat current noise at zero frequency ω = 0, (left panel) and power spectral density of the electric current noise (right panel) as functions of the wire orbital site energy ε0 and wire-lead coupling strength Γ . The parameters employed are TL = 6.2K and TR = 2.2K. ergy are different. The electric current is quantized by the electron charge e while, in contrast, the energy carried by the electron is continuous and can assume principally an arbitrary value. Notably, the main contribution to the electronic noise power across the wire stems from those electrons occupying energy levels around the chemical po- tential µ = 0. When ε0 deviates from the chemical poten- tial, increasingly less electrons participate in the transport. The flow of electron becomes diminished, and since both, the electric current and the electric noise are insensitive to the electron kinetic energies, they both decrease with increasing ε0. This scenario differs for heat flow: There, the deviation from the chemical potential increases the possibility that successive electrons will carry different en- ergies. This in turn causes an increase of heat current noise. With further deviation of the orbital energy from the chem- ical potential, the occupancy difference [fL(E) − fR(E)] decreases monotonically; consequently the noise power Sh(ω = 0) decreases again. 3.4 Electronic heat current noise in thermal equi- librium Next, let us focus on thermal equilibrium which is attained when the two temperatures are set equal, i.e. if TL = TR. In this case the average heat current vanishes identically, while its fluctuations remain finite. The zero-frequency spectra of both noise spectra for heat and electric current noise increase upon increasing the coupling strength Γ . This is so because the transmission probability increases. The corresponding heat noise power is nonzero in equilib- rium, however, as depicted with Fig. 2(b). In thermal equilibrium with TL = TR = T the nonequilibrium zero frequency PSD in Eq. (33) simpli- fies further, obeying Sh(ω = 0, T ) = 2kBT 2 Gh(ω = 0) , (35) where Gh(ω = 0) denotes the static, linear heat con- ductance, obtained from expanding the result in Eq. (28) Copyright line will be provided by the publisher around a small temperature bias and comparing with Eq. (33). This result is therefore in agreement with the fluctuation-dissipation-theorem (FDT) for the static heat conductance. Note that an extension to a Green-Kubo-like, but now frequency dependent conductance, however, would intrin- sically require also intermediate time-varying temperatures T (t). Such a concept with a time-dependent, nonequilib- rium temperature, cannot be justified in the coherent quan- tum regime of an open system with only one level ǫ0 con- necting the two leads. In fact even for a different setup with a spatially extended intermediate thermal conductor it has been found in Ref. [44] that at finite frequencies ω the PSD is not related to the corresponding linear heat conductance in the ballistic, low temperature transport regime. This vio- lation of the FDT is thus far from being fully settled in the literature. The properties at zero absolute temperature, TL = TR = 0, become even more subtle. Here, the heat current PSD at finite frequencies ω still depends on frequency. This dependence originates from quantum fluctuations where virtual transitions of electrons from lead-to-lead occur [44]. The Fermi distribution equals the Heaviside step function in this case. Therefore, the contributions to the integrand in Eq. (30) stems from the interval [−Ω, 0]. After an integration of Eq. (30), one finds for the frequency dependent PSD the expression: Sh(ω, TL = TR = 0) Γ = 4π¯h(cid:26)(cid:2)(2Ω)2 − 2Γ 2(cid:3) arctan(cid:18) Ω Γ (cid:19) + 2 ΩΓ "1 + log (Ω2 + Γ 2)2!#) , Ω ≡ ¯hω. Γ 4 In the limit Γ → ∞ the zero-temperature PSD scales like Sh(ω) ∝ ω3. This is in full agreement with results obtained in the work [44] for a different setup, where such (36) pss header will be provided by the publisher 9 an asymptotic behavior is found uniformly throughout the whole frequency region. This uniform feature no longer holds true when Γ is finite: The second term in the rhs of (36) introduces a linear cutoff in the limit ω → 0, so that [ Sh(ω) − Sh(ω = 0)] ∝ ω in the extreme low frequency limit. In distinct contrast, in the high-frequency region, the first term in the rhs of (36) becomes dominat- ing. As a consequence, the PSD (36) in the high frequency limit approaches a square-law asymptotic crossover depen- dence, Sh(ω) ∝ ω2. 4 Conclusions and sundry topics By using the Green function formalism we have investi- gated electronic heat current. Our focus centered on the is- sue of the heat current fluctuations in a molecular junction model composed of a single orbital molecular wire. For the noninteracting case we succeeded in deriving a closed form for the frequency dependence of heat current noise; i.e. the heat noise PSD, both in nonequilibrium TL 6= TR and in thermal equilibrium TL = TR. The dependence of the heat current noise on the orbital energy ε0 is qualitatively differ- ent from that for the accompanying electric current noise, see Fig. 5. Moreover, the heat current fluctuation proper- ties depend strongly on the the overall tunneling coupling strengths ΓL = ΓR = Γ . In the zero-temperature limit, the PSD of the heat cur- rent noise obeys two distinctive asymptotic behaviors, be- ing different in the intermediate-low frequency and in the high-frequency regimes. The particular square-law behav- ior of the PSD in the high-frequency region is due to the Lorentzian shape of the transmission coefficient T (E) in Eq. (29). Yet, the general effect would remain for any choice of the coefficient in the form of a localized, bell- shaped function: the noise spectrum will deviate from a cu- bic power-law asymptotic behavior upon entering the high- frequency region. As emphasized in our introduction, with this work only the electron subsystem has been considered. Realistic heat transport in real molecular junctions would involve the complexity of interacting electrons and electron-phonon interactions [8]. This electronic heat transport may dom- inate in certain situations so that the measured heat noise can be attributed approximately to the electronic compo- nent only. The unified approach, which would include both the electron and the phonon subsystems, as well as the ef- fects of their interactions, presents a future challenge al- though several contributions in this direction for the aver- age heat current (but not the heat current noise PSD) have already been undertaken before [8,30,33,51]. 4.1 Open issues We conclude this study with further remarks that may shed light on challenging open problems and in addition may invigorate others to pursue future research in objectives addressed with our study. A first observation is that we obtained within the Green function analysis tractable ex- pressions for quantum transport in the steady state without ever having to invoke the explicit knowledge of the inher- ent nonequilibrium density operator. Naturally, the quan- tum averages for the current and the auto-correlation of the quantum fluctuations carry less information as encom- passed with the full steady state nonequilibrium density operator. The latter nonequilibrium density operator is typ- ically very difficult to obtain and explicit results are known for tailored situations only. In fact, explicit results are very intricate already for those cases with overall quadratic Hamiltonians [52]. As discussed above, a much more subtle issue refers to the experimental detection of quantum correlations. In clear contrast to the case with a quantum, single-time ex- pectation of a quantum observable, the issue of measure- ment of manifest quantum correlations is a delicate ob- jective that is only rarely addressed with sufficient care in the literature. This is so because the mere calculation of a theoretical two-time quantum correlations does not say anything about its feasible experimental measurement sce- nario. Either strong (i.e. von Neumann-type) or weak quan- tum measurements impact the dynamics as clearly mani- fested with the example of the Zeno-effect [53,54]. With more than one time present this objective re- lates to the problem of measurements of quantities that are not given in terms of quantum observables [49,50,55,56]. To appreciate the complexity somewhat in more detail let us first consider the case with classical random variables. Then the PSD can be obtained experimentally as the limit of a time average of the classical random process J h(t), via considering the expression Sh t→∞(ω) = limt→∞ Z t −t 1 2t(cid:12)(cid:12)(cid:12)(cid:12) 2 . ds J h(s) exp(iωs)(cid:12)(cid:12)(cid:12)(cid:12) (37) Note that classically the measurement of the stochastic variable of the instantaneous heat flow J h(t) presents no serious problem while the same is not straightforward for a quantum dynamics. Moreover, even classically, the re- sult in (37) holds true only when the stochastic, finite value t (ω) tends to the exact ensemble averaged value Sh(ω), Sh with its variance approaching zero as t → ∞. The latter implies conditions of higher, fourth-order correlations to be satisfied [57]. With the feature of dealing with the non- commutation property of quantum observables at different times no such direct scenario is available for experiment. Here the complexity of quantum measurements will enter in its full generality. Only for tailored situations this task may simplify further, as it was the case in Sects. 3.2, 3.4 for the zero frequency limit. As mentioned already above, the case of quantum lin- ear response theory may come as support also for nonequi- librium: The measurement of a single observable (here the heat flux operator) due to an external perturbation is typi- cally related to the evaluation of a specific quantum corre- lation function [47]. The case of the quantum-dissipation relation of Callen-Welton in thermal equilibrium presents Copyright line will be provided by the publisher 10 First author et al.: Short title such a celebrated case [58,59,60]. There, the dissipative part of the measurable, frequency-dependent susceptibil- ity of a perturbed observable B is uniquely related to the power spectral density SBA(ω) of quantum fluctuations of the observable B and the fluctuations of observable A to which an applied external conjugate force couples. In our case it remains therefore a formidable task to identify the corresponding variable for the nonequilibrium situation so that the single-time measurement of its linear response be- comes related to the heat PSD in Eq. (30) in a prescribed manner. This at best is possible for the thermal equilib- rium PSD in which an imposed energy perturbation cou- ples to the thermal affinity ∆T /T ; cf. in Refs. [61,62, 63]. This is not possible, however, for the equilibrium heat flow fluctuations at absolute T = 0, with the inherent ther- mal affinity being divergent. In presence of quantum co- herence destroying phenomena, such as high temperature or disorder, the nature of quantum correlations becomes suppressed. Then, the classical scenario can be used again to validate the theoretical predictions in thermal equilib- rium [4,60] and for tailored steady-state nonequilibrium situations; note the nonequilibrium fluctuation theorems in Ref. [47]. Acknowledgements Work supported by the German Ex- cellence Initiative via the "Nanosystems Initiative Munich" (NIM) (P.H.), the DFG priority program DFG-1243 "Quantum transport at the molecular scale" (F.Z., P.H.), and the Volkswagen Foundation (Project No. I/83902) (P. H. and S.D). References [1] M. A. Reed, C. Zhou, C. J. Muller, T. P. Burgin, and J. M. Tour, Science 278, 252 (1997). [2] A. Nitzan and M. A. Ratner, Science 300, 1384 (2003). [3] C. Joachim and M. A. Ratner, Proc. Natl. Acad. Sciences 102, 8801 (2005). [4] Y. M. Blanter and M. Buttiker, Phys. Rep. 336, 1 (2000). [5] P. Hanggi, M. Ratner, and S. Yaliraki, Chem. Phys. 281, 111 (2002). [6] S. Kohler, J. Lehmann, and P. Hanggi, Phys. Rep. 406, 379 (2005). [7] G. Cuniberti, G. Fagas, and K. Richter, (eds.), Lect. Notes Phys. 680, 1 -- 518 (2005). [8] Y. Dubi and M. D. Ventra, Rev. Mod. Phys. 83, 131 (2011). [9] S. R. Forrest, Nature 428, 911 (2004). [10] C. Joachim, J. K. Gimzewski, and A. Aviram, Nature 408, 541 (2000). [11] N. J. Tao, Nanotechnology 1, 173 (2006). [12] X. D. Cui, A. Primak, X. Zarate, J. Tomfohr, O. F. Sankey, A. L. Moore, T. A. Moore, D. Gust, G. Harris, and S. M. Lindsay, Science 294, 571 (2001). [13] J. Reichert, R. Ochs, D. Beckmann, H. B. Weber, M. Mayor, and H. v. Lohneysen, Phys. Rev. Lett. 88, 176804 (2002). [14] L. S. Levitov and M. Reznikov, Phys. Rev. B 70, 115305 (2004). [15] J. P. Morten, D. Huertas-Hernando, W. Belzig, and A. Brataas, Phys. Rev. B 78, 224515 (2008). [16] D. A. Bagrets and Y. V. Nazarov, Phys. Rev. B 67, 085316 (2003). Copyright line will be provided by the publisher [17] M. Esposito, U. Harbola, and S. Mukamel, Rev. Mod. Phys. 81, 1665 (2010). [18] N. Cl´ement, S. Pleutin, O. Seitz, S. Lenfant, and D. Vuil- laume, Phys. Rev. B 76, 205407 (2007). [19] C. Beenakker and C. Schonenberger, Physics Today 56, 37 (2003). [20] Y. P. Li, D. C. Tsui, J. J. Heremans, J. A. Simmons, and G. W. Weimann, Appl. Phys. Lett. 57, 774 (1990). [21] M. Buttiker, Phys. Rev. B 45, 3807 (1992). [22] S. Camalet, S. Kohler, and P. Hanggi, Phys. Rev. B 70, 155326 (2004). [23] F. Zhan, N. Li, S. Kohler, and P. Hanggi, Phys. Rev. E 80, 061115 (2009). [24] H. L. Engquist and P. W. Anderson, Phys. Rev. B 24, 1151 (1981). [25] U. Sivan and Y. Imry, Phys. Rev. B 33, 551 (1986). [26] P. N. Butcher, J. Phys. Condens. Matter 2, 4869 (1990). [27] J. Koch, F. von Oppen, Y. Oreg, and E. Sela, Phys. Rev. B 70, 195107 (2004). [28] Y. C. Chen and M. Di Ventra, Phys. Rev. Lett. 95, 166802 (2005). [29] M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B 73, 045314 (2006). [30] M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B 75, 155312 (2007). [31] D. Segal, Phys. Rev. B 73, 205415 (2006). [32] M. Paulsson and S. Datta, Phys. Rev. B 67, 241403 (2003). [33] M. Galperin, M. A. Ratner, and A. Nitzan, J. Phys.: Con- dens. Matter 19, 103207 (2007). [34] J. S. Wang, J. Wang, and J. T. Lu, Eur. Phys. J. B 62, 381 (2008). [35] N. Li, J. Ren, L. Wang, G. Zhang, P. Hanggi, and B. Li, Rev. Mod. Phys. 84, 1045 (2012). [36] J. Lehmann, S. Kohler, P. Hanggi, and A. Nitzan, Phys. Rev. Lett. 88, 228305 (2002). [37] J. Lehmann, S. Camalet, S. Kohler, and P. Hanggi, Chem. Phys. Lett. 368, 282 (2003). [38] C. W. Chang, D. Okawa, A. Majumdar, and A. Zettl, Science 314, 1121 (2006). [39] L. Wang and B. Li, Phys. Rev. Lett. 99, 177208 (2007). [40] B. Li, L. Wang, and G. Casati, Phys. Rev. Lett. 93, 184301 (2004). [41] B. Li, L. Wang, and G. Casati, Appl. Phys. Lett. 88, 143501 (2006). [42] M. Rey, M. Strass, S. Kohler, P. Hanggi, and F. Sols, Phys. Rev. B 76, 085337 (2007). [43] I. V. Krive, E. N. Bogachek, A. G. Scherbakov, and U. Land- man, Phys. Rev. B 64, 233304 (2001). [44] D. V. Averin and J. P. Pekola, Phys. Rev. Lett. 104, 220601 (2010). [45] F. Zhan, S. Denisov, and P. Hanggi, Phys. Rev. B 84, 195117 (2011). [46] Ø. L. Bø and Y. Galperin, J. Phys.: Condens. Matter 8, 3033 (1996). [47] P. Hanggi and H. Thomas, Phys. Rep. 88. 207 (1982). [48] V. S. Vladimirov, Equations of Mathematical Physics (New York, Dekker, 1971). [49] M. Campisi, P. Talkner, and P. Hanggi, Phys. Rev. Lett. 105, 140601 (2010). pss header will be provided by the publisher 11 [50] M. Campisi, P. Talkner, and P. Hanggi, Phys. Rev. E 83, 041114 (2011). [51] D. Segal, A. Nitzan, and P. Hanggi, J. Chem. Phys. 119, 6840 (2003). [52] A. Dhar, K. Saito, and P. Hanggi, Phys. Rev. E 85, 011126 (2012). [53] W. M. Itano, D. J. Heinzen, J. J. Bollinger, and D. Wineland, Phys. Rev. A 41, 2295 (1990). [54] P. Facchi and S. Pascazio, Progress in Optics, 42, 147 (2001). [55] P. Talkner, E. Lutz, and P. Hanggi, Phys. Rev. E 75, 050102 (2007). [56] A.A. Clerk, M.H. Devoret, S.M. Girvin, F. Marquardt, and R.J. Schoelkopf, Rev. Mod. Phys. 82, 1155 (2010). [57] A. Papoulis, Probability, Random Variables, and Stochastic Processes (McGraw-Hill Book Company, New York, 1965); see pp. 343-ff therein. [58] H. B. Callen and T. Welton, Phys. Rev. 83, 34 (1951). [59] J. Des Cloizeaux, Linear response, generalized susceptibil- ity and dispersion theory, In: Theory of condensed matter, F. Bassani, G. Cagliotto, J. Ziman, eds. (International Atomic Energy Agency, Vienna 1968), pp. 325 -- 354. [60] P. Hanggi and G. L. Ingold, Chaos 15, 026105 (2005). [61] W. M. Visscher, Phys. Rev. A 10, 2461 (1974). [62] P. B. Allen and J. L. Feldman, Phys. Rev. B 48, 12581 (1993). [63] S. Liu, P. Hanggi, N. Li, J. Ren, and B. Li, Anomalous heat diffusion, arXiv:1306.3167. Copyright line will be provided by the publisher 

1502.02938

2

1502

Time correlators from deferred measurements

Repeated measurements as typically occurring in two- or multi-time correlators rely on von Neumann's projection postulate, telling how to restart the system after an intermediate measurement. We invoke the principle of deferred measurement to describe an alternative procedure where co-evolving quantum memories extract system information through entanglement, combined with a final readout of the memories described by Born's rule. The new approach to repeated quantum measurements respects the unitary evolution of quantum mechanics during intermediate times, unifies the treatment of strong and weak measurements, and reproduces the projected and (anti-) symmetrized correlators in the two limits. As an illustration, we apply our formalism to the calculation of the electron charge correlator in a mesoscopic physics setting, where single electron pulses assume the role of flying memory qubits. We propose an experimental setup which reduces the measurement of the time correlator to the measurement of currents and noise, exploiting the (pulsed) injection of electrons to cope with the challenge of performing short-time measurements.

cond-mat.mes-hall

cond-mat

a Time correlators from deferred measurements D. Oehri,1 A.V. Lebedev,1 G.B. Lesovik,2 and G. Blatter1 1Theoretische Physik, ETH Zurich, CH-8093 Zurich, Switzerland 2L.D. Landau Institute for Theoretical Physics, RAS, 142432 Chernogolovka, Russia (Dated: October 16, 2018) Repeated measurements as typically occurring in two- or multi-time correlators rely on von Neu- mann's projection postulate, telling how to restart the system after an intermediate measurement. We invoke the principle of deferred measurement to describe an alternative procedure where co- evolving quantum memories extract system information through entanglement, combined with a final readout of the memories described by Born's rule. The new approach to repeated quantum measurements respects the unitary evolution of quantum mechanics during intermediate times, uni- fies the treatment of strong and weak measurements, and reproduces the projected and (anti-) symmetrized correlators in the two limits. As an illustration, we apply our formalism to the calcu- lation of the electron charge correlator in a mesoscopic physics setting, where single electron pulses assume the role of flying memory qubits. We propose an experimental setup which reduces the mea- surement of the time correlator to the measurement of currents and noise, exploiting the (pulsed) injection of electrons to cope with the challenge of performing short-time measurements. PACS numbers: 03.65.Ta, 73.23.-b I. INTRODUCTION Within the quantum world, the question what quan- tities can be measured in an experiment is often a non-trivial one, e.g., measuring time correlators (with times τj) requires finding the correct ordering of op- erators. Concrete examples in mesoscopic physics and quantum optics are the measurements of charge correlators1 -- 4 or full counting statistics5 -- 7 and that of photon correlators8,9. The question is usually resolved by including the measurement apparatus in the descrip- tion and its internal workings decide upon the form of the measured correlator. Examples are the Amp`eremeter, double-dot detector, and spin counter used in Refs. 1, 3, and 5, or the different photodetectors introduced by Glauber and by Mandel10. These detectors then act back on the system, thereby influencing the measurement out- come, i.e., the specific form of the correlator. E.g., a weak measurement as used in Ref. 1, see also Ref. 11, response functions2,12, while a strong measurement pro- leads to symmetrized (RS(τ1, τ2)) and antisymmetrized (IS(τ1, τ2)) correlators weighted with different detector duces a projected correlator SP(τ1, τ2). at times τj< τn in a n-th order correlator). The measure- These different forms of measured correlators can be derived13 by invoking the von Neumann projection postulate14, telling how to restart the system after the first measurement at τ1 (or after previous measurements ment with a weakly coupled detector can be treated per- turbatively, with the von Neumann projection excerted on the detector and no back action on the system1. In a strong measurement with a large system -- detector cou- pling, the projection formally can be applied directly to the system, therefore producing a maximal back action13. In this paper, we invoke the principle of deferred measurement15 known from quantum information theory, cedures for a two-time (τ1, τ2 > τ1) correlator: (a) strong FIG. 1: Schematic illustration of different measurement pro- measurement described by von Neumann projection acting directly on the system at time τ1 and providing a projected correlator SP after readout at τ2; for a weak measurement, the von Neumann projection at time τ1 acts on the weakly coupled detector. (b) Repeated measurement without von Neumann projection at τ1: unitary co-evolution of system and quantum memories which are entangled at times τ1 and τ2 and final readout after τ2. The coupling strength between the system and the quantum memories determines the degree of entanglement. where it can be used for quantum computing, and apply it to the problem of repeated measurements, specifically, of time-correlators. We replace the von Neumann projec- tion by entangling the measured system with co-evolving quantum memories, see Fig. 1, thereby (effectively) ex- panding the Hilbert space of the total system in every measurement step. The desired correlator then is de- rived from a final measurement of all the quantum mem- ories by invoking Born's rule16. Hence the entire system plus memories undergoes a unitary quantum evolution until the very end, where the Born rule takes us from the quantum to the classical world. The new scheme cap- (b)SysMemMemSys11ττ22(a)tures the cases of weak and strong measurement within a unique formalism by merely changing the degree of en- tanglement between the system and the quantum mem- ory. In the limits of weak and strong entanglement, we reproduce the results previously derived via use of the projection postulate. No simple physical form for the time-correlators could be found so far in the intermedi- ate coupling regime. Describing a measurement by entangling the system with a detector and including a (dissipative) bath in the evolution of the density matrix is a concept that has been well developed over the past two decades17 -- 20. Here, we extend the idea of system -- detector entanglement to the case of repeated measurement. Thereby, the quantum memories evolve coherently during and after the infor- mation transfer from the system due to entanglement, with the (dissipative) measurement deferred to the very end of the process. The proposed scheme for the measurement of time- correlated observables finds an interesting application in mesoscopic physics. In particular, measuring the charge Q dynamics of a quantum dot with the help of a nearby quantum point contact is a classic problem by now21. In such a setup, single electron pulses5,22 assume the role of flying qubit memories which are either transmitted or re- flected by the quantum point contact (QPC), depending on the dot's charge state, see Fig. 2. Analyzing the charge transmitted across the QPC then provides the desired in- formation on the dot's charge correlator. Making use of recent developments in electron quantum optics23,24, we propose a setup, see Fig. 3, that shifts the task of re- solving short times, typically done on the level of the detection, to the proper timing of electron pulses and gate operations. a system observable O with an eigenbasis{n} that is to We briefly sketch the main idea of the paper: Consider be measured. We start with a system state at the initial time τin ψ(τin)=Q ψn(τin)n and a quantum memory in the initial state φ n (1) in and (1) have them transiently interact at time τ1 with the help of an externally controlled interaction (to be identified with a quantum detector). The system and memory then become entangled,
Q ψn(τ1)n(cid:6)φ whereφ(1) n  denote memory states after interaction with the system in staten and we assume a negligible evolu- ψn(τ1)nφ in→Q n , (1) (1) (2) n n Q tion of the system during the time of interaction. Evolv- ing the system to the later time τ2 and entangling it with a second memory, we obtain the state Umn(τ21)ψn(τ1)mφ n φ (1) m, (2) with Umn(τ) the matrix elements of the system propa- gator U(τ) and τ21 = τ2− τ1. The unitary evolution of (3) m,n 2 the memory states φ(j) n  preserves the system informa- tion gained at times τj. At time τfin> τ2 we measure the memory observables a(1) and a(2) with discrete spectra a(1,2) α . Making use of Born's rule, we find the probability distribution function Pαβ for the measurement outcomes α and β on the two memory observables; this proba- bility distribution contains the desired information on the system's two-time correlator. The specific relation between the distribution function Pαβ of measurement outcomes and the correlator of system observables O(τ1) and O(τ2) depends on the system -- detector coupling and the observables measured on the memories; we will show below how to extract the well known (anti-)symmetrized and projected correlators from the probabilities Pαβ in the limits of weak and strong measurements. In the end, by making use of quantum memories which store the information acquired from the system at the quantum level at earlier times τ1 and τ2 until the final time τfin, we have avoided the intermediate readout which requires the use of the projection postulate. Hence the entire measurement process follows a unitary quantum evolution until the transition to the classical world is done via Born's rule. In the following, we will first derive the general frame- work describing the deferred measurement of a correlator with the intermediate von Neumann projection replaced by a system -- detector entanglement (Sec. II). In Section II A, we discuss the limit of weak measurement and use qubits as quantum memories to arrive at a simple re- lation between the measurement outcome on the qubits and the (anti-) symmetrized correlators of the system. In Sec. II B, we first discuss a strong measurement at strong coupling using qudit memories and then invoke (weakly coupled) qubit registers to show that both types of strong measurements produce the projected time cor- relator of the system. An illustration of our formalism is given in Sec. III, where we describe the measurement of the charge correlator in a mesoscopic setting, specifically, the two-time charge correlator of a quantum dot (QD) as measured by a quantum point contact (QPC). Sec. III E describes a possible experimental implementation and in Sec. IV we summarize our results. II. CORRELATOR MEASUREMENTS BY QUANTUM MEMORIES (j) given by Eq. (1), while the memories are described by We consider the situation where a two-time correla- tor of a system operator O is measured with the help of two quantum memories; the system's initial stateψ is in, j= 1, 2 (see below for the discussion initial statesφ ing memory statesφ(j) in depend on the system state n, where un describes the time evolution of the of an open system described by a density matrix ρ). The memories interact with the system at times τ1,2 during a small time intervall δτ . After this interaction, the result- n = unφ (j) quantum memories during the interaction with the sys- l,m,n (1) (2) n φ tem (we assume a trivial evolution of the free memories). to keep their system information after their interaction. Ulm(τf 2)Umn(τ21)ψn(τ1)lφ remain unchanged during the time δτ of the individual interaction events, see Sec. III C for an extended discus- sion of this point. After the second interaction event at The system state ψ(τ) = ∑n ψn(τ)n is assumed to τ2, the wave functionψ(τfin> τ2) of the system is en- tangled with the states φ(j) n  of the memories and the combined wave functionΨf reads m, (4) Ψf= Q with τf 2= τfin− τ2. The quantum memories are supposed At time τfin, we measure the operators a(1) and a(2) on nϕ(j) n =∑α sα ϕ(j) α , we rewrite the memory statesφ(j) α . (discrete) eigenvalues and eigenstates of a(j) by a(j) Pαβ(τ21)=Ψfp β Ψf, α =ϕ(j) α ϕ(j) α . Making use of the where p(j) lm′ = δmm′ (rendering the unitarity condition ∑l UlmU∗ j-th memory, i.e., p(j) evolution Ulm(τf 2) in Eq. (4) irrelevant), we obtain the mUmn(τ21)sα Q Pαβ(τ21)=Q the first and second memory, respectively. Denoting the α and Applying Born's rule to the final state (4) provides us with the probability distribution α is the projector onto the eigenstate α of the nψn(τ1)2 Pαβ(τ21) provide us with the desired information on the The above expressions (5) and (6) for the probabilities (2) (1) α p probabilities (5) (6) sβ m n . (1) (1) (2) (2) two-time correlator of the system. They are easily gen- eralized to the case of open systems by introducing the combined system plus bath (the open system) density matrix ρ and evolve it in time including the subsequent entanglement with the quantum memories: Starting from the initial density matrix ρ0⊗φ in⊗φ in de- inφ inφ scribing the open system plus memories at time τin, we proceed as in the case of isolated systems by condition- ing the time evolution of the memory states on the corre- sponding system states and obtain the final density ma- trix at time τf Ukm(τf 2) Umn(τ21) ρnn′(τ1) U ρf= Q n′m′(τ21) (7) m′, mφ n′⊗φ n φ × U m′k′(τf 2)kk k,k′,n,n′,m,m with the open system's density matrix ρ(τ1) at time τ1, its reduced part ρnn′(τ1)=nρn′, and the reduced op- erators Uil=i Ul with U the evolution operator of the open system. Note that here, the outgoing statesφ n(′) m(′) are conditioned on the system states n(′) and andφ (1) m(′) at times τ1 and τ2. We define the probabilities Pαβ ′⊗φ (2) (2) (1) (1) (2) † † as Pαβ(τ21)= Tr
p β ρf(cid:6) (2) (1) α p (8) with the trace taken over both the open system and the memory states. Calculating this expression with the final density matrix Eq. (7), we obtain Pαβ(τ21)= Q n,n′,m Tr
sβ m Umn(τ21) sα n ρnn′(τ1) m(cid:6), n′m(τ21) sβ∗ × sα∗ n′ U † 3 (9) with the remaining trace taken over the bath degrees of freedom. This result is the direct generalization of (6) to the case of open systems. The expressions (5) and (6) as well as (8) and (9) constitute the basic formulas which we will further develop in the following sections. Indeed, as expressed in terms of evolution amplitudes of system and detectors, it is difficult to appreciate the physical meaning and content of these results. In order to make progress, we consider next the two cases of weak and strong measurements. A. Weak measurement Given a weak system -- detector coupling, the most direct way to find the probabilities Pαβ in terms of physically transparent quantities is to start from Eq. (5) and eval- uate this expression perturbatively in the linear system -- detector coupling Hsd = ∑j b(j)(τ) O, where the time- dependent coupling b(j)(τ) acts on the j-th memory dur- (2) ing a time δτ around τj. The unperturbed evolution of the memories is described by the Hamiltonian h0 and we make use of the interaction representation. We go over to irreducible quantities by subtracting the uncorrelated contribution, αβ= Pαβ− P f p(1) α Ψ f  and P P irr (1) (1) α P (2) β P (10) (2) (1) (2) (1) (2) Pαβ, f  de- f p β =Ψ β Ψ scribing measurements involving a single entanglement at time τ1 or τ2 with only one memory, respectively. The α can be obtained by a simple summation of α =Ψ with P(1) quantity P(1) α =Q α = P(1) and P(2) erwise, the determination of P(2) for a time-independent problem (oth- α necessitates a second β ≠∑α Pαβ. measurement). Note that the sum over the first index of (2) Pαβ already includes correlations, see Eq. (6), and hence P αβ=Ψ U β (τf) UD(τf , τin)Ψ, (12) system stateψ(τin), ⋅ refers to the irreducible part, with the expectation value to be taken over the initial The task then is to evaluate the irreducible expression D(τf , τin)p α (τf)p Pαβ, P irr (2) (1) (11) α β † and the time evolution operator reads UD(τf , τin)=T exp− iŏh S τf τin ′ Hsd(τ ′), dτ (13) withT denoting time-ordering. Evaluating (12) to lowest relevant order in the coupling, we find 4 P irr αβ=(−i)2ŏh2 S τf =(−i)2ŏh2 S τ τin τin dτ dτ dτ dτ ′ ′S τ ′S τ τin ′ τin (1) ′′Ψ[[p ′′
ψ O(τ −ψ O(τ −ψ O(τ +ψ O(τ α (τf)p ′) O(τ ′′) O(τ ′) O(τ ′′) O(τ (2) (1) (1) β (τf), Hsd(τ ′)], Hsd(τ ′′)]Ψ ′′)ψφ (1)(τ α (τf)b inp ′′)φ inφ ′)φ (2)(τ β (τf)b inp in ′)ψφ (1)(τ inb ′′)p α (τf)φ inφ ′)φ (2)(τ in ip β (τf)b in ′′)ψφ in ip α (τf)b (1)(τ ′′)φ inφ inb (2)(τ ′)p β (τf)iφ in in(cid:6), inb ′)p inb inφ ′′)p ′)ψφ β (τf)φ (2)(τ α (τf)φ (1)(τ Hsd(τ)= Ω(τ)σx O, where the coupling Ω(τ) is switched (2) (2) (2) (2) (1) (1) (1) (1) (2) (2) (2) (2) (1) (1) (1) (1) (2) (2) (1) (1) (2) (2) (14) (j) (17) (16) (15) (2) (1) (2) (1) P irr anti-commutator) to arrive at the final result with the detector response functions where we made sure that the first memory interacts with the system at the earlier time τ′′. For a slow system dy- namics and exploiting that b(j)(τ)φ in≠ 0 only for τ≈ τj, we can replace O(τ′′) → O(τ1) and O(τ′) → O(τ2). (with[⋅,⋅] and{⋅,⋅} denoting the usual commutator and We make use of the standard definitions for the sym- metrized and anti-symmetrized irreducible correlators OO(τ1, τ2)={ O(τ1), O(τ2)}~2, RSirr OO(τ1, τ2)=−i[ O(τ1), O(τ2)], ISirr αβ(τ21)=IS det,αIS OO(τ1, τ2) det,βRSirr det,αIS +RS OO(τ1, τ2), det,βISirr dτφ α (τf), b(τ)}φ in{p in, RS det,α=− 1 S τf 2ŏh det,α=−iŏh in. in[p dτφ α (τf), b(τ)]φ IS S τf The symbols R and I address symmetrized and anti- RSirr OO andISirr obtainsRSirr OO=[RS ISirr OO=[−IS with D(1) = RS In a situation where the full information Pαβ can be extracted from the memories, the individual correlators OO can be obtained from (17) by combining ¯αβ one (1) det, ¯α. Alterna- tively, one may have preferential access to combinations of probabilities Pαβ (see Sec. III E for an example) or make use of specific detector properties, see below and the appendix for examples. αβ−RS αβ+IS det,α−RS det, ¯αIS symmetrized quantities (or equivalently, up to factors of 2, real- and imaginary parts). ¯αβ]~D (1)IS ¯αβ]~D (1)IS det,αIS (1) det, ¯αP irr (1) det, ¯αP irr (1) (1) det,αP irr (1) det,αP irr (1) two different probabilities, e.g., using P irr (2) det,β, (20) (2) det,β, (21) (j) (j) (j) (j) αβ and P irr (j) (j) (1) (18) (19) (j) (j) τin τin A generic choice for the quantum memories are qubit devices that couple to the system via the Hamiltonian (j)ϕ1]~√ φ in=[ϕ0+ eiθ (j) on during a short time δτ around τj. Assuming initial states . , , τin 2, (j) (j) (j) (j) (j) (22) (25) (24) (23) (j) (j) (j) (j) (j) (j) volve integrals of the type and we obtain the response functions OO) from the memory correlator P irr will rotate the qubit around the x-axis by δϑ On with in= δϑ ei(1−2α)θ inpα(τf)Ω(τ)σxφ , RS det,1= δϑ cos θ det,1=−2δϑ sin θ IS a system residing in a state n with eigenvalue On δϑ = Ωδτ~ŏh. The response functions (18) and (19) in- dτφ S τf RS det,0= δϑ cos θ IS det,0= 2δϑ sin θ Choosing θ(2) = π~2, OO (ISirr tor RSirr choosing θ(1)= π~2 (θ(1)= 0). Alternatively, we may use the results (20) and (21) and θ(1)= θ(2)= π~4 to find (we choose α= 0, ¯α= 1, and β= 0) OO= P irr RSirr OO=− P irr ISirr 00 + P irr i.e., polarizing the second mem- ory along the y-axis, we can directly find the correla- αβ by Note that the sums P irr 01 contain very different types of information, once a correlation, the other time only a mean value, as discussed in more detail in Sec. III E. Hence, we find that the delayed measure- ment of the two quantum memories provides the sym- metrized and anti-symmetrized correlators (15) and (16). Finally, we note that a weak linear coupling between the system observable and the detector/memory variable canonically conjugated to the detector readout provides a more effective entanglement. Such a von Neumann like interaction allows to produce a strong entanglement and a strong measurement even for weak coupling if sufficient time is available for the entanglement process25. 00 − P irr 00 + P irr 00 + P irr 10 and P irr 2δϑ2 4δϑ2 (26) (27) 10 10 , . B. Strong measurement αn (j) (j) (j) (j) (29) (28) , φ φ We are now going to show that a strong system -- detector coupling naturally leads to projective correla- tors. Consider a situation where the system operator O is measured via entanglement with two quantum memories. We first consider an operator O with a non-degenerate spectrum and comment on the general case in the end. A strong coupling between the system and the memory im- plies that the memory statesφ(j) n  after interaction with the system in staten are fully distiguishable, i.e, mφ n = δnm. statesϕ(j) α  and we assume a one-to-one relation with the The observables a(j) distinguish between memory eigen- evoluted memory statesφ(j) n , n =ϕ with αn≠ αm for n≠ m (otherwise, the observable a(j) m= δββm for the second measures linear combinations of eigenstates of O and thus is not suitable for a measurement of this observable). Un- der these (strong coupling) conditions, the amplitudes sα n reduce to sα memory). In order to describe the strong coupling situ- ation it is favorable to proceed with the expression (6) and we obtain the result n= δααn (for j= 1, sβ Pαnβm(τ21)=
Umn(τ21)ψn(τ1)
2 l Pm(τ2) Pn(τ1)ρ0 Pn(τ1)l (τ21)=Q =Umn(τ21)ψn(τ1)2, with the projected density matrix ρP(τ1) = ∑k Pk(τ1) ρ0 Pk(τ1) and the projectors Pk = kk onto different (τ21) are easily combined into the desired two-time system states, see Eq. (1). The projected correlators SP correlator SP The right hand side of the above expression is nothing but the projected system correlator (30) (31) PnPm PnPm SP . l OO(τ21) OO(τ21)= Tr[ O(τ2) O(τ1)ρP(τ1)] (τ21) OnOmPαnβm(τ21), =Q =Q OnOmSP PnPm nm SP (32) nm thereby establishing the general result relating the sys- tem correlator to the memory readings. A further sim- O obey a linear relation On= η(j)a(j) plification can be achieved if the eigenvalues of a(j) and (2), OO(τ21)=Q = η Pαnβm(τ21) obtain the simple result , in which case we (2)a (1) (1) (1) (2) (2) (1) (33) SP nm βm αn αn a a a η η η 5 The system evolution of directly relating the projected system correlator to the These results can easily be generalized to the case where the observable O involves a degenerate spec- trum: is conditioned on the eigenvalue On rather than the memory correlatora(1)a(2). state n of the system. Degenerate eigenstates pro- to be replaced by PαOβO′(τ21) = Pαnβm(τ21) has ∑{mOm=O′}∑{nOn=O} Umn(τ21)ψn(τ1)2 and ρP(τ1) is PO(τ1)ρ0 PO(τ1) with the projection substituted by ∑O operator PO = ∑{nOn=O}nn. With these replace- duce equal evolutions of the memories, the memories implying that ments, the above results remain valid also for a degener- ate spectrum. Hence, the perfect entanglement between the system and the memories arising due to strong coupling is equiv- alent to a von Neumann projection applied to the system. While no back action is apparent on the level of the sys- tem dynamics, the strong back action of this maximal en- tanglement with the memories manifests itself in a strong change of the systems density matrix when tracing over the memories. above requires equal or more memory states ϕα than system statesn, hence quantum memories with dimen- The realization of a strong measurement as described sionality d or qudits are required. Besides the obvious difficulty in realizing qudit memories, their coupling to the system in order to serve as a measurement device is equally nontrivial. As an alternative, we discuss a mea- surement scheme involving qubit registers, instead. Such an alternative setup implementing a strong mea- surement involves a weak system -- detector coupling but invokes multiple measurements. We replace the strongly coupled qudit memories by weakly coupled qubit regis- ters with J qubits each, probing the system state close to τ1 and τ2 within a short time interval δτ . For a strong measurement, J is chosen sufficiently large to distinguish the different eigenvalues On of the system. Assuming the system not to change during the interaction time Jδτ with one register, the final state of system and memory registers is of the same form as in Eq. (4) with the out- going individual memory statesφ(1) n  andφ(2) m replaced by the outgoing register states Φ(1) n  and j=1φ(j) n  = ∏J Φ(2) m=∏2J m. Making use of Born's rule, we ob- j=J+1φ(j) αβ(τ21) for finding µ (ν) qubits of the first (second) register in states α∈{0, 1} (β∈{0, 1}), (sβ Q Q J αβ=J m)J−νUmn(sα m)ν(s n)J−µψn2 n)µ(s ¯α n= s1−β n= s1−α
sα α(n)=
J n2(J−µ) for measuring µ n2µs ¯α system in staten, we can separate the system- and de- n′]P µ [UmnψnU αβ= Q α(n, n ∗ ∗ mn′ ψ n,n′,m with s ¯α probability P µ of the J qubits in the state α after interaction with the tector response in the above equation, n . Introducing the conditional ′)P ν β(m) tain the probabilities P µν m and s ¯β P µν P µν (34) µ ν ¯β n n µ , P µ (j) p α (35) (j) states. µ,ν νP ν (j) p µP µ n,m µ ν nsα µ with P µ P µ the 'off-diagonal'
(sα conditional ns ¯α probabilities (note that µνP µν αβ β(m). around the x-axis, the evolution α(n)P µ n′∗)µ(s ¯α n′∗)(J−µ) In the non-degenerate case (i) and for a strong two cases: (i) all system eigenvalues are non-degenerate, measurement, the probability distributions P µ P µ tions of µ (this is the very definition of this measure- ment being a strong one) and the 'off-diagonal' ele- ments P µ
J = α(n, n′) α(n, n)= P µ α(n)). The conditional probabilities P µ α(n) depend only on the eigenvalue On of the staten (and not on the staten itself). We then have to distinguish i.e., On≠ On′ for n≠ n′, (ii) there are degenerate system α(n) and α(n′) for different n ≠ n′ do not overlap as func- α(n, n′) for n′ ≠ n are suppressed, as follows α(n, n′)=
α(n′). We then from the relationP µ αβ ≈ ∑n,mUmnψn2P µ β(m). The register correlators α(n)P ν Sαβ(τ21)=Ψf JQ 2JQ j=1 j=J+1 Umnψn2Q = Q Assuming again a linear system -- qubit coupling Hsd(τ)= Ω(τ)σx O that rotates the qubits by an angle δϑ On β Ψf=Q α(n)Q can simplify the expression (34) to the form P µν (replacing the distribution functions Pαβ) take the form again assume initial states polarized in the xy-plane, see With the qubits initially polarized along the y-axis, i.e., un= cos(δϑ On) −i sin(δϑ On) cos(δϑ On)  n = unφ i sin(δϑ On) produces the memory states φ(j) in, where we (22). The probabilities Pα(n) for an individual qubit to reside in state α= 0, 1 are given by +(1− 2α)δϑ On sin θ. P0(n)= 1− P1(n)≈ 1 θ(j)= π~2, we define the register's 'magnetizations' 1(n)]= 2Jδϑ On, µ[P µ M(n)=Q 0(n)− P µ combination S11− S10− S01+ S00 then involves the prod- uct of register polarizations M(n)M(m) and using the OO(τ21)=∑nm OnOmUmnψn2, see Eq. (32), OO(τ21)= S11− S10− S01+ S00 Note the normalization S11+ S10+ S01+ S00= J 2, which follows from replacing M(n)=∑µ µ[P µ 0(n)− P µ 1(n)] by Σ(n)=∑µ µ[P µ 1(n)]=µ0+µ1= J. 0(n)+ P µ separate only for states with different eigenvalues On≠ relation SP we can relate the system correlator to the register corre- lators via where we have made use of (37) in the last equation. The In the degenerate case (ii), the distribution functions . (39) (36) (37) (38) SP 4J 2δϑ2 2 µ 6 αβ then are given by α(n, n′) ∼ 0, while for degenerate eigenvalues with On= On′, P µ α(n, n′)= P µ α(n). The prob- On′, implying that P µ α(n)P ν n′]P µ β(m). Q{n′On′=On}[UmnψnU αβ≈ Q abilities P µν ∗ ∗ mn′ψ On the other hand, the projected density matrix ρP(τ1) OO(τ21) involves the projectors PO = ∑{nOn=O}nn, such that the result appearing in the system correlator SP P µν m,n (39) remains unchanged. Summarizing, we have seen that the measurement scheme invoking entanglement with quantum memories and their delayed measurement provides us, in the weak- and strong measurement limits, with the same results as obtained via the traditional route using an intermedi- ate von Neumann projection. Furthermore, these results have been obtained within a unified description starting from the same initial formula in the form (5) or (6). The above general theoretical considerations are rather non-trivial to implement in a practical situation, as the preparation, entanglement, and measurement of many quantum memories is often a non-trivial task. The im- plementation of these ideas is less demanding, though still challenging, when considering specific examples. In- deed, quantum memories for delayed measurement are naturally provided in a scattering geometry, where indi- vidual scattered particles take the role of flying qubits. In the following, we focus on a specific example in meso- scopic physics, the measurement of a charge correlator of a quantum dot by scattering electrons in a nearby quan- tum point contact. We first analyze the situation for a simple two-state system with charges Q= ne, n= 0, 1, and then extend these considerations to arbitrary charge states. III. CHARGE CORRELATOR MEASUREMENT We consider a classic problem21, the charge Q dynam- ics of a quantum dot (QD) (attached to leads or coupled to another dot in an isolated double-dot system) mea- sured by a nearby quantum point contact (QPC), see Fig. 2 and Ref. 13 for a recent discussion of this sys- tem. Here, we want to characterize the dot's dynamics by its two-time charge correlator. In this system, the mea- surement is executed by the electrons which are trans- mitted across/reflected by the QPC with probabilities that depend on the dot's charge state. For the present discussion, it is convenient to view the QPC current as a sequence of individual electron pulses; during recent years, this theoretical idea5,26,27 has progressed to an ex- perimental reality22 -- 24, opening the new field of electron quantum optics28. The quantum memories then can be viewed as flying qubits, individual electron pulses arriv- ing at the QPC at times τ1 and τ2 that probe the charge state of the QD through the capacitive coupling between the QD and the QPC, see Fig. 2. 7 Second, we generalize the discussion to a QD with mul- tiple charge states as described by the charge operator Q = ∑n Qnnn. While two-state quantum memories (qubits) are always sufficient for a complete description when the measurement is weak, for a strong measure- ment, a QD with multiple charge states will require the use of qubit registers, i.e., finite trains of electron pulses. A. Weak measurement τin † † corrections δT , δθ, and δχ. T− δT ei(θ+δθ), r1=√ more, we parametrize the scattering matrices for the sys- R eiχ, It remains to determine ues α, β equal to r and t. the detector response functions (19) and (18). We con- For the case of a weak measurement, we can make immediate use of the general results Eqs. (17), (20), and (21) by replacing O → Q and choosing the val- sider a linear system -- detector coupling of the form Hsd= e2 q Q~C = v Q with q the charge on the QPC. Further- tem in states 0 and 1 by t0 = √ T eiθ, r0 = √ t1=√ R+ δT ei(χ+δχ), with small gral (−i~ŏh)∫ τf dτφinpα(j)(τf)v(τ)φin (we drop the In order to find the detector response functions (18) and (19) we replace b by v and determine the inte- memory index (j) as the electrons are scattered by the ten in the formφinu 0 pα u0φin with u0= e−ih0(τf−τin)~ŏh and u1= e−i(h0+v)(τf−τin)~ŏh describing respectively. Furthermore, we have used that u1= u0 uD the dynamics of the detector in the Heisenberg represen- tation in the absence and presence of a charge on the dot, 0 pα u1φin−φinu same QPC). To lowest order in v, this can be writ- 1 in δT , δθ, and δχ, we find that dτ v(τ)~ŏh
uD=T exp
−iS τf and v(τ) in the interaction representation. The initial memory statesφin evolve under u1 and u0 according to unφin=φn and henceφinu m pα unφin=φmpαφn= n, where we have used that pα=ϕαϕα. Expand- sα∗ −iŏh S τf −iŏh S τf final resultsRSdet,t= T δθ, RSdet,r= Rδχ, ISdet,r= δT. dτφinpt(τf)v(τ)φin=− 1 dτφinpr(τf)v(τ)φin= 1 δT+ iT δθ, δT+ iRδχ, ISdet,t=−δT, and taking real and imaginary parts, we arrive at the m sα ing sα (43) (44) with τin τin (41) (42) 2 2 τin † (40) Using these detector response functions in the general expressions (20) and (21) one easily arrives at the system (j) (j) positions on the right of the QPC, provides information on the two-time charge correlator. FIG. 2: Quantum dot system (QD) measured by a capaci- tively coupled quantum point contact (QPC): Single-electron pulses incident on the QPC from the left (at τin) are either transmitted (with amplitude t) or reflected (amplitude r); the outgoing Lippmann-Schwinger wave functions describe flying qubits without own dynamics and serve as quantum memo- measure the two-time correlator of the dot's charge. After scattering at the QPC, the two electrons (flying qubits) are entangled with the quantum dot system and carry information on its dynamics. Simultaneous detection of the two scattered ries. Two pulses separated in time by τ2− τ1 are needed to electrons (at τfin), e.g., a distance vF(τ2− τ1) away with both The memory statesϕ(j) α  are the two scattered states where the electron is reflected (α = r) or transmitted (α = t), i.e., the outgoing state is given by φ(j) n  = tnϕ t + rnϕ(j) r  with scattering coefficients tn↔ st rn↔ sr r,t emanating from the lution of the scattered waves ϕ information, in particular,φ(2) n = 0. This allows us mφ(1) n and n depending on the charge state of the system. We assume well separated single-electron pulses and an evo- QPC at times τj that preserves the corresponding system to envision an individual detection of the electrons24,29. In the final readout, the flying qubits are detected on the right or left side of the QPC, telling whether the two electrons have been transmitted (with probability Ptt), reflected (Prr) or mixed (Ptr and Prt). These probabil- ities then contain the information about the two-time charge correlator of the QD. Formally, such a final-state analysis corresponds to α , with measuring the (charge) operators a(j)=∑α a(j) α =ϕ(j) α providing the transmit- α ϕ(j) α p(j) ted (α= t) or reflected (α= r) components of the j-th the projectors p(j) t = 1 and a(j) r = 0 if the transmitted charge electron. The eigenvalues a(j) r = 1 and (j) t = 0 if the reflected charge is measured on the left. is measured on the right of the QPC, while a(j) (j) probability distributions (5) or (6) with α, β= r, t. binary charge states0 and1 and eigenvalues Q0= 0, Q1 = 1, Q = 11 (defined in units of e) and an ini- tial state ψ(τin)= ψ0(τin)0+ ψ1(τin)1, see Eq. (1). a Both types of measurements provide us with the same In the following, we first analyze the case of a QD with α depend on the measured charge, e.g., a 1221τfin(τ −τ )21vFQDin21QPCτQPCQD, rt rt (46) (45) P irr QQ. QQ or ISirr Alternatively, the QPC can be tuned to deliver the Indeed, using a detector with high transmission, e.g., a QPC with correlators (we choose α= t, ¯α= r, and β= t), QQ= R(δχ~δT) P irr tt − T(δθ~δT) P irr RSirr δT(Rδχ+ T δθ) tt + P irr ISirr QQ=− δT(Rδχ+ T δθ) . individual system correlators RSirr energetic (E) single-electron pulses Eâ V0, V0 the QPC barrier, one easily finds that (although δTâδθ,δχ, we have δTâ Rδχ) RSirr Eâ V0, the situation is reverse,RSdet,râISdet,r~tâ RSdet,t, andISirr tudes tn = √ 1− Tneiχn of the QPC Tn = T− QnδT , θn = θ+ Qnδθ, and χn = χ+ Qnδχ. A RSdet,tâISdet,r~tâRSdet,r, tβ measuresISirr QQ (RSirr Tneiθn and rn = √ rβ measures QQ. When the detector predominantly reflects par- ticles, e.g., for low energy single electron pulses with tβ ), see the appendix for further details on the QPC detector response. For a quantum dot with multiple charge states we have to require that the expansion of the scattering ampli- detector scale linearly in the charge Qn of the dot, i.e., QQ) is measured by P irr and therefore P irr QQ, while P irr rβ (P irr (47) straightforward calculation then shows that the results (43) and (44) for the detector response functions as well as the final results (45) and (46) remain unchanged. Similarly, the probability to find no charge on the dot in either of the two measurements is Ptt =U00ψ02, while the mixed results are Ptr=U10ψ02 and Prt=U01ψ12. 8 The strong measurement of the charge correlator for a multi-charge quantum dot quite naturally involves trains of electron pulses19, with the number J of electrons in each train sufficiently large to distinguish the different charge eigenvalues Qn of the dot. The separation δτ between electron pulses within a train has to be suffi- ciently long in order to allow for their separate detection (i.e., counting), while the train duration J δτ must re- main small on the scale τsys of the dot's dynamics. µ[P µ r(n)− P µ When going over from qubit registers to electron trains scattered at the QPC we replace the 'magnetization' (38) by the disbalance between reflected and transmitted elec- trons, t(n)]= J[R− T+ 2QnδR], (50) D(n)=Q charge Qn, Rn= R+ QnδR. Operating the QPC at the symmetry point T= R= 1~2, we can determine the com- bination Srr− Srt− Str+ Stt and relate this quantity to where we have assumed a linear QPC characteristic with a reflection probability scaling linearly with the dot's µ the projected charge correlator SP QQ(τ21), QQ(τ21)= Srr− Srt− Str+ Stt . SP Operating the QPC away from the symmetric point, one has to determine the weighted sum T 2Srr− T R Srt− RT Str+ R2Stt instead and divide by J 2δR2 to arrive 4J 2δR2 (51) at the projected correlator SP QQ. B. Strong measurement C. Finite-width memory wave-packets When performing a strong measurement of a quantum dot with a binary charge it is sufficient to invoke individ- ual electron pulses as quantum memories. For a strong dot -- QPC coupling, we require a one-to-one relation be- tween the presence of a charge on the dot and the out- come of the measurement, i.e.,φ0=ϕt andφ1=ϕr, 0= 0. This is achieved by outcome withr1= 1,r0= 0 andt0= 1,t1= 0, i.e., the presence of a charge Q1= 1 on the dot reflects the QPC probability Prr(τ21) that directly traces the charge Q and see Eq. (29), or sr tuning the QPC such as to generate a unique scattering electron back to the left. In this case, it is the reflection 1= 1 and sr according to (6), we have to determine the expression Prr(τ21)=Q Q rmUmn(τ21)rnψn(τ1)2 m With rm = δm1, we obtain the simple result Prr(τ21)= U11(τ21)ψ12 (see also (30) with n, m= 1 and α1= β1= r) and since Qn= δn1, we find the projected correlator (see (48) n . (49) Eq. (32)) QQ(τ1, τ2)= Prr(τ21)=U11(τ21)ψ12. SP time scale of the system. Here, we allow for a spread in time of the detector's electron wave function and drop the In general terms, this corresponds to a measurement which Above we have assumed an instantaneous (within a short time δτ ) entanglement between the system and the memory states, requiring that both the width τwp of the wave-packet and the scattering time τsca at the QPC sat- isfy τwp, τscaâ τsys, where τsys denotes the characteristic condition τwpâ τsys, i.e., we assume that τwp࣠ τsys while the scattering event itself remains fast, τsca â τsys. probes the system sharply (τscaâ τsys) during some finite time (τwp࣠ τsys; longer measurement times τwp> τsys do tion f(j)(τ) which is normalized (∫ dτf(j)(τ)2= 1) and Ψf=S dτ ′ × Q Ulm(τ m(τ 2), ′ (2)(τ 1)S dτ (1)(τ 2) ′ ′ ′ 21)ψn(τ f 2)Umn(τ 1)lφ ′ ′ ′ Let us suppose that the j-th wave-packet incident on the QPC around τj is described by the wave func- peaked at the time τj. Assuming instantaneous scatter- ing, we obtain the final state n (τ 1)φ ′ not provide meaningful results). (2) (1) (52) 1 f 2 f l,m,n f 2 = τf − τ′ 21 = τ′ 2− τ′ with τ′ 2 and τ′ j) that have been scattered at the n (τ′ memory statesφ(j) j. Making use ofφ n′(τ′′ j)φ(j) n (τ′ j)= δ(τ′ j− QPC at time τ′ j)φ n′φ(j) n , we obtain the smeared probabilities τ′′ (j) Pαβ(τ21)=S dτ 2f (1)(τ 21) (53) 1)2f (2)(τ 2)2Pαβ(τ 1S dτ ′ ′ ′ ′ ′ 1 and the outgoing (j) with Pαβ given by Eq. (6). The finite width wave-packets enter as integration kernels. While for sharp wave- packets, the entanglement between system and memories (i.e., the 'measurement') takes place at times τ1 and τ2, for broader wave-packets, the entanglement arises within a finite time τwp around τ1 and τ2 with distribution func- 2)2. As a consequence, for the 1)2 andf(2)(τ′ tionsf(1)(τ′ case of strong coupling where the entanglement gives rise to projective measurements, the times of projection are not fixed but distributed with the distribution functions above. Note that the result (53) is only valid for negligible scattering time τsca in the QPC. If the scattering time τsca is finite compared to the system time τsys, the effect of interaction cannot be accounted for by the scattering matrices sn of the QPC depending on the system state, but the interaction during the scattering has to be treated in more detail. The limit of τscaâ τsys considered above has also im- action for τsca â τsys only consists of dephasing, i.e., a plications for the resulting back-action: While the back- suppression of off-diagonal elements of the system's den- sity matrix, for finite τsca, the back-action of the mea- surement on the system goes beyond pure dephasing and alters the system's dynamics. D. Higher-order correlators 9 in Sirr QQQ (encoded in σ) relate to opposite detector re- sponse functions (encoded by ¯σ). This result agrees with the one in Ref. 4 obtained with the help of the von Neu- mann projection postulate and shows that only Keldysh time-ordered charge correlators are measurable. E. Experimental implementation Our general concept of deferred repeated measure- ments has been formulated with quantum memories, e.g., qubits, qubit registers, and qudits, see Sec. II. In our application of these general considerations, the measure- ment of a charge correlator with the help of a quantum point contact (Sec. III), the quantum memories have been replaced by scattered electrons (flying qubits). It then is natural to seek for an experimental implementation, where the final measurement of the quantum memories, i.e., the scattered electrons, can be cast into a measure- ment of currents and noise rather than an individual de- tection of qubit states. Such an implementation is pro- posed below. By now, several experiments have demonstrated the controlled generation of individual electron pulses22 -- 24 and even the detection of such pulses via qubit detectors seems to be in reach29. A setup particularly well suited within the current context is the one of Fletcher et al.23, involving a single electron pump and a time-correlated detector setup in a quantum Hall device, see Fig. 3 for an illustration. Pairs of electron pulses of width∼ 100 ps are injected at a typical rate νp∼(10 ns)−1 and are well τsys∼ 100− 1000 ns. The detector involves a dynamically suited to probe the dynamics of a dot with a system time switchable barrier (dsb in Fig. 3) that can be switched on or off with a nanosecond precision in time. the single electron pump (with injection currents in the We propose an experiment implementing a weak mea- surement of a charge correlator along the lines of Sec. III A, where electron pairs with a tunable delay time in the range τ21= 10− 100 nanoseconds are injected by I = 2eνp ∼ 10 pA regime) and scattered at the QPC tector barrier dsb is switched on at a time τd+ τ21~2 in probing the quantum dot QD. The transmitted electrons arrive at the detector dsb after a delay time τd. The de- t (2) (1) order to let the first electron pass along the path 1 and deflect the second electron along 2, see figure 3. With the static barrier (ssb) closed, the measured currents I t and I resulting from the transmission of the first and second electron, respectively, can be related to the prob- abilities Pαβ via t ~eνp= Ptt+ Ptr, t ~eνp= Ptt+ Prt. The two sums∑β Pαβ and∑α Pαβ in the above equations α Ψf for the (independent) transmis- sion (α= t) or reflection (α= r) of the first electron by ity P(1) are fundamentally different in a subtle way: summing over the second particle β produces the trivial probabil- α = Ψfp(1) (1) (2) (58) (57) I I (54) (55) sγ l m,n nψn(τ1)2 its irreducible part can be recast in the form l Ulm(τ32)sβ The study of higher-order correlators is straightfor- ward, e.g., to measure a third-order charge correlator, we send three electrons scattering from the QPC at times τ1< τ2< τ3 and obtain the probabilities Pαβγ=Q mUmn(τ21)sα Q describing electrons transmitted across (α, β, γ = t) or reflected from (α, β, γ= r) the QPC. For weak coupling, αβγ= Q QQQ (τ1, τ2, τ3) σσ′=± S with ¯σ=−σ, the detector responses S det,α=RS det,α=IS (j) QQQ = cσcσ′[ Q(τ1),[ Q(τ2), Q(τ3)]σ′]σ det,α, and the third-order correlators with the constants c+= 1~2 and c−=−i and[⋅,⋅]−=[⋅,⋅] resp. [⋅,⋅]+={⋅,⋅} (note that (anti-)symmetrized charges (j) det,α and (3),− det,γ Sσσ (2),¯σ′ det,βS (1),¯σ det,αS (j),− (j),+ Sσσ P irr (56) ′ ,irr ′ ,irr S 10 QQ. thogonality of these four components, the particle num- electron state after the scattering events at the QPC, barrier (ssb) is left open, such that the two particles inter- We can calculate the evolution of the state through the Hong-Ou-Mandel setup using the wave function in channel 3 as a function of mutual delay τ (tuned via an additional gate in loop 1, see figure) then provides all periment of Bocquillon et al.24 and sketched in Fig. 3. In this experiment, the dynamically switchable barrier (dsb) again splits the two electrons in each pair to propagate along the paths 1 and 2 → 2′, respectively. The static fere in the splitter. Measuring the current noise S(3)(τ) information needed to constructRSirr Ψf=Ψftt+Ψftr+Ψfrt+Ψfrr describing the two- with Pαβ= αβΨfΨfαβ, i.e., the individual components are not normalized and onlyΨfΨf= 1. Due to the or- bers N(i), i= 3, 4, emerging from our Hong-Ou-Mandel the particle number fluctuationsΨf(δ N3)2Ψf in chan- tions, rtΨf N3Ψfrt=(1~2)⋅ 1⋅ Prt and rtΨf N 2 3Ψfrt= (1~2)⋅ 12⋅ Prt, hence, and the same result holds true for theΨftr scattering component. While there is no contribution fromΨfrr, the one originating fromΨftt depends on the time delay τ . Let f1(x) and f2(x) denote the two electron wave- relate the number fluctuation in the componentΨftt to packets propagating along the incoming paths 1 and 2 of the HOM-interferometer. As shown in Ref. 30, we can rtΨf(δ N3)2Ψfrt= 1 nel 3 involve single-particle and two-particle contribu- splitter can be analyzed term by term. ⋅ Prt, In particular, (60) 4 the overlap of wavefunctions as ttΨf(δ N3)2Ψftt= 1 1−f1f22Ptt, wheref1f2=∫ dxf∗ 1(x)f2(x). For a large time delay fere, f1f2= 0, such that the fluctuations are just the between the first and second electron, these do not inter- (61) 2 double of (60), ttΨf(δ N3)2Ψftt= 1 Compensating the original time delay τ=−τ21 (such that (62) Ptt. 2 the two electrons arrive simultaneously at the splitter), we have ttΨf(δ N3)2Ψftt= 0, (63) since the Pauli exclusion forces the two particles to propagate to different channels. result Ψf(δ N3)2Ψf then involves separately the probabili- ties Prt+ Ptr and Ptt, The final Ψf(δ N3)2Ψf∞= 1 (Prt+ Ptr)+ 1 4 Ptt, (64) 2 t t t QQ. (1) (4) (3) and I +I FIG. 3: Schematic illustration of a quantum Hall device with a single electron pump (left), the QPC -- QD setup (mid- dle) and a detector arrangement (right), inspired by Refs. 23,24. Pairs of electrons are injected by the single electron pump with a time delay τ21 and scattered by the quantum point contact with scattering coefficients r and t depending tron (flying qubits j= 1 and j= 2) into arms 1 and 2. Closing on the charge state of the quantum dot. The dynamically switchable barrier23 (dsb) separates the first and second elec- = I charge correlatorISirr (1) (2) the static switchable barrier (ssb) and measuring the currents t provides the anti-symmetrized charge- I t QQ. Opening the static switchable barrier (3)(τ) as a function of the tunable delay τ (ssb), having the two particle streams along the trajectories in the detector 3 provides the symmetrized correlatorRSirr 1 and 2' interfere at the 50/50 beam splitter, and measuring the current noise S the QPC,∑β Pαβ= P(1) t ~eνp = P (1) α . Hence, the first equation (57) and thus contains only infor- reduces to I mation about the mean charge on the dot. normalization ∑α P(j) = P (2) On the contrary, summing over the first particle α does not generate P β as the interaction of the first electron at the earlier time τ1 introduces nontrivial correlations, see (2) Eq. (6). This becomes apparent when going over to irre- ducible probabilities P irr α P β and using the (2) (58) becomes I t which includes information about the dot's charge correlator. Assum- ing a time-independent mean charge on the dot, we have P the measured currents are easily transformed to provide the antisymmetrized correlator αβ= Pαβ− P(1) α = 1. Then, the second equation tt + P irr rt + P t ~eνp= P irr = I t ~eνp and using the general result (17), t − I t )~eνp QQ=(I ISirr δT(Rδχ+ T δθ) . tr ∝(ISdet,r+ISdet,t)= 0, explicitly tt + P irr tt + P irr tt + P irr In order to find the symmetric correlator RSirr Note that the evaluation of the irreducible probability P irr tr with the help of (17) indeed provides a vanish- ing result, P irr demonstrating that the sum P irr relations. QQ one has to measure a time correlator on the transmitted chan- nel. This can be conveniently done with the help of a Hong-Ou-Mandel type splitter as implemented in the ex- tr contains no cor- (2) (2) (1) (1) (2) (1) (59) t t It(3)It(1)pumpIsplitterQPCQDrtssbdsbI(2)tdelay(1)(2')(3)(2)(4)when the delay between electron pulses is not compen- sated and Ψf(δ N3)2Ψf0= 1 (Prt+ Ptr), 4 W (65) (3)(t1) I In a next step, we express the charge fluctuations in channel 3 through the irreducible current-current correlator,30 where the time-window W is centered around the arrival time of the wave-packets at the detector 3 and has a p , with νp the rate of pair injection by the pump. The particle number fluctuations then can be expressed through the low-frequency current when the time delay is properly compensated, τ=−τ21. (3)(t2), (66) Ψf(δ N3)2Ψf= 1 dt1dt2 I e2S width of the order of ∆τ = ν−1 noise S(3)(ω), sin2(ω~2νp) S(3)(ω) Ψf(δ N3)2Ψf=S dω (ω~2)2 with ω≤ νp. Neglecting the frequency dependence of the noise at small ω, S(3)(ω)∼ S(3)(0), we obtain (3)~e2νp, S(3)∞ − S Ψf(δ N3)2Ψf= S , Ptt= 2 and hence (3) (67) (68) (69) 2π e2 0 , . e2νp (3) S 0 e2νp Prt+ Ptr= 4 rt + P irr tt − T δθ Rδχ(P irr QQ (70) tr and P irr tt , rt + P irr (δT)2(Rδχ+ T δθ)2 Using Eq. (17), we can derive an alternative (more symmetrized) version of Eq. (45) which involves just the combinations P irr RSirr =(Rδχ)2 P irr tr )+(T δθ)2 P irr Defining the detector parameters κ= T δθ~(Rδχ+ T δθ) = t = I t ~eνp independent of time), we obtain the final result(δT)2RSirr QQ= = 2(1−2κ) S(3)∞ − S − κ2 and going over to irreducible probabilities (with P P + κ2− I + 4κ (2) (1) (3) (1) (71) rr 0 . . t (3) S 0 e2νp e2νp (1) t eνp Note that in this setup, the final measurement of qubit memories does not require fast or time resolved detection schemes, but merely relies on the measurement of aver- age currents and low-frequency noise. This is due to the fact, that all timing tasks are realized by the properly time-delayed electron pulses in the incoming channel and the dynamically switchable gate (dsb) which separates the electron pairs; both elements have been realized in an experiment23. Hence, the new measurement scheme, 11 combined with novel elements from electron quantum op- tics, allows to shift the (difficult) timing issues in the measurement of a time-correlator from the detector to the source. is given by Eq. (49), SP A strong coupling between the quantum dot and the quantum point contact provides us with a projected cor- relator. For the simplest case of a binary charge on the dot with values Q= 0, 1, the projected charge correlator QQ= Prr. Making use of the nor- malization Ptt+ Prt+ Ptr+ Prr= 1 and the result Eq. (69), QQ= 1− 2 0 + S(3)∞ we find that (72) SP S . (3) e2νp IV. CONCLUSION In conclusion, we have applied the principle of deferred measurement to the problem of repeated measurement and have derived physical expressions for the two- and multi-time correlators. The measurement involves the inclusion of quantum memories which are entangled with the system at specific times τj where the system observ- able is to be probed. The expanded system plus mem- ories undergoes a unitary evolution until the very end, where the result is extracted via application of Born's rule to the memories. The measured probabilities Pαβ (see Eq. (6)) or memory correlators Sαβ (see Eq. (35)) then can be combined to extract the desired system corre- lators. The limits of weak and strong measurements pro- vide the standard (anti-) symmetrized and projected time correlators previously obtained by invoking the (non- unitary) von Neumann projection. The general results have been illustrated by using qubits and qubit registers as quantum memories. Our analysis sheds new light on the problem of repeated measurement and illustrates the usefulness of qubits as sensitive measurement devices. Although our paper's main results are rather on the conceptual side, one could imagine an implementation of such a deferred measurement in an experiment. A sys- tem that naturally lends itself for a realization of these ideas is the classic mesoscopic setup which probes the charge of a quantum dot through a quantum point con- tact. The individual scattered electrons in the QPC can be understood as flying qubits which are either trans- mitted or reflected, with amplitudes depending on the charge state of the quantum dot. In particular, the qubit register required in the strong measurement of a dot with a multi-valued charge is easily implemented in terms of finite trains of electrons. We have applied our formalism to this situation and derived the corresponding expres- sions for a weak and strong measurement. The experimental implementation of these ideas re- quires a system control that can only be met with a modern quantum engineering approach. Recent de- velopments in electron quantum optics provide con- trolled single electron pulses and allow for their time re- solved manipulation/detection on a sub-nanosecond time scale23,24. Using the setup of Ref. 23 as a base and aug- menting it with a Hong-Ou-Mandel type analyzer24, we propose to include a quantum point contact and a quan- tum dot in order to realize the principle of repeated mea- surement by a deferred measurement of quantum memo- ries. As a final remark, one may appreciate the relation of the deferred measurement principle to Everett's idea of a multiverse31,32. Rather than applying a projection af- ter the first measurement and pursuing a single further evolution (of the system = 'universe'), the principle of a deferred measurement involving the system's entangle- ment with a quantum memory enhances the overall di- mensionality, e.g., for a qubit memory the dimensional- ity is doubled (with two 'universes' evolving in parallel). It is then only the final measurement which determines which evolution (i.e., which 'universe') actually has been realized. Acknowledgments We thank Renato Renner for discussions and acknowl- edge financial support from the Swiss National Sci- ence Foundation through the National Center of Compe- tence in Research on Quantum Science and Technology (QSIT), the Pauli Center for Theoretical Studies at ETH Zurich, and the RFBR Grant No. 14-02-01287. Appendix: Detector properties A quick overview is provided by the example of a δ- function scatterer: Expressing the strength of the scat- terer ŏh2λ~m for the two charge states by λ0 = λ and λ1 = λ+ δλ, an incoming state with wave vector k is transmitted with amplitude tn= k~(k+ iλn), n∈{0, 1}. Expanding the transmission Tn= k2~(k2+ λ2 n) and phase θn =− arctan(λn~k), we find the modifications δT and δθ= δχ (for a symmetric scatterer) in the scattering char- acteristic of the QPC upon charging the dot δT≈− 2k2λ2 (k2+ λ2)2 δθ= δχ=− kλ k2+ λ2 In the limit of a large incoming energy, i.e., kâ λ, we find δT ≈ −2λ δλ~k2 and δθ = δχ ≈ −δλ~k and hence δθ, δχ â δT ; with T ≈ 1 and R ≈ λ2~k2 â 1, we have Tδθ â δT â Rδχ and therefore ISdet,t â RSdet,r~t â ISdet,r. For a small k â λ, we obtain δT ≈ −2k2δλ~λ3 and δθ ≈ −kδλ~λ2 and using T ≈ k2~λ2 and R ≈ 1, we find ISdet,r â RSdet,r~t â ISdet,t. When k ≈ λ, all response func- incoming energy δλ λ δλ λ (A.1) (A.2) , . tions are of the same order. Alternatively, we can consider a single electron tran- sistor (SET) with the level position kres,n affected by the 12 . , (A.4) (A.3) γδkres capacitive coupling and depending on the dot's charge state n, i.e., kres,0= kres and kres,1= kres+ δkres. The transmission coefficient is given by tn= iγ~(k−kres,n+iγ), where γ is the level width. Again expanding Tn = γ2~((k− kres,n)2+ γ2) and tan θn=(k− kres,n)~γ for small δkresâ kres, we find δT=− 2(k− kres)γ2δkres [(k− kres)2+ γ2]2 δθ= δχ=− (k− kres)2+ γ2 For incoming electrons on resonance with the level, i.e., k−kresâ γ, we obtain δT≈−2(k−kres)δkres~γ2 and δθ= δχ≈−δkres~γ, such that δθ, δχâ δT and using T≈ 1 and R≈(k− kres)2~γ2, we find thatISdet,tâRSdet,r~tâ ISdet,r. On the other hand, for off-resonant electrons δT ≈ −2γ2δkres~(k− kres)3 and δθ = δχ ≈ −γδkres~(k− kres)2, such that δθ, δχâ δT and using T≈ γ2~(k−kres)2 and R ≈ 1, we find ISdet,r â RSdet,r~t â ISdet,t. Whenk− kres≈ γ, all response functions are of the same ing potential Vn(x)= Vn−kx2~2 where the offset Vn is the n. Here, we assume a quasi-classical description and consider the two limits of electrons with energy Eâ Vn resp. Eâ Vn, see Fig. 4. A more realistic description for the quantum point con- tact (QPC) is achieved by considering a parabolic scatter- QPC barrier height when the dot is in the charge state order. FIG. 4: QPC modeled by a parabolic potential.
Using the Kemble formula33, we obtain the trans- mL2~8ŏh2(E− Vn)~√ mission Tn= 1~(1+ exp[−2π Vn]), where we have chosen V(±L~2)= 0. For weak coupling δV = V1− V0 â V0, we obtain the shift (we define the energy scale EL=ŏh2~2mL2) E+ V0 δV√ which is suppressed exponentially for Eâ V0 and Eâ V0 The change in phase at large energies E â V0 is de- region [−L~2, L~2]; within a quasi-classical description, termined by the transmission phase accumulated in the due to an exponentially small reflection or transmission. δT= π TERE ELV0 (A.5) V0 4 , 0L2L2V0EVV1dx S L~2  −L~2 θn= 1ŏh = 1 in phase δθ= δχ, this is given by (ε≡ E~Vn)
2m(E− Vn(x))
√ ε−(ε− 1) log(ε− 1)1~2 +(ε− 1) log(1+√ ε)(cid:6).  δV√ δθ≈− 1 Vn EL 2 . V0 E Expanding this result for small δV , we obtain the change (A.6) (A.7) 3 ELV0 Given the exponential suppression of δT at large ener- gies Eâ V0, we find that δT âδθ=δχ and for large transmission we have ISdet,tâRSdet,r~tâISdet,r. In the opposite regime of small energies Eâ V0, we de- termine the change in phase (within quasi-classics) from 1 G.B. Lesovik and R. Loosen, JETP Lett. 65, 295 (1997). 2 G.B. Lesovik, Phys.-Usp. 41, 145 (1998). 3 R. Aguado and L.P. Kouwenhoven, Phys. Rev. Lett. 84, 1986 (2000). 4 K.V. Bayandin, A.V. Lebedev, and G.B. Lesovik, JETP 106, 117 (2008). 5 L.S. Levitov, H.W. Lee, and G.B. Lesovik, J. Math. Phys. 37, 4845 (1996). 6 D.A. Bagrets and Yu.V. Nazarov, Phys. Rev. B 67, 085316 (2003). 7 A. Di Lorenzo, G. Campagnano, and Y.V. Nazarov, Phys. Rev. B 73, 125311 (2006). 8 R.J. Glauber, Phys. Rev. Lett. 10, 84 (1963). 9 L. Mandel, Phys. Rev. 152, 438 (1966). 10 L. Mandel and E. Wolf, Optical coherence and quantum optics (Cambridge University Press, Cambridge, 1995). 11 A. Bednorz, C. Bruder, B. Reulet, and W. Belzig, Phys. Rev. Lett. 110, 250404 (2013). 12 G.B. Lesovik and I.A. Sadovskyy, Phys. Usp. 54, 1007 (2011). 13 D. Oehri, A.V. Lebedev, G.B. Lesovik, and G. Blatter, 13
S x0−L~2 χn≈ 2ŏh the phase of the reflection amplitude, 2m(E− Vn(x)), where the reversal point x0 < 0 is characterized by V(x0)= E. To leading order in δV we find that (A.8) dx δχ≈− 1  E 3~2 δV√ . (A.9) 3 V0 ELV0 ponential suppression of T and the response functions re- Once more, it follows thatδθ,δχâ δT due to the ex- spect the orderRSdet,râISdet,r~tâRSdet,t. At in- tions satisfyingRSdet,tâISdet,r~tâRSdet,r while at small transmissionRSdet,râISdet,r~tâRSdet,t. termediate energies, the response functions are of similar magnitude. Summarizing, we find that a scatterer with large transmission is characterized by the response func- Frost, G.A.C. Jones, and D.G. Hasko Phys. Rev. Lett. 70, 1311 (1993); L.M.K. Vandersypen, J.M. Elzerman, R.N. Schouten, L.H. Willems van Beveren, R. Hanson, and L.P. Kouwenhoven, Appl. Phys. Lett. 85, 4394 (2004); S. Gus- tavsson, R. Leturcq, B. Simovi c, R. Schleser, T. Ihn, P. Studerus, K. Ensslin, D.C. Driscoll, and A.C. Gossard, Phys. Rev. Lett. 96, 076605 (2006). 22 G. Feve, A. Mahe, J.-M. Berroir, T. Kontos, B. Placais, D.C. Glattli, A. Cavanna, B. Etienne, and Y. Jin, Sci- ence 316, 1169 (2007); J. Dubois, T. Jullien, F. Portier, P. Roche, A. Cavanna, Y. Jin, W. Wegscheider, P. Roulleau, and D.C. Glattli, Nature 502, 659 (2013). 23 J.D. Fletcher, P. See, H. Howe, M. Pepper, S.P. Giblin, J.P. Griffiths, G.A.C. Jones, I. Farrer, D.A. Ritchie, T.J.B.M. Janssen, and M. Kataoka1, Phys. Rev. Lett. 111, 216807 (2013). 24 E. Bocquillon, V. Freulon, J.-M. Berroir, P. Degiovanni, B. Pla¸cais, A. Cavanna, Y. Jin, G. F`eve, Science 339, 1054 (2013). 25 S.A. Gurvitz, Int. J. of Mod. Phys. B 20, 1363 (2006). 26 A.V. Lebedev, G.B. Lesovik, and G. Blatter, Phys. Rev. B Phys. Rev. B 90, 075312 (2014). 14 J. von Neumann, Mathematische Grundlagen der Quan- 72, 245314 (2005). 27 J. Keeling, I. Klich, and L.S. Levitov, Phys. Rev. Lett. 97, tentheorie (Springer, Berlin, 1931). 15 M. Nielsen and I. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, Cam- bridge, 2000). 16 M. Born, Z. Phys. 37, 863 (1926). 17 S.A. Gurvitz, Phys. Rev. B 56, 15215 (1997). 18 Y. Makhlin, G. Schon, and A. Shnirman, Phys. Rev. Lett. 85, 4578 (2000). 19 D.V. Averin and E.V. Sukhorukov, Phys. Rev. Lett. 95, 126803 (2005). 20 H.M. Wiseman and G.J. Milburn, Quantum measurement and control (Cambridge University Press, Cambridge, 2010). 21 M. Field, C.G. Smith, M. Pepper, D.A. Ritchie, J.E.F. 116403 (2006). 28 C. Grenier, R. H´erv´e, G. F`eve, P. Degiovanni, Mod. Phys. Lett. B 25, 1053 (2011). 29 R. Thalineau, A.D. Wieck, Ch. Bauerle, and T. Meunier, arXiv:1403.7770. 30 A.V. Lebedev and G. Blatter, Phys. Rev. B 77, 035301 (2008). 31 M. Tegmark, Nature 448, 23 (2007). 32 The Many-Worlds Interpretation of Quantum Mechanics, edited by B.S. De Witt and N. Graham (Princeton Uni- versity Press, Princeton, New Jersey, 1973). 33 E.C. Kemble, Phys. Rev. 48, 549 (1935). 

1610.08218

1

1610

Resistive Switching and Voltage Induced Modulation of Tunneling Magnetoresistance in Nanosized Perpendicular Organic Spin Valves

Nanoscale multifunctional perpendicular organic spin valves have been fabricated. The devices based on an La$_{0.7}$Sr$_{0.3}$MnO$_3$/Alq$_3$/Co trilayer show resistive switching of up to 4-5 orders of magnitude and magnetoresistance as high as -70% the latter even changing sign when voltage pulses are applied. This combination of phenomena is typically observed in multiferroic tunnel junctions where it is attributed to magnetoelectric coupling between a ferromagnet and a ferroelectric material. Modeling indicates that here the switching originates from a modification of the La$_{0.7}$Sr$_{0.3}$MnO$_3$ surface. This modification influences the tunneling of charge carriers and thus both the electrical resistance and the tunneling magnetoresistance which occurs at pinholes in the organic layer.

cond-mat.mes-hall

cond-mat

Resistive Switching and Voltage Induced Modulation of Tunneling Magnetoresistance in Nanosized Perpendicular Organic Spin Valves Robert Göckeritz,1 Nico Homonnay,1 Alexander Müller,1 Bodo Fuhrmann2, and Georg Schmidt1,2,a) 1Institut für Physik, Martin-Luther-Universität Halle-Wittenberg, 06099 Halle (Saale), Germany 2Interdisziplinäres Zentrum für Materialwissenschaften, Martin-Luther-Universität Halle-Wittenberg, 06099 Halle (Saale), Germany a)Electronic mail: [email protected]. Nanoscale multifunctional perpendicular organic spin valves have been fabricated. The devices based on an La0.7Sr0.3MnO3/Alq3/Co trilayer show resistive switching of up to 4-5 orders of magnitude and magnetoresistance as high as -70% the latter even changing sign when voltage pulses are applied. This combination of phenomena is typically observed in multiferroic tunnel junctions where it is attributed to magnetoelectric coupling between a ferromagnet and a ferroelectric material. Modeling indicates that here the switching originates from a modification of the La0.7Sr0.3MnO3 surface. This modification influences the tunneling of charge carriers and thus both the electrical resistance and the tunneling magnetoresistance which occurs at pinholes in the organic layer. In the past years a number of multiferroic tunnel junctions have been demonstrated in which tunneling magnetoresistance (TMR) and total device resistance can be modulated by a voltage pulse.1,2 The effects are typically explained by tunneling electroresistance (TER) due to a ferroelectric barrier which changes the total resistance and magnetoelectric coupling at the interface between ferroelectric barrier and ferromagnetic contact which changes the TMR in magnitude and sometimes in sign. We observe the same functionality in organic spin valves (OSVs, Fig. 1), which after applying a voltage pulse may change the device resistance by three orders of magnitude or more and modulate their magnetoresistance (MR) from +26% to -38% which is a much larger effect than observed in Refs. 1 or 2. Nevertheless, the absence of a ferroelectric layer in our devices excludes both TER and magnetoelectric coupling as possible explanation. It should, however, be noted that our devices and those from Refs. 1 and 2 have a La0.7Sr0.3MnO3 (LSMO) bottom electrode as a common property. FIG. 1: MR traces of a nanosized OSV (LSMO/Alq3/MgO/Co/Ru with 20/12/3/30/10 nm in thickness) for two different resistance states after different voltage pulses taken at 4.3 K. The resistance changes by approx. three decades and the relative MR exhibit a sign reversal from +26% to -38%. Already in 2011 the simultaneous observation of magnetoresistance and resistive switching (RS) has been reported for organic spin valves by Prezioso et al.3,4 In this case the devices were LSMO/Alq3/Co-based spin valves showing a relative MR of -20% in the initial state. By applying 1 voltage pulses the overall device resistance was increased while the relative MR decreased without changing shape or sign. The device resistance could be increased by two decades while the MR was completely suppressed. A possible explanation suggested by Prezioso et al. was the blocking of filaments or charge trapping in combination with giant magnetoresistance (GMR) and spin injection as a prevalent transport mechanism, however, no clear identification of the underlying physics was possible. Recent results from our own group in structures with only one ferromagnetic electrode (LSMO) also demonstrated RS. However, in this case the RS could clearly be linked to the modification of the LSMO surface which creates and modifies a tunnel barrier between the LSMO and the organic semiconductor.5 The modification was shown to originate from the creation of mobile oxygen vacancies in the LSMO. In these experiments the MR could clearly be identified as tunneling anisotropic magnetoresistance (TAMR). Any increase in device resistance also resulted in increasing MR with a maximum value of 20%. Similar to the OSV from Prezioso et al. the devices presented here have a second magnetic electrode. In recently published experiments we have already identified the origin of the magnetoresistance in our devices as tunneling via pinholes through the Alq3 layer resulting in TMR.6,7 Also with respect to the interplay of RS and MR they differ considerably from the OSVs by Prezioso et al. as we do not only observe a change in magnitude but also a sign change of the MR. All samples are fabricated using our recently reported technique7 which allows to define perpendicular OSVs with nanosized active device area. The samples consist of 7 to 14 devices, respectively, with a layer stack of LSMO/Alq3/MgO/Co/Ru (thicknesses 20/12/3/30/10 nm, respectively). The active device area is lithographically defined by a window of approx. 500 x 500 nm² through an insulating aluminum oxide (AlOx) layer on top of the LSMO. All other active layers are consecutively deposited through different large-area shadow masks by thermal evaporation (Alq3), sputtering (MgO, Co, Ru) and e-beam evaporation (Ti/Au contacts). During the last step the organic is shielded by the shadow mask to avoid any radiation damage.8 Characterization is done in a 4He bath cryostat equipped with a 3D vector magnet. All results presented here were obtained at a temperature of 4.3 K. The resistance is measured at 10 mV bias (applied to the cobalt top contact) using a current amplifier. For RS effects short voltage pulses (200- 500 ms) of up to ±2.5 V are applied, followed by a resistance measurement. For selected resistance values an I/V curve is taken and an MR scan is performed. A total of 69 devices with nanosized active area on 15 different samples, each sample with respective different fabrication details were fabricated and characterized. Similar to Ref. 7 we achieve a yield of approx. 55% of working devices, while the other devices show an immeasurably high (>100 GΩ) or very low (<5 kΩ) resistance. This variation of the electrical device properties among a series of samples has recently been discussed7 and indicates that tunneling through pinholes is the dominating transport mechanism. The presence of TMR and the absence of GMR are also confirmed by Hanle-measurements which are routinely performed on all devices.6 For more than half of the working devices it is possible to change the device resistance reversibly by applying voltage pulses of different height and polarity and all of these devices exhibit MR. Though strong variations in total resistance and MR appear from device to device as mentioned above the qualitative behavior is comparable and is described below using two examples. 2  FIG. 2. Resistance versus pulse height (XY plane with its projection to the bottom plane) for two nanosized (500 x 500 nm²) perpendicular OSVs with dAlq3 ≈ 12 nm. A closed loop over 3 decades of resistance (5 for device 'B') is possible. The bars indicate the maximum relative MR at selected points of the loop. The MR is always negative except for a single positive result close to the maximum resistance state (see Fig. 3 for more details). All measurements were done at 4.3 K. Fig. 2 shows the changes of the device resistance and the relative MR for a representative device (Device A) when voltage pulses between -1.6 V and +1.3 V are applied. The initial resistance before any pulse (8.1 MΩ) only changes when Vpulse<-0.7 V. Further decrease of the pulse voltage down to - 1.6 V results in a decrease of resistance by more than two decades (≈50 kΩ). From this point on the pulse voltage is increased. Massive increase of resistance of more than three decades can be observed for Vpulse>+0.5 V, resulting in 270 MΩ after Vpulse=+1.3 V. Decreasing the pulse voltage again shows no effect down to -1.0 V when the resistance decreases again in a sharp switching event, finally reaching the lowest state (Vpulse=-1.6 V). We thus obtain a closed hysteresis loop which can be repeatedly driven through nearly 4 decades of resistance. At the two threshold values of Vpulse the resistance rises smoothly (≈+0.5 V) but decreases in an abrupt manner (≈-1.0 V). The loop was not extended further in order to avoid damage to the device by excessive voltages. Another device 'B' is located on another sample, however, fabricated using the same parameters. Here the resistance change extends over more than 5 decades (right side of Fig. 2), although the general behavior is the same as for device 'A'. The corresponding threshold voltages are slightly different and the measured resistances exhibit a distinct fluctuation for higher Vpulse, nevertheless a closed loop can be measured. At selected points of the hysteresis loop (marked with colored bars along the MR axis in Fig. 2) MR measurements have been carried out. The bars indicate the respective relative magnitude of the MR. Already here it is visible that the MR not only changes its amplitude but also its sign, much in contrast to results of Prezioso et al.3,4 and Grünewald et al.5 The corresponding full MR traces are plotted in Fig. 3. The value for the maximum MR for each trace relates to the height and sign of the corresponding bar with the same color in Fig. 2. 3  FIG. 3. MR traces of one magnetic field sweep direction for different voltage pulses (bottom to top) at T = 4.3 K for two devices 'A' and 'B', respectively. The MR traces are captured directly after resistance measurements which are depicted in Fig. 2 (according color code). The size, coercive fields and even the sign of the MR changes for both devices. The black curve for device 'A' (bottom left in Fig. 3) is taken before any pulse is applied. The MR trace is typical for an OSV as it displays a negative MR (-19.8%) with two distinct resistance values and two clear coercive fields (+2 mT and +20 mT). The back sweep looks similar with reversed magnetic fields and is not shown here. Asymmetries in MR traces (compare to Fig. 1) often occur. They can also be seen elsewhere1 and usually no clear explanation is given. In our devices they may for example be attributed to different pinning of domain walls between opposite magnetic field sweep directions, which could be related to the inhomogeneous Co top electrode or to a very sensitive probing of the complex density of states in the LSMO which depends on the exact direction of the magnetization with respect to the lattice and which may not vary uniformly during the reorientation process. While the device resistance is changed by the voltage pulses (between -1.6 V and +1.3 V, bottom to top in Fig. 3 for device 'A') also all three features of the MR, namely the sign, the relative magnitude, and the coercive fields change. It should, however, be noted that all MR traces have a negative sign, except after Vpulse=+1.2 V where a pronounced positive MR signal of +23.5% can be observed. It is also noteworthy that in all MR traces the lower coercive field remains unchanged (≈2 mT) while the upper coercive field varies (17-140 mT). In some cases the switching at higher fields even occurs in several steps. Starting with an MR of -20% (before any voltage pulse) it reaches -70% after the first pulse (-1.6 V). Successively a reduction (from -30% to <-4%) is observed until at +1.2 V even the sign is reversed (+23.5%) before the MR is finally reduced to almost zero. After closing the hysteresis loop and setting the device back to the initial low resistance state the MR also goes back to -28%, however, it then lacks the step-wise switching which could be observed in the beginning of the loop. Qualitatively the same features as explained above can be observed for device 'B' (right side of Fig. 3) with two noticeable differences. Firstly, a MR trace can only be observed after an initial -2.0 V pulse and secondly, both coercive fields vary for the different states. It does not surprise that repeating the complete measurement cycles does not yield the exact same values of the MR, coercive fields and threshold voltages for the sign reversal as the preparation of each resistance state depends on the samples' history. Especially shape and magnitude of the MR can strongly depend on minor changes of the procedure while the overall qualitative behavior during similar cycles is identical. Indeed this 4 directly supports our line of arguments about the presence of tunneling and its high sensitivity to any change of material properties. We expect the RS in our device to originate from a tunnel barrier which is modified by a voltage pulse similar to the one described in Ref. 5. In order to analyze the barrier properties first I/V characterization is performed at those points of the hysteresis loop where also the MR is characterized. Small excitation voltages (±10 mV) are used in order to avoid modifications of the barrier during the measurement. The I/V curves are non-linear and the dI/dV curves exhibit a strongly parabolic character in the voltage range of approx. ±4 mV. This parabolic behavior is a typical indicator for tunneling processes as will be discussed below. The obtained curves, however, exhibit a certain asymmetry indicating that in contrast to Ref. 5 they are not caused and cannot be modeled by a simple rectangular barrier. In the following explanation this is taken into account by considering two adjacent barriers with different origin and voltage dependence. First we consider the tunneling through the Alq3 layer at pinholes which we do model by a fixed barrier for a given device and which is the major cause of TMR. But secondly our devices also exhibit RS which modulates the MR. This effect is well investigated by Grünewald et al.5 and has successfully been modeled by a variable rectangular tunneling barrier created by oxygen vacancies at the Alq3/LSMO interface. These positive oxygen vacancies are formed at the LSMO/Alq3 interface. They are mobile and can easily be moved by electric fields. During a hysteresis loop vacancies and interstitials of oxygen ions are separated and redistributed within a certain depth from the interface which results in a variable tunnel barrier and causes the change in resistance which can be considered as a local metal-to-insulator transition.9 It should be noted that the bulk of the LSMO electrode itself behaves fully metallic. Its in-plane resistance is in the range of a few kΩ and is slightly reduced during cool-down from room temperature to 4.3 K. This second effect is implemented here by an additional second variable tunneling barrier in the same way as Grünewald et al. did for their devices. Together with the aforementioned fixed Alq3 barrier we thus obtain a transport model of two adjacent tunnel barriers that we can simulate with a suitable fitting model, which is explained later. While in Ref. 5 the variable barrier resulted in additional TAMR only, here we observe a much stronger effect. The features of the TMR now depend on the spin dependent coupling between the LSMO and the Co through the barrier. As the LSMO surface is part of the barrier its modification can change this coupling which can result in massive changes of the TMR, even in a sign reversal. Both barrier thickness10 and electrode surface11 are known to influence TMR with organic spacers in magnitude and sign. A modification below the LSMO surface can even lead to a modified magnetization reversal which may account for the observed changes in coercive field. In order to implement the above explained transport model into an I/V fitting model with two adjacent tunnel barriers we use an expanded Simmons model12 (with error corrected formulas). Each rectangular barrier is defined by the thickness, height, permittivity and charge carrier mass. For the first barrier all parameters are kept constant at reasonable values to account for the fixed Alq3 barrier for the different resistance states in one hysteresis loop (0.9 nm, 21.9 meV, 50 x ε0 and 3.5 times the electron mass, respectively). For the second barrier at the surface of the LSMO, however, voltage pulses modify the barrier properties and cause the resistive switching. Therefore these parameters are kept free within adequate limits. As there is no global optimum for all nine fitting parameters any results need to be considered with caution and merely to extract trends rather than exact quantities. 5  FIG. 4. Experimental data (circles) for I/V curve (black) and differential resistance (red) together with simulated curves for a simple Simmons fit (light colored curves) and for the two-barrier-model (dark colored curves) after a +1.2 V pulse for device 'A'. The latter one respects for asymmetries and achieves a better compliance with the measurement. In Fig. 4 the experimental I/V and differential resistance curves are displayed after a +1.2 V pulse together with a simple Simmons fit (light colored) and a fit for the two-barrier-model (dark color). Obviously a better compliance is possible for the two-barrier-model especially as the asymmetry of the I/V characteristics is taken into account. FIG. 5. Thickness and height of the second tunnel barrier versus the applied pulse voltage for device 'A'. Values have been extracted from data fitting using the expanded (and corrected) Simmons model12 with two barriers. Parameters of the first barrier are fixed, most notably the thickness at 0.89 nm. The variable thickness and height of the second barrier (1.5 - 5.8 nm) are strongly influenced by different voltage pulses. Arrows indicate the sequence of the measurement in the resistance hysteresis loop beginning at 'start'. For the variable barrier the most significant parameter which influences the transmission is the barrier thickness, for which we plotted the calculated value of each fit for different voltage pulses in Fig. 5. In our model allowed values are between 0.1 - 20 nm, while barrier height, effective mass of charge carriers and dielectric constant are allowed to vary within the ranges of 9.4 meV - 4 eV, 10 - 100 x ε0 and 3 - 4 times the electron mass, respectively. According to the fit the barrier thickness changes from 1.5 nm to 5.8 nm and shows a clear trend of increasing thickness with higher pulse voltages. For the resetting pulse (-1.6 V) the barrier thickness also resets to a lower value of about 2.3 nm. These results support the theory about a variable tunnel barrier in the LSMO in addition to the fixed tunnel barrier of the Alq3 pinholes. 6  Although the explanation given above is plausible other explanations shall briefly be discussed. The Co contact itself is a possible candidate, however, nothing indicates a contribution from this side while we know that the LSMO contact can exhibit resistive switching.13–16 In addition RS in oxides has often been described17–21 while for metallic cobalt RS has not been reported. Modification of conducting filaments in the Alq3 can be excluded because we already know that the MR is based on tunneling. Other modifications of the Alq3 cannot be completely excluded, though no suitable mechanism comes to mind. In reference to the explanation of the MR and RS effects provided in Refs. 3 and 4 our alternative explanation is entirely evident as the basic magnetoresistive functionality of our devices differs considerably. In addition our theory can explain the finding that an MR effect in device 'B' could only be measured after applying a negative voltage pulse by the need to first sufficiently create and separate oxygen vacancies and interstitials to realize a suitable barrier for the occurrence of tunneling assisted MR effects as described in Ref. 5 again supporting the presence of a similar mechanism. Furthermore, recent measurements on similar devices containing metal free phthalocyanine (H2Pc) as spacer layer which will be published elsewhere show very similar results in terms of MR and RS properties supporting that the effect originates from the electrode rather than from properties of the organic semiconductor. In summary we have fabricated nanosized perpendicular organic spin valve devices which exhibit a unique interplay of resistive switching and magnetoresistance effects up to now primarily observed in multiferroic tunnel junctions. Transport occurs locally through pinholes providing a tunneling path through the organic material. The MR thus originates only from TMR and possibly from small contributions from TAMR. The tunnel barrier consists of two parts, namely the organic semiconductor in the pinhole (fixed barrier) and a tunnel barrier induced by oxygen vacancies at the LSMO surface which can be modified by a voltage pulse leading to resistive switching and modifications of the MR (variable barrier). These devices demonstrate the potential multifunctionality of LSMO in transport structures and present the opportunity for multi-state memory because in contrast to Prezioso et al.3,4 the MR persists in different resistive switching states. A major challenge, however, for applications will be to harness the MR which is currently based on pinholes and thus difficult to control. Acknowledgments This work was supported by the European Commission within the 7FP project HINTS (Project No. NMP-CT-2006-033370) and by the DFG in the SFB 762. 7 (2011). 1V. Garcia, M. Bibes, L. Bocher, S. Valencia, F. Kronast, A. Crassous, X. Moya, S. Enouz-Vedrenne, A. Gloter, D. Imhoff, C. Deranlot, N.D. Mathur, S. Fusil, K. Bouzehouane, and A. Barthelemy, Science 327, 1106 (2010). 2D. Pantel, S. Goetze, D. Hesse, and M. Alexe, Nat. Mater. 11, 289 (2012). 3M. Prezioso, A. Riminucci, I. Bergenti, P. Graziosi, D. Brunel, and V.A. Dediu, Adv. Mater. 23, 1371 4M. Prezioso, A. Riminucci, P. Graziosi, I. Bergenti, R. Rakshit, R. Cecchini, A. Vianelli, F. Borgatti, N. Haag, M. Willis, A.J. Drew, W.P. Gillin, and V.A. Dediu, Adv. Mater. 25, 534 (2013). 5M. Grünewald, N. Homonnay, J. Kleinlein, and G. Schmidt, Phys. Rev. B 90, 205208 (2014). 6M. Grünewald, R. Göckeritz, N. Homonnay, F. Würthner, L.W. Molenkamp, and G. Schmidt, Phys. 7R. Göckeritz, N. Homonnay, A. Müller, T. Richter, B. Fuhrmann, and G. Schmidt, Appl. Phys. Lett. 8J. Rybicki, R. Lin, F. Wang, M. Wohlgenannt, C. He, T. Sanders, and Y. Suzuki, Phys. Rev. Lett. 109, 9P. Orgiani, A.Y. Petrov, R. Ciancio, A. Galdi, L. Maritato, and B.A. Davidson, Appl. Phys. Lett. 100, 10S. Mandal and R. Pati, ACS Nano 6, 3580 (2012). 11S.W. Jiang, B.B. Chen, P. Wang, Y. Zhou, Y.J. Shi, F.J. Yue, H.F. Ding, and D. Wu, Appl. Phys. Lett. Rev. B 88, 85319 (2013). 106, 102403 (2015). 76603 (2012). 42404 (2012). 104, 262402 (2014). 12M.G. Chapline and S.X. Wang, J. Appl. Phys. 101, 83706 (2007). 13Y.W. Xie, J.R. Sun, D.J. Wang, S. Liang, and B.G. Shen, J. Appl. Phys. 100, 33704 (2006). 14L. Huang, B. Qu, L. Liu, and L. Zhang, Solid State Communications 143, 382 (2007). 15C. Moreno, C. Munuera, S. Valencia, F. Kronast, X. Obradors, and C. Ocal, Nano Lett. 10, 3828 (2010). (2013). 16D. Liu, N. Wang, G. Wang, Z. Shao, X. Zhu, C. Zhang, and H. Cheng, Appl. Phys. Lett. 102, 134105 17G. Dearnaley, A.M. Stoneham, and D.V. Morgan, Rep. Prog. Phys. 33, 1129 (1970). 18C.-Y. Lin, C.-Y. Wu, C.-Y. Wu, C. Hu, and T.-Y. Tseng, J. Electrochem. Soc. 154, G189 (2007). 19R. Waser and M. Aono, Nat. Mater. 6, 833 (2007). 20J. Maier, K. Kern, R. Waser, R. Dittmann, G. Staikov, and K. Szot, Adv. Mater. 21, 2632 (2009). 21K. Nagashima, T. Yanagida, K. Oka, M. Taniguchi, T. Kawai, J.-S. Kim, and B.H. Park, Nano Lett. 10, 1359 (2010). 8 

1201.3115

2

1201

Measurement and simulation of anisotropic magnetoresistance in single GaAs/MnAs core/shell nanowires

We report four probe measurements of the low field magnetoresistance in single core/shell GaAs/MnAs nanowires synthesized by molecular beam epitaxy, demonstrating clear signatures of anisotropic magnetoresistance that track the field-dependent magnetization. A comparison with micromagnetic simulations reveals that the principal characteristics of the magnetoresistance data can be unambiguously attributed to the nanowire segments with a zinc blende GaAs core. The direct correlation between magnetoresistance, magnetization and crystal structure provides a powerful means of characterizing individual hybrid ferromagnet/semiconductor nanostructures.

cond-mat.mes-hall

cond-mat

Measurement and simulation of anisotropic magnetoresistance in single GaAs/MnAs core/shell nanowires J. Liang,1, 2 J. Wang,1, 2 A. Paul,1, 3 B.J. Cooley,1, 2 D. W. Rench,1, 2 N. S. Dellas,1, 3 S. E. Mohney,1, 3 R. Engel-Herbert,1, 3 and N. Samarth1, 2, a) 1)Materials Research Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, USA 2)Department of Physics, The Pennsylvania State University, University Park, Pennsylvania 16802, USA 3)Department of Materials Science and Engineering, The Pennsylvania State University, University Park, Pennsylvania 16802, USA (Dated: 6 June 2021) We report four probe measurements of the low field magnetoresistance in single core/shell GaAs/MnAs nanowires synthesized by molecular beam epitaxy, demonstrating clear signatures of anisotropic magne- toresistance that track the field-dependent magnetization. A comparison with micromagnetic simulations reveals that the principal characteristics of the magnetoresistance data can be unambiguously attributed to the nanowire segments with a zinc blende GaAs core. The direct correlation between magnetoresistance, magnetization and crystal structure provides a powerful means of characterizing individual hybrid ferromag- net/semiconductor nanostructures. PACS numbers: 75.75.-c,75.78.Cd,85.75.-d The incorporation of spin-related functionality into semiconductor nanostructures provides an exciting new route for nanospintronic devices.1 Nanodevices derived from MnAs/GaAs heterostructures present an interest- ing opportunity in this context because GaAs is an im- portant semiconductor for optoelectronics, while MnAs is a ferromagnetic metal with a Curie temperature above room temperature (∼313-350 K, depending on the strain). Indeed, MnAs/GaAs heterostructures have ex- cellent compatibility with commonly used semiconductor devices.2,3 In addition, MnAs is a fundamentally inter- esting ferromagnet because of the unique competing in- terplay between the magnetocrystalline anisotropy and the shape anisotropy.4 Recent work has shown that the heteroepitaxy of MnAs on GaAs can also be realized in core/shell nanowires (NWs).5 -- 7 Such NWs constitute a novel arena for studying magnetization dynamics in restricted nanoscale geometries. However, probing the magnetization in such individual NWs is a challenge for conventional magnetometry techniques. Here, we use magnetoresistance (MR) measurements of single NW de- vices in conjunction with micromagnetic simulations to gain insights into the magnetization switching process of core/shell GaAs/MnAs NWs. The methodology pre- sented here is also applicable to other hybrid core/shell semiconductor/ferromagnet NWs of current interest.8 -- 10 The core/shell NW samples studied here were synthe- sized on GaAs (111)B substrates in an EPI 930 molecular beam epitaxy chamber. We used a catalyst-free growth technique for the GaAs NWs,11 followed by thin film growth of a MnAs shell, as detailed in an earlier report.6 This growth technique contrasts with other approaches a)Electronic mail: [email protected] wherein GaAs/MnAs core/shell NWs are synthesized us- ing a Au catalyst.5,7 Additionally, we note that the epi- taxial orientation relationship between GaAs and MnAs is different from NWs synthesized using a Au catalyst due to the different crystal structures of the GaAs core. This creates a difference in magnetocrystalline anisotropy that has a significant impact on the magnetic domain struc- ture of the MnAs shell and the low field magnetotrans- port properties. Figure 1(a) shows a cross-sectional transmission elec- tron microscope image of a single core/shell NW with a zinc-blende (ZB) GaAs core of ∼200 nm diameter. The MnAs shell thickness is estimated to be ∼10 nm. The GaAs core is mostly in the ZB structure with small seg- ments of the wurtzite (WZ) phase; the MnAs shell is crystalline with a hexagonal NiAs structure. For the seg- ments of the NW with ZB core the growth direction is along the [111] direction with six facets belonging to the {110} family. The c-axis (hard axis) of MnAs lies in plane with the NW facets, at an angle of ∼±53◦ with respect to the wire axis. The c-axis of MnAs mirrors itself on adjacent facets. For the WZ part of the NW, the growth direction is along [001] and the c-axis of MnAs is along the NW axis.6 We ultrasonically removed the GaAs/MnAs core/shell NWs from the substrate and dispersed them onto a Si/Si3N4 substrate. The sample was then transferred into a dual-beam focused ion beam (FIB) system (FEI Quanta 200 3D) with in situ scanning electron microscopy (SEM) capabilities. After oxide layer milling, we deposited four Pt electrodes on single GaAs/MnAs core/shell NWs for electrical measurements. We minimized the Ga+ ion imaging time to reduce contamination and also kept the Ga+ ion deposition current and chamber pressure low to minimize the spreading of Pt. Figure 1(b) shows an 2 1 0 2 n a J 0 2 ] l l a h - s e m . t a m - d n o c [ 2 v 5 1 1 3 . 1 0 2 1 : v i X r a 2 FIG. 1. (a) Cross-sectional TEM image of a GaAs/MnAs core/shell NW. The GaAs core is in the ZB phase. (b) Single GaAs/MnAs core/shell NW device contacted by four FIB- assisted Pt electrodes, with L = 1.96 µm. (c) GaAs/MnAs core/shell structure used in micromagnetic simulations. FIG. 2. (a) Resistivity ρ of the device in Fig. 1(b) as a func- tion of the temperature with an excitation current of 0.5 µA. (b) Resistivity vs. temperature (plotted on a log scale). Ex- perimental MR loops with a magnetic field applied (c) parallel and (d) perpendicular to wire axis. SEM image of a typical device. Subsequent electrical transport measurements were carried out in a Quantum Design Physical Properties Measurement system using a standard four-probe AC resistance bridge. We made sure that the contacts are ohmic and used a typical ex- citation current of 0.5 µA. We measured the MR over a temperature range 500 mK to 300 K and in magnetic fields up to 80 kOe. In total, we fabricated and measured four devices. In this Letter, we focus on data from only one of these devices; the other devices show qualitatively similar behavior. In addition, we carried out control mea- surements using a bare GaAs NW device (i.e. without any MnAs shell) using the same FIB contacting technique with Pt electrodes. This control experiment shows that the nominally undoped GaAs core is highly insulating. Thus, the GaAs core merely serves as a NW template that supports the metallic MnAs shell which dominates the transport data discussed in this paper. Figure 2(a) shows the temperature-dependent resistiv- ity of a GaAs/MnAs core/shell NW with length ∼1.96 µm from room temperature (∼298 K) down to ∼500 mK. The resistivity of the MnAs shell was ∼ 5 × 10−4 Ω·cm at room temperature, which is ∼5 times higher than a typical MnAs epilayer grown on GaAs(001). The tem- perature dependence of the resistivity shows metallic be- havior, similar to that of a MnAs epilayer between room temperature and 20 K. However, the somewhat smaller residual resistivity ratio (defined as the ratio of the re- sistivity at 300 K to that at 4.2 K) indicates that these MnAs shells are more disordered than epitaxial films of similar thicknesses.12 Below T ≈ 15 K, the resistivity in- creases with decreasing temperature (Fig. 2(b)). Anal- ysis to be reported elsewhere shows a temperature de- pendence of the conductivity σ ∼ ln T , consistent with the onset of localization in a diffusive two dimensional system. The saturation of the resistivity at even lower temperatures (T (cid:46) 1.4 K), shown in Fig. 2(b), is not yet understood, but could arise either from trivial heating effects or from more interesting dimensional crossover as the relevant length scales (such as the phase breaking length) increase with lowering temperature. Figures 2(c) and 2(d) show the MR of a NW device measured at low temperature (T = 10 K) in magnetic fields applied parallel and perpendicular to the wire axis, respectively. These measurements were carried out af- ter first saturating the magnetization of the MnAs shell at 80 kOe. We observe a hysteretic MR at low fields, superimposed upon a linear negative MR that does not saturate even at the highest fields used in the present measurements (high field data not shown). The physi- cal origin of this interesting non-saturating MR will be discussed elsewhere. We focus here on the low field MR. For magnetic fields along the wire axis, as we decrease the magnitude of the field from its maximum value to zero, the resistance initially increases (linear background) and reaches a maximum after field reversal at ∼3.5 kOe (Fig. 2(c)). The MR then abruptly changes, indicating domain switching. From the epitaxial relationship be- tween the MnAs shell and the GaAs core, we conclude that this MR feature originates from wire segments with a ZB core. Our reasoning is as follows: for segments with a WZ core, the MnAs hard axis is along the wire axis; the strong magnetocrystalline anisotropy energeti- cally disfavors magnetization in that direction and chang- ing the applied field in that direction would not give rise to abrupt changes in the MR. Figure 2(d) shows the MR with the magnetic field perpendicular to the wire axis. Here, the MR is more complex: as we decrease the mag- nitude of the field from its maximum value to zero, the resistance again initially increases (linear background); the resistance then drops sharply after field reversal to a minimum at ∼8 kOe before recovering at ∼4 kOe. This behavior suggests a two step reversal process. Again, contributions from wire segments with a WZ core cannot cause the abrupt changes observed in the MR data: the magnetization would be exclusively in the cross section of the WZ wire segments favored by both the external applied field and magnetocrystalline anisotropy. Both hysteretic effects in the measured MR loops per- sist up to room temperature and are classic signatures of anisotropic magnetoresistance (AMR). Since AMR is di- rectly connected to the magnetization of a ferromagnetic sample, we can exploit the effect as a sensitive probe of the field induced rotation and switching of the magneti- zation in these nanostructures.13 To further gain insight into the magnetization reversal process of GaAs/MnAs NWs we carried out micromagnetic simulations. The magnetic domain structure of MnAs thin films has al- ready been the subject of micromagnetic studies using finite difference based solvers.4,14 This approach how- ever is not suitable due to the geometry of the core/shell structure. We therefore employed the open source fi- nite element code MAGPAR15 to avoid erroneous re- sults from the staircase approximation.16 The domain configuration was calculated using the Landau-Lifshitz- Gilbert equation with the damping constant α = 0.1. The following micromagnetic parameters for MnAs were used: exchange stiffness constant A = 1 × 10−11 J/m, uniaxial magnetocrystalline anisotropy constant Ku = −7.2 × 105 J/m3, and saturation magnetization Ms = 8 × 105 A/m. These parameters were successfully em- ployed previously to simulate the magnetic domain struc- ture of MnAs thin films grown on GaAs(001)4 and on GaAs(111) substrates.14 We varied MnAs shell thickness and GaAs core diameter from 10 to 20 nm and from 120 to 160 nm, respectively. Here we present the results of a 10 nm thick MnAs shell on a ZB GaAs core (160 nm in diameter) being closest to the NW geometry investi- gated. The core/shell NW geometry caused a very large boundary element matrix and we therefore limited the wire length to 250 nm. The finite element mesh was gen- erated using GMESH.17 The average edge length of the tetrahedral elements around 5.5 nm was chosen close to the micromagnetic exchange lengths and the field step- ping was as small as 25 Oe. Figure 3(a) shows the simulated hysteresis curve with the magnetic field applied along the wire axis. The hys- teresis shows a single domain reversal with a coercive field of 4.85 kOe in good agreement with the measure- ments. The energy barrier separating the two stable do- 3 FIG. 3. Simulated magnetic hysteresis curves of GaAs/MnAs core shell nanowires with a ZB core and magnetic fields ap- plied (a) along the wire axis (z) and (b) perpendicular to the wire axis (x). In panel (b), we show hysteresis loops for the magnetization component along the applied field direction x and perpendicular to the wire axis M⊥, and along the wire axis M(cid:107). AMR effect extracted from the simulated hysteresis curves for magnetic fields (c) parallel and (d) perpendicular to the wire axis. main states is caused by the magnetocrystalline hard axis being inclined with the wire axis. We used the simulation results as the input to the standard heuristic description of AMR:18 ρ = ρ⊥ +(cid:0)ρ(cid:107) − ρ⊥(cid:1) cos2 ϕ, (1) where ρ(cid:107) and ρ⊥ are the longitudinal and transversal resistivities with respect to magnetization and ϕ the an- gle between the magnetization and current density. For the NW, the angle is given by the normalized magneti- zation along the wire axis: cos ϕ = M(cid:107). Note that for MnAs ρ(cid:107) < ρ⊥.19 Figure 3(c) shows that the AMR effect calculated by Eqn. 1 is in very good agreement with the measured MR loop in Fig. 2(c), reproducing shape and coercive field well. Note that although there is a reduc- tion in MR with increasing field, it does not explain the linear background at higher magnetic fields. The overes- timation in coercive field is attributed to an underestima- tion of the MnAs shell width. Simulations with varying NW geometries showed a decrease in coercive field with increasing shell thickness. The experimental determina- tion of the MnAs thickness from the TEM image was challenging due to the low contrast between the core and the shell and is held responsible for this discrepancy. Figures 3(b) and 3(d) show the simulated hysteresis curves and AMR loops for magnetic fields applied per- pendicular to the wire axis [x-axis, see Fig. 1(c)]. Two magnetization curves, M⊥ in the applied field direction and M(cid:107) along the wire axis, are overlayed. The hys- teresis reveals two discontinuous changes at 2.1 kOe and 3.4 kOe: the two facets aligned within the applied field [upper and lower facet, cf. Fig. 1(c)] reverse at a lower field, whereas all other facets that are inclined to the ap- plied field direction switch at the higher field value at once. The increase in magnetization along the wire axis M(cid:107) between the two switching events originates from the epitaxial relationship, i.e., the inclination of the hard axis with respect to the wire axis and the alternation of the angle in adjacent facets. Reducing the magnetic field from saturation, the magnetization in the facets deviates from the x-direction and aligns in the respective facet planes, perpendicular to the "local" hard axis, to mini- mize demagnetization and magnetocrystalline anisotropy energies. This causes M(cid:107) of the upper and lower facet to be antiparallel with M(cid:107) of all other facets, which in turn gives rise to a small net M(cid:107) of the wire. Revers- ing M⊥ is accompanied with reversing M(cid:107). If the upper and lower facets reverse, M(cid:107) has the same orientation in all facets between the two switching fields, giving rise to a large M(cid:107). At larger fields the magnetization of the slanted facets reverses, reducing the net M(cid:107). The cal- culated switching fields are smaller than measured. A possible source of error is the alignment of the NW in the applied field. Although a sufficiently precise line-up of the nanowire axis was easily achieved, the control over the azimuthal orientation is challenging and might cause the discrepancy. Disregarding the linear background, the measured MR is well reproduced by the AMR effect de- rived from the magnetization curve M(cid:107). A two-step re- versal process and the reduction in MR between the two reversal field steps due to a large M(cid:107) are correctly pre- dicted. In conclusion, low field MR measurements of single GaAs/MnAs core/shell NWs reveal the AMR effect of the wire segments with a ZB GaAs core superimposed on a linear background. Both MR loops measured for fields perpendicular and parallel to the wire axis are well reproduced by micromagnetic simulations, even for a rel- 4 atively complex geometry. The combination of MR mea- surements with micromagnetic simulations thus provides a powerful means to gain insight into the domain struc- ture and dynamic properties of functional ferromagnetic nanostructures. This work is supported by the Penn State Center for Nanoscale Science under NSF DMR-0820404, and by NSF ECCS-0609282, ONR N00014-09-1-0221, and the Penn State Nanofabrication Facility NSF NNUN ECCS- 35765. 1D. D. Awschalom and M. E. Flatte, Nature Phys. 3, 153 (2007). 2M. Tanaka, Semicond. Sci. Tech. 17, 327 (2002). 3L. Daweritz, Rep. Prog. Phys. 69, 2581 (2006). 4R. Engel-Herbert, T. Hesjedal, and D. M. Schaadt, Phys. Rev. B 75, 094430 (2007). 5M. Hilse, Y. Takagaki, J. Herfort, M. Ramsteiner, C. Herrmann, S. Breuer, L. Geelhaar, and H. Riechert, Appl. Phys. Lett. 95, 133126 (2009). 6N. S. Dellas, J. Liang, B. J. Cooley, N. Samarth, and S. E. Mohney, Appl. Phys. Lett. 97, 072505 (2010). 7Y. Takagaki, J. Herfort, M. Hilse, L. Geelhaar, and H. Riechert, J. Phys.: Cond. Matter 23, 126002 (2011). 8A. Rudolph, M. Soda, M. Kiessling, T. Wojtowicz, D. Schuh, W. Wegscheider, J. Zweck, C. Back, and E. Reiger, Nanoletters 9, 3860 (2009). 9C. Gao, R. Farshchi, C. Roder, P. Dogan, and O. Brandt, Phys. Rev. B 83, 245323 (2011). 10C. H. Butschkow, E. Reiger, S. Geissler, A. Rudolph, M. Soda, and D. Weiss, D. Schuh, G. Woltersdorf, W. Wegscheider, arXiv:1110.5507. 11C. Colombo, D. Spirkoska, M. Frimmer, G. Abstreiter, A. Fontcuberta i Morral, Phys. Rev. B 77, 155326 (2008). and 12J. J. Berry, S. J. Potashnik, S. H. Chun, K. C. Ku, P. Schiffer, and N. Samarth, Phys. Rev. B 64, 052408 (2001). 13J.-E. Wegrowe, D. Kelly, A. Franck, S. E. Gilbert, and J.-P. Ansermet, Phys. Rev. Lett. 82, 3681 (1999). 14R. Engel-Herbert, T. Hesjedal, D. M. Schaadt, L. Daweritz, and K. H. Ploog, Appl. Phys. Lett. 88, 052505 (2006). 15W. Scholz, J. Fidler, T. Schrefl, D. Suess, R. Dittrich, H. Forster, and V. Tsiantos, Comp. Mat. Sci 28, 366 (2003). 16M. Donahue and R. McMichael, IEEE Trans. Mag. 43, 2878 (2007). 17C. Geuzaine and J.-F. Remacle, Internat. J. Numer. Methods Engrg. 79, 1309 (2009). 18T. McGuire and R. Potter, IEEE Trans. Mag. 11, 1018 (1975). 19Y. Takagaki and K.-J. Friedland, J. Appl. Phys. 101, 113916 (2007). 

1510.09176

1

1510

Electron-phonon coupling in suspended graphene: supercollisions by ripples

Using electrical transport experiments and shot noise thermometry, we find strong evidence that "supercollision" scattering processes by flexural modes are the dominant electron-phonon energy transfer mechanism in high-quality, suspended graphene around room temperature. The power law dependence of the electron-phonon coupling changes from cubic to quintic with temperature. The change of the temperature exponent by two is reflected in the quadratic dependence on chemical potential, which is an inherent feature of two-phonon quantum processes.

cond-mat.mes-hall

cond-mat

Electron-phonon coupling in suspended graphene: supercollisions by ripples Antti Laitinen,1 Mika Oksanen,1 Aurélien Fay,1 Daniel Cox,1 Matti Tomi,1 Pauli Virtanen,1 and Pertti J. Hakonen1, ∗ 1Low Temperature Laboratory, Aalto University, P.O. Box 15100, FI-00076 AALTO, Finland Abstract Using electrical transport experiments and shot noise thermometry, we find strong evidence that "supercollision" scattering processes by flexural modes are the dominant electron-phonon energy transfer mechanism in high-quality, suspended graphene around room temperature. The power law dependence of the electron-phonon coupling changes from cubic to quintic with temperature. The change of the temperature exponent by two is reflected in the quadratic dependence on chemical potential, which is an inherent feature of two-phonon quantum processes. 5 1 0 2 t c O 0 3 ] l l a h - s e m . t a m - d n o c [ 1 v 6 7 1 9 0 . 0 1 5 1 : v i X r a 1 Electron-phonon coupling is the basic means to control energy transport in a variety of devices which provide extreme sensitivity in calorimetry, bolometry, and radiation detection (infrared/THz). Owing to thermal noise, such devices are typically operated at cryogenic temperatures, at which the coupling between electrons and phonons becomes weak. Under these conditions, graphene is expected to have an advantage owing to its small heat capacity that allows fast operation even though its electron-phonon coupling becomes exceedingly small near the Dirac point1 -- 7. The weak electron-acoustic phonon coupling in graphene stems from the restrictions in energy transfer caused by momentum conservation7. The maximum change of momentum at the Fermi level is twice the Fermi momentum 2kF , which corresponds to phonon energy ω2kF . This energy defines a characteristic temperature, the Bloch-Grüneisen temperature by ω2kF = kBTBG, above which only a fraction of phonons are available for scattering with electrons in the thermal window. Manifestations of this has been observed in resistance vs. temperature measurements8, especially on electrolytically-gated graphene9. The energy transfer limits due to momentum conservation are circumvented when acoustic phonon scattering is assisted by other scattering processes, a.k.a. supercollision events, or when energy is transferred to flexural phonons, which is a two-phonon process.7,10 -- 16 The flexural phonon mechanism can dominate over the single acoustic phonon process, but to win over supercollision cooling, extremely clean samples are required.10 Moreover, even if a sample is free of impurities, other scattering mechanisms such as static or dynamic ripples can still facilitate supercollisions. e −T δ The heat flow from the conduction electrons/holes to the lattice can be characterized by a power law of the form P = Σ(T δ ph), where Te is the electron temperature, Tph the phonon temperature, Σ the coupling constant and δ a characteristic exponent17. Our monolayer graphene experiments have been conducted near the Dirac point at charge densities n < 0.1−4.5·1011 cm−2, which corresponds to TBG < 42 K for longitudinal acoustic phonons (and even lower for transverse acoustic and flexural phonons). Consequently, all our results have been measured in the regime of T > TBG, where the scattering of electrons from acoustic phonons leads to δ = 1 or δ = 5, depending whether Te << µ or Te > µ, respectively4. Non-conventional pathways for cooling, the supercollision cooling or flexural two phonon processes10, yield δ = 3 or δ = 5 for Te < µ or Te > µ, respectively. In this work, we have employed shot noise thermometry in combination with conduc- 2 tance measurements to determine the electron-phonon coupling in high-quality, suspended graphene monolayers. We demonstrate that cooling in our graphene samples, even though their field effect mobility is µf > 105 cm2/Vs at low temperatures, takes place via supercolli- sion phonon processes at temperatures Te = 200−600 K. We observe a power-law dependence having δ (cid:39) 3 − 5, depending on the magnitude of the chemical potential µ = 10 − 73 meV relative to kBTe. The change in the exponent agrees well with the cross-over behavior be- tween δ = 5 and δ = 3 for the non-degenerate (kBTe > µ) and degenerate cases (kBTe < µ), which also leads to a µ2 dependence seen in the data as a function of chemical potential. We also find that the location of the cross-over between the two regimes is shifted by the Fermi velocity renormalization caused by interactions. For the deformation potential, we obtain D (cid:39) 64 eV over the full range of data. No increased relaxation rate due to optical phonon scattering was found in our experiments at Te < 600 Kelvin. The origin of the supercollision scattering can be understood to be small-scale ripples in the suspended graphene. 1 cm2 , which is close to 6 nF The studied sample, with length L = 1.1 µm and width W = 4.5 µm, was manufactured from exfoliated graphene using e-beam lithography and etching in hydrofluoric acid. Raman spectroscopy was employed to verify that the sample was a monolayer graphene sheet. Before measurements, the graphene was annealed by passing a current of 1.1 mA through the sample (nearly 0.9 V in voltage) which evaporated resist residuals and resulted in a nearly neutral, g = +0.4 V. high-quality sample with the charge neutrality (Dirac) point located at V D The gate capacitance was determined from the parallel model (and verified by the µ scale in Fabry-Pérot interference patterns18): Cg = 4.7 nF cm2 in a similar device reported in Ref. 19. The residual charge carrier density was determined to be n0 = 0.8· 1010 cm2 from log(G) vs log(n) plot (see Fig. 1b) as a point where the linear behaviour levels to a constant value at low charge densities n = Cg(Vg − V D g ). Note that the slope of G(n) on the log-log plot is nearly 1/2 which indicates ballistic behavior. The effective n0 is 20% larger when making a similar log-log plot for the electron-phonon coupling vs. n. The initial slope of G(n) was employed to determine the field effect mobility of µf > 105 cm2/Vs. The inset of Fig. 1b displays the variation of zero-bias resistance R0 = dV /dIV =0 over chemical potentials ranging ±73 meV across the Dirac point. The measured minimum conductivity falls below 4e2 πh for high aspect h approaching theoretical minimum conductivity σmin = 4e2 ratio samples20,21; the actual maximum resistance of the sample was 2.2 kΩ. In addition to IV -characteristics and differential conductance Gd = dI dV , we measured 3 zero-frequency shot noise SI and dS/dI (over 600-900 MHz) calibrated against a AlOx tunnel junction in the same cooldown (see Ref. 22 for details). Poissonian noise is given by Sp = 2q(cid:104)I(cid:105) where (cid:104)I(cid:105) is the average current. The Fano factor defines the noise level SI compared to the Poissonian noise23, F = SI/Sp. In addition to gate voltage, the Fano factor depends also on the bias voltage V . When the bias voltage is increased, electron-electron interaction effects (in the so called "hot electron regime") and inelastic scattering effects start to influence the Fano factor. For diffusive transport in local equilibrium, the Fano factor is related to electronic temperature:24,25 Te = F eV . This equation is the basis of our noise thermometry on graphene electrons. Our experiments were performed on a dilution refrigerator operated typically around 0.5 K in our large bias experiments. For experimental details, we refer to Ref. 18. 2kB Figure 1. a) Schematic view of a suspended graphene sample, where metallic leads contact a graphene flake (top part), and an optical image of the sample (lower part). A layer of silicon oxide, part of which is etched away, separates the circuit from the back gate made of heavily doped silicon. b) Logarithm of the conductivity as a function of the logarithmic charge carrier density n = Cg(Vg − V D g )/e. The solid line extrapolates towards the residual charge carrier density of n0 = 0.8 · 1010 cm−2; the offset of the Dirac point due to charge doping is corrected. The inset displays zero-bias resistance versus Vg. c) Thermal model of the sample: electrons are thermalized to metallic leads via heat diffusion and electron-phonon coupling conduction in series with Kapitza boundary resistance to the leads/substrate phonons. Measured shot noise SI up to high bias is illustrated in Fig. 2a. Near the Dirac point, strong initial increase in SI is found, which reflects a clearly larger Fano factor at small charge densities n0 than at n >> n0. Rather smooth, symmetric increase in SI is observed with bias current over the full range of bias conditions. In our experimentally determined 4 108109101010111012100.3100.6100.9n (1/cm2)G (mS)JouleheatingElectron system, TePhononsystem, TphElectron diffusionInelastic scatteringThermalizationto the leadsLeads, Tleads−1001005001000Vg (V)R (Ω)a)b)c)−505010002000R (Ω)Vg− VDirac (V)e-ph coupling, there is a difference in the prefactor of the power laws obtained at equal hole and electron densities. We assign this to the presence of pn-junctions at Vg > V D g , which undermines the accuracy of our shot noise thermometry: at large bias voltages with enhanced phonon scattering, the noise generated by the pn-junctions is an additive quantity to the local thermal noise, and the resulting electronic temperatures are higher than what they really should be. Consequently, we rely only on the data without pn-junction in our actual analysis. The shot noise temperature is corrected for contact resistance, which increases the measured temperatures by 20% in overall. The Joule heating power to graphene electrons equals to Pe = (V − ∆Vcc)I where ∆Vcc denotes the total voltage drop over the contacts ∆Vcc = RcI, where Rc = 40 Ω. The heat flow paths in the sample, namely diffusive transport and inelastic scattering, that balance Pe are depicted in Fig. 1 c. When the graphene lattice is at liquid helium temperatures or below and the electron system is heated to 200−300 K, the electronic heat will flow mostly to acoustic/flexural phonons at rate P = Qe−ph = Σ(cid:0)T δ (cid:1), not to contacts by electronic e − T δ ph diffusion. Under this Joule heat flux, elevation of the graphene acoustic phonon temperature should remain below a few tens of Kelvins26, and the latter term T δ ph can be neglected. In order to improve the precision in the determination of δ, we subtract the electronic heat conduction to obtain P from Pe. At Te < 100 K, we observe power-law behavior in Pe vs Te with an exponent δ (cid:39) 1.627, but our accuracy is not enough to test modifications of the Wiedemann-Franz law as achieved in Ref.28. g < 0.6 V while, at Vg − V D As indicated by the data on P vs Te in Fig. 2b, we observe for the electron-phonon heat flow a power-law dependence having δ (cid:39) 3 − 5, depending on the chemical potential µ = 10 − 73 meV: we find δ = 5 at Vg − V D g > 5 V, we observe δ = 3. At 0.6 V< Vg − V D g < 5 V, our data display a cross-over from δ = 5 to δ = 3. This cross-over takes place when the chemical potential reaches (cid:39) 20 meV, i.e. when the electronic system starts to become degenerate (kBTe/µ < 1). This finding δ = 5 at µ << kBTe is consistent with single acoustic phonon scattering at high temperatures1,4, but the theoretical expectation for acoustic phonons at µ >> kBTe disagrees with our results. Furthermore, the magnitude of the predicted e-ph coupling for δ = 5 is too small unless an unrealistically large value for the deformation potential is adopted. Moreover, instead of the expected µ-dependence of acoustic phonon scattering (P ∝ µ4 at TBG < T < µ; P ∝ µ0 at TBG, µ < T ), we find µ2-dependence (or more precisely, linear in n, see below) 5 which contradicts single phonon scattering but agrees with flexural phonon and supercollision scattering. Hence, we may rule out single phonon scattering events1,4 as the dominant e-ph coupling in our suspended sample. Figure 2. a) Shot noise vs. the bias current I at a few values of gate voltage near the Dirac point at Vg = +0.4 V (Vg is given in the figure). b) Heat flow to phonons P as a function electron temperature Te = F eV /2kB. The green, blue, and red curves represent measured data while the dashed lines display the theoretical behavior from Eq. 1 with kF (cid:96) = 3 and D = 64 eV; for other parameters, see text. The dotted lines denote power laws with exponents δ = 3 (upper) and δ = 5 (lower). c) Electron-phonon coupling power as a function of the gate-induced charge carrier g = +0.4 V (no pn-junction). The density n at temperature Te = 150, 200, and 250 K for Vg < V D theoretical lines are obtained from Eq. 1 using the same parameters as in Fig. 2b: the lines denote the expansions at µ << kBTe (dashed lines) and at µ >> kBTe (solid lines). The electron-phonon energy flow via supercollisions is predicted in Ref. 10. Taking into account the energy dependence of the density of states in graphene, which leads to the cross-over between the T 3 e power laws, the result reads: √ e and T 5 QSTph=0 (cid:39) − πn vF kBTe N (2π2v2 F )2 g2 Dk5 BkF (cid:96) T 5 e q( ) , (1) where q(z) (cid:39) 9.69 + 1.93z2 for z (cid:28) 5 and q(z) (cid:39) 2ζ(3)z2 ≈ 2.40z2 for z (cid:29) 5, and N = 4 is the degeneracy in graphene. The electron-phonon coupling constant for deformation potential is gD = D/(cid:112)2ρs2 where D is the deformation potential and ρ the mass density of the graphene sheet. The quantity kF (cid:96) is the dimensionless mean free path associated with the additional scattering mechanism enabling the supercollisions. For experimental convenience, we have also here written the above in terms of the particle density n which is kept fixed by gate voltage. We take renormalization of the Fermi velocity by electron- 6 −1−0.500.51x 10−300.511.522.533.5x 10−5I (A)SI (W) +0.4 V−1.2 V−8 Va)01234x 1011020040060080010001200P (W/cm2)n (1/cm2) 150 K200 K250 K200600101102103104Te (K)P (W/cm2) +0.4 V (Dirac)−0.8 V−10 Vb)c)electron interactions29,30 into account by setting vF = vF (n, Te) with dependence on both charge density and electronic temperature as detailed in the supplementary material. Supercollision and flexural phonon scattering yield similar power law predictions10, but can be distinguished from each other by comparing numerical estimations against measured data. Although flexural phonon scattering may dominate over single acoustic phonon scat- tering (when Te > 10TBG, where TBG refers to acoustic phonons), the ratio of heat flux via flexural modes QF to that of supercollision heat flow QS is estimated as QF / QS = kF (cid:96)/200. We may fit the supercollision formula Eq. 1 to our data ∝ T 3 g > 5 V using the parameters vF = 1.0 · 106 m/s (at high energy) and D/(kF (cid:96))1/2 = 37 eV. Using the the theoretical estimate D = 20 − 30 eV6,7, we may conclude that only supercollision cooling is compatible with the results, as the heat flux is 100 times larger than what can be produced by flexural two-phonon scattering. e at Vg − V D Fig. 2c displays the charge carrier density dependence of the electron-phonon cooling power P at a fixed temperature. The calculated curves indicate the variation on n obtained from the low- and high-µ expansions of Eq. 1 using the same parameters as above. Compar- ison of the data at 150, 200, and 250 K indicate that the data scale with the ratio kBTe/µ in accordance with the theory. The evident bumps at low/intermediate particle density are assigned to resistance fluctuations that may cause bias-dependent modification of noise even at the band 600-900 MHz31. The linear slope in n corresponds to µ2 dependence away from the Dirac point where the velocity renormalization is negligible. The value for kF (cid:96) for supercollision scattering is small for our sample at high bias. Possible mechanisms contributing to this, in addition to short-range impurities, are dynamic and static ripples32, as well as scattering from the edges (possibly magnetic) of the sample. Coulomb impurities are not expected to contribute significantly to supercollisions. Ripples can yield quite small kF (cid:96) for scattering, as estimated in Ref.33; STM results of Ref.32 indicate kF (cid:96) (cid:39) 2 for a typical current-cleaned graphene sample such as ours at room temperature (kF (cid:96) = 1.8 using amplitude of corrugation Z = 0.4 nm and radius of curvature R = 1.5 nm, i.e. wave length ∼ 5 nm32). Both static and dynamic ripples are known to influence the resistance of suspended graphene samples15,16. Further information of the origin of the short kF (cid:96) in our experiment is found from the results of resistance measurements in the same sample. At low bias, the results indicate nearly ballistic behavior, which rules out scattering from static frozen-in ripples or short- 7 ranged impurities. Using our total square resistance R(cid:3) = (V /I)W/L vs. Te curve at n (cid:39) 0.8· 1011 cm−2 (see the supplementary information), we find the temperature-dependent e having a magnitude of ρi = 0.01(Te/K)2. component in the resistivity that increases as T 2 This value for ρi is quite close to the results of Ref.16 on flexural phonon limited mobility in suspended graphene. By interpreting ρi directly as a scattering length via kf (cid:96)r = h/(4e2ρi), we find a small value kF (cid:96)r ∼ 7 at Te = 300 K, indicating an effective scattering mechanism activated by the large bias; the subscript r refers to scattering events governing the resistance. −1) denotes the T 2 The effect of quasielastic scattering from thermally excited flexular phonons on the su- percollision energy flow is analyzed in Ref. 34. An effective kF (cid:96)eff for supercollisions can be estimated from the resistance, given an adjustment for differences in characteristic wave vector scales. Scattering leading to resistance occurs mainly on scales of max(kF , q∗), where √ us/κ denotes the cut-off of the flexural modes due to strain u in the membrane.16 q∗ = Supercollision energy transport is dominated by thermal phonons with qT = kBT /s. Here, s = 2.1 · 104 m/s is the speed of the acoustic mode in graphene, and κ = 4.6 · 10−7 m2/s is specified by the dispersion relation of the flexural modes ω = κq2. When kF < q∗ < qT −1 (cid:39) 10(q∗/qT )2(4e2/h)ρiG, (regime VI of Ref. 15), this leads to an effective value [kF (cid:96)E] e dependent part of the resistivity and G ∝ log(q∗/qT ) is where ρi (∝ [kF (cid:96)r] a factor of the order of unity; the subscript E denotes that these scattering events govern the energy exchange. We have estimated the strain u ≈ 4− 8· 10−4 from the gate voltage depen- dence of the measured fundamental flexural mode frequency fres = 77 MHz: f − fres < 0.1 MHz over Vg = 0 − 10 V. In our analysis we employ u = 4 · 10−4, which minimizes the u. Combining u = 4· 10−4 difference between our measured ρi and the calculated result ρi ∝ 1 with the value for ρi yields a temperature-independent result kF (cid:96)E (cid:39) 3 that can be employed in Eq. 1; this bias-induced value for kF (cid:96)E is much smaller than expected for short-range scatterers in high-quality graphene. However, smaller strain and q∗ and hence larger kF (cid:96)E would be necessary for the results to be in line with previous results on resistance due to flexural phonons, although these results are not fully comparable as they were measured at small bias voltage.16 Taking kF (cid:96) as a materials parameter independent of the variation in the charge carrier density, as is the case above, and using the above parameters in Eq. 1, we are able to fit our results also in the non-degenerate case ∝ T 5 e and in the cross-over regime as depicted in Fig. 2b. The fit indicates D = 64 eV which is by a factor of two larger than the theoretical 8 estimate D = 20 − 30 eV for the unscreened case6,7. Altogether, once the values of D = 64 eV and kF (cid:96) = 3 are fixed, the overall features of our data are accounted for by Eq. 1. Compared with previous transport experiments5,12, our results on the electron-phonon coupling are rather close in magnitude in the degenerate limit at high bias, even though our samples are ballistic at low bias. This suggests that the mechanism for the electron-phonon scattering may be similar, i.e. it could be ripples in both experiments. A strong contribution is expected to arise from excited flexural modes, since they can be in a "hot-phonon-like" non-equilibrium state as has been observed with optical phonons in carbon nanotubes under similar conditions35,36. Such non-equilibrium state of ripples, however, is more likely in a suspended sample than in a supported device. Finally, photocurrent in suspended graphene p-n junctions has been found to be by an or- der of magnitude larger than in supported structures; this enhancement has been attributed to the elimination of a dominant electronic cooling channel via the surface phonons of the polar SiO2 substrate37. Similar direct coupling to phonons has also been observed in single walled carbon nanotubes38. Our experiments avoid these problems and address truly the inherent electron-phonon coupling in graphene. In summary, our experiments indicate strong supercollision cooling in the presence of ripples in suspended graphene. We have achieved the first results on graphene cooling in the high temperature limit demonstrating the T 5 e dependence for the electron-phonon heat transfer. In the low-Te limit, our results indicate quadratic dependence on the chemical potential, which is a characteristic signature of non-conventional cooling processes. This µ2 behavior is in line with the cross-over, found at Te (cid:39) µ/kB, from the quintic high-Te behavior to cubic in the low-Te regime. Our analysis yields for the deformation potential D = 64 eV. Most likely, the observed strong electron-phonon coupling originates from supercollision events that are facilitated by a high-bias-induced ripple structure with a magnitude in line with recent room temperature STM experiments on suspended graphene. We acknowledge fruitful discussions with T. Heikkilä and M. Katsnelson. We have ben- efited from interaction with M.F. Craciun and S. Russo within a scientific exchange pro- gramme between Low Temperature Laboratory and Centre for Graphene Science at Exeter University. Our work was supported by the Academy of Finland (contracts no. 135908 and 250280, LTQ CoE), by the EU-project RODIN FP7-246026, and by the European Science Foundation (ESF) under the EUROCORES Programme EuroGRAPHENE. This research 9 project made use of the Aalto University Cryohall infrastructure. MO is grateful to Väisälä Foundation of the Finnish Academy of Science and Letters for a scholarship. ∗ [email protected] 1 Kubakaddi, S. S. Phys. Rev. B 2009, 79, 75417. 2 Tse, W.-K.; Das Sarma, S. Phys. Rev. B 2009, 79, 235406. 3 Bistritzer, R.; MacDonald, A. Phys. Rev. Lett. 2009, 102, 206410. 4 Viljas, J. K.; Heikkila, T. T. Phys. Rev. B 2010, 81, 245404. 5 Betz, A. C.; Jhang, S. H.; Pallecchi, E.; Ferreira, R.; Fève, G.; Berroir, J.-M.; Plaçais, B. Nat. Phys. 2012, 9, 109 -- 112. 6 Suzuura, H.; Ando, T. Phys. Rev. B 2002, 65, 235412. 7 Katsnelson, M. I. Graphene: Carbon in Two Dimensions, 1st ed.; Cambridge University Press, 2012. 8 Chen, J.-H.; Jang, C.; Xiao, S.; Ishigami, M.; Fuhrer, M. S. Nat. Nanotechnol. 2008, 3, 206 -- 9. 9 Efetov, D. K.; Kim, P. Phys. Rev. Lett. 2010, 105, 256805. 10 Song, J. C. W.; Reizer, M. Y.; Levitov, L. S. Phys. Rev. Lett. 2012, 109, 106602. 11 Graham, M. W.; Shi, S.-F.; Ralph, D. C.; Park, J.; McEuen, P. L. Nat. Phys. 2012, 9, 103 -- 108. 12 Betz, A. C. Phys. Rev. Lett. 2012, 109, 56805. 13 Graham, M. W.; Shi, S.-F.; Wang, Z.; Ralph, D. C.; Park, J.; McEuen, P. L. Nano Lett. 2013, 13, 5497 -- 502. 14 Chen, W.; Clerk, A. a. Phys. Rev. B 2012, 86, 125443. 15 Mariani, E.; von Oppen, F. Phys. Rev. B 2010, 82, 195403. 16 Castro, E.; Ochoa, H.; Katsnelson, M.; Gorbachev, R.; Elias, D.; Novoselov, K.; Geim, a.; Guinea, F. Phys. Rev. Lett. 2010, 105, 16 -- 18. 17 Giazotto, F.; Heikkilä, T. T.; Luukanen, A.; Savin, A. M.; Pekola, J. P. Rev. Mod. Phys. 2006, 78, 217 -- 274. 18 Oksanen, M.; Uppstu, A.; Laitinen, A.; Cox, D. J.; Craciun, M.; Russo, S.; Harju, A.; Hako- nen, P. arXiv.org:1306.1212 2013. 19 Bolotin, K. I.; Sikes, K. J.; Jiang, Z.; Klima, M.; Fudenberg, G.; Hone, J.; Kim, P.; Stormer, H. L. Solid State Commun. 2008, 146, 351 -- 355. 10 20 Katsnelson, M. I.; Novoselov, K. S.; Geim, A. K. Nat. Phys. 2006, 2, 620 -- 625. 21 Tworzydlo, J.; Trauzettel, B.; Titov, M.; Rycerz, A.; Beenakker, C. W. J.; J., B. C. W. Phys. Rev. Lett. 2006, 96, 246802. 22 Danneau, R.; Wu, F.; Craciun, M. F.; Russo, S.; Tomi, M. Y.; Salmilehto, J.; Morpurgo, A. F.; Hakonen, P. J. J. Low Temp. Phys. 2008, 153, 374 -- 392. 23 Blanter, Y. M.; Büttiker, M. Phys. Rep. 2000, 336, 1. 24 Wu, F.; Virtanen, P.; Andresen, S.; Placais, B.; Hakonen, P. J. Appl. Phys. Lett. 2010, 97, 262115. 25 Santavicca, D. F.; Chudow, J. D.; Prober, D. E.; Purewal, M. S.; Kim, P. Nano Lett. 2010, 10, 4538 -- 43. 26 Balandin, A. A. Nat. Mater. 2011, 10, 569 -- 581. 27 Compare with Tarkiainen, R.; Ahlskog, M.; Hakonen, P.; Paalanen, M. J. Phys. Soc. JAPAN 2003, 72, 100. 28 Fong, K. C.; Wollman, E. E.; Ravi, H.; Chen, W.; Clerk, A. A.; Shaw, M. D.; Leduc, H. G.; Schwab, K. C. Phys. Rev. X 2013, 3, 041008. 29 Kotov, V. N.; Uchoa, B.; Pereira, V. M.; Guinea, F.; Castro Neto, A. H. Rev. Mod. Phys. 2012, 84, 1067 -- 1125. 30 Elias, D. C.; Gorbachev, R. V.; Mayorov, A. S.; Morozov, S. V.; Zhukov, A. A.; Blake, P.; Ponomarenko, L. A.; Grigorieva, I. V.; Novoselov, K. S.; Guinea, F.; Geim, A. K. Nat. Phys. 2011, 7, 701 -- 704. 31 Wu, F.; Tsuneta, T.; Tarkiainen, R.; Gunnarsson, D.; Wang, T.-H.; Hakonen, P. J. Phys. Rev. B 2007, 75, 125419. 32 Zan, R.; Muryn, C.; Bangert, U.; Mattocks, P.; Wincott, P.; Vaughan, D.; Li, X.; Colombo, L.; Ruoff, R. S.; Hamilton, B.; Novoselov, K. S. Nanoscale 2012, 4, 3065 -- 8. 33 Song, J. C. W.; Reizer, M. Y.; Levitov, L. S. arXiv.org:1111.4678v1 2011. 34 Virtanen, P. arXiv.org:1312.3833 2013. 35 Yao, Z.; Kane, C. L.; Dekker, C. Phys. Rev. Lett. 2000, 84, 2941 -- 2944. 36 Lazzeri, M.; Piscanec, S.; Mauri, F.; Ferrari, A. C.; Robertson, J. Phys. Rev. Lett. 2005, 95, 236802. 37 Freitag, M.; Low, T.; Avouris, P. Nano Lett. 2013, 13, 1644 -- 1648. 38 Baloch, K. H.; Voskanian, N.; Bronsgeest, M.; Cumings, J. Nat. Nanotechnol. 2012, 7, 316 -- 319. 11 SUPPLEMENTARY INFORMATION FOR "ELECTRON-PHONON COUPLING IN SUSPENDED GRAPHENE: SUPERCOL- LISIONS BY RIPPLES" I. RENORMALIZATION OF THE FERMI VELOCITY (cid:16) vF ((cid:126)k) ∼ vF,∗ 1 + ln α∗ 4 (cid:17) In suspended graphene samples close to the Dirac point, the electronic spectrum and the density of states is renormalized by electron-electron interactions, as discussed in Ref. 29. Effectively, the Fermi velocity increases at low energies, which then is also reflected in the supercollision mechanism. Here, we account for the interaction effects via a simple renormalization group (RG) procedure29. As electron states (cid:126)k > Λ are integrated out, the effective electron-electron interaction α = e2/(4π
0vF ) flows towards zero. The correspond- ing velocity renormalization in the simplest approximation reads Λ∗ √ πn, kBTe/vF,∗) , max((cid:126)k, (S1) where the renormalized velocity at scale Λ∗ is set as vF,∗ = 1.0· 106 m/s according to regular tight binding parameters7, α∗ = α = 1 with
G = 2.2,30 and the scale Λ∗ is calculated at the
G charge density 5× 1012 cm−2 yielding Λ∗ = kBvF,∗ × 2200 K. The Fermi velocity is momentum- dependent and renormalized at µ = 0 to vF ((cid:126)k) at (cid:126)k < Λ∗. Temperature and chemical potential both cut off the flow of vF , although the cross-over region when kF ∼ kBTe/vF,∗ is not completely accurately handled by this equation. Electron-phonon supercollisions involve, in addition to initial and final low-energy states whose energies are renormalized as described by Eq. (S1), a high-energy virtual electron state at (cid:126)k ∼ (cid:126)q ∼ kBTe/(s) (cid:29) kF ; here s stands for the speed of sound. For typical temperatures in our case, this is close to or above the cutoff Λ∗, where the variation of vF is small, and we take vF ≈ vF,∗ in this regime. As the velocity renormalization results to a linear √ spectrum below the Fermi level, the µ vs. n relationship remains simple, µ = vF (n, Te) πn. With these two provisions, the effect of velocity renormalization on supercollisions is obtained by substituting Eq. (S1) for k = kF into Eq. (1) of the main text. The main effect of the electron-electron interactions is to shift the cross-over point between degenerate and non-degenerate regimes towards smaller charge densities as compared to the noninteracting situation. For n ∼ 1010 cm−2, we have vF /vF,∗ ∼ 1.3 . . . 1.6 between T = 10 . . . 500 K. 12 II. DETAILS OF THE ELECTRICAL CHARACTERISTICS In our experiments, we measured the electrical characteristics of the sample both at DC and at low-frequency AC (dynamic resistance). The strength of the flexural phonon scattering was determined from the measured total square resistance R(cid:3) = (V /I)W/L il- lustrated in Fig. 3 at 0.8 · 1011 cm2 (Vg = -2.4 V) as a function of Te, with Te determined from the shot noise thermometry. The behavior of R(cid:3) is well fit with a quadratic tem- perature dependence as expected for flexural phonon scattering16. From the fit we obtain R(cid:3) = [1390 + 0.01(Te/K)2] Ω which was employed for determining the pure temperature- dependent part ρi = 0.01(Te/K)2 employed in the main text. Figure 3. Total square resistance R(cid:3) = (V /I)W/L as a function of the electronic temperature deduced from the Fano factor using Te = F eV /2kB (n = 0.8 · 1011 cm−2). The fit function is R(cid:3) = [1390 + 0.01(Te/K)2] Ω. We also measured the zero-bias AC resistance R0 as a function of temperature which turned out to be irregular because residual gas got desorbed from surfaces in the vacuum chamber while the cryostat was warming up. The desorbed gas became partly readsorbed on to the clean, current-annealed sample, causing a shift of the Dirac point due to charge doping, most likely due to oxygen. Nevertheless, the data could be fit pretty well with 13 05010015020025030014001600180020002200Te (K)R (Ω)e -term close to the above ρi. However, we quadratic temperature dependence, yielding a T 2 consider the determination based on V /I more reliable than our R0 analysis, because it probes the scattering under the same conditions as those prevailing in the actual electron- phonon coupling measurement. The IV-curve measured at the same charge density n = 0.8 · 1011 cm−2 is illustrated in Fig. 4. The IV-curve reflects the clear increase of R(cid:3) along with the bias voltage V . The differential resistance Rd = dV /dI, measured by lock-in methods, corresponds to the inverse slope of the IV-curve. The main purpose of Rd(V ) measurements was to determine the coupling strength of the noise from the sample to the preamplifier. Figure 4. IV-curve at Vg = -2.4 V measured over a voltage range in which the electronic temperature Te, determined from the shot noise thermometry, traverses the same T−range as displayed in Fig. 3. III. ANALYSIS OF THE POWER LAWS The heat flow from electrons to the phonon system is characterized by the power law P ∝ T δ ph). We observe power laws δ = 5 and 3, near and far from the Dirac point, respectively. This transition was e (where we have dropped the small phonon temperature term T δ 14 −0.2−0.100.10.2−0.5−0.2500.250.5V (V)I (mA)Figure 5. Measured heat flow from electrons to phonons (data from Fig. 2c in the main text) e and displayed as a function of Te for three gate voltage values indicated in the normalized by T 5 figure. illustrated in the main text using direct plotting on log-log frame. Here we replot the data e . The curve closest to the Dirac point in Fig. 5 by normalizing the power flow P with T 5 is almost completely flat, which indicates T 5 e -dependence. Slightly off from the Dirac point (Vg = -0.8 V) we observe a weakly declining curve, which belongs to the cross-over region from δ = 5 towards δ = 3. Far away from the Dirac point (Vg = -10 V), we obtain steeper behavior, close to 1/T 2 e , which corresponds to δ = 3. 15 20025030035040045050000.20.40.60.81x 10−9Te (K)P/Te5 (W/(K5cm2))+0.4 V (Dirac)−0.8 V−10 V

1102.0039

1

1102

Spontaneous Emergence of Persistent Spin Helix from Homogeneous Spin Polarization

We demonstrate that a homogeneous spin polarization in one-dimensional structures of finite length in the presence of Bychkov-Rashba spin-orbit coupling decays spontaneously toward a persistent spin helix. The analysis of formation of spin helical state is presented within a novel approach based on a mapping of spin drift-diffusion equations into a heat equation for a complex field. Such a strikingly different and simple method allows generating robust spin structures whose properties can be tuned by the strength of the spin orbit interaction and/or structure's length. We generalize our results for two-dimensional case predicting formation of persistent spin helix in two-dimensional channels from homogeneous spin polarization.

cond-mat.mes-hall

cond-mat

Spontaneous Emergence of Persistent Spin Helix from Homogeneous Spin Polarization Valeriy A. Slipko,1, 2 Ibrahim Savran,3 and Yuriy V. Pershin1, ∗ 2 Department of Physics and Technology, V. N. Karazin Kharkov National University, Kharkov 61077, Ukraine 1Department of Physics and Astronomy and USC Nanocenter, University of South Carolina, Columbia, SC 29208, USA 3Department of Computer Science and Engineering, University of South Carolina, Columbia, SC 29208, USA 1 1 0 2 n a J 1 3 ] l l a h - s e m . t a m - d n o c [ 1 v 9 3 0 0 . 2 0 1 1 : v i X r a We demonstrate that a homogeneous spin polarization in one-dimensional structures of finite length in the presence of Bychkov-Rashba spin-orbit coupling decays spontaneously toward a persis- tent spin helix. The analysis of formation of spin helical state is presented within a novel approach based on a mapping of spin drift-diffusion equations into a heat equation for a complex field. Such a strikingly different and simple method allows generating robust spin structures whose properties can be tuned by the strength of the spin orbit interaction and/or structure's length. We generalize our results for two-dimensional case predicting formation of persistent spin helix in two-dimensional channels from homogeneous spin polarization. PACS numbers: 72.15.Lh, 72.25.Dc, 85.75.2d The helical wave of rotating spin orientation is re- ferred to as the spin helix. There is a significant interest to spin helix configurations in semiconductor materials since the electron spin relaxation of such spin configura- tions can be partially [1 -- 3] or even completely suppressed [4, 5]. While a partial suppression of spin relaxation in two-dimensional systems becomes possible in the pres- ence of only Bychkov-Rashba [6] spin-orbit coupling (see Refs. [1, 2]), the complete suppression of spin relaxation requires a specific combination of Bychkov-Rashba and Dresselhaus [7] interactions as it was demonstrated in Ref. [4]. More generally, the relaxation of the spin he- lix is an example of situations [1 -- 5, 8 -- 17] when electron spin relaxation scenario deviates from the predictions of D'yakonov-Perel' theory [18]. Experimentally, the spin grating technique [19] is typ- ically used [3, 5] to create spin helical configurations in semiconductors. In this method, a sample is illuminated by a pair of pump beams with orthogonal linear polariza- tions. The interference of such beams results in a spacial modulation of light helicity. Correspondingly, through the optical orientation effect, a modulation of spin polar- ization in the form of spin helix is produced. Moreover, a spin injection from a ferromagnetic material into a semi- conductor can also be used to excite a spin helix [20]. In this approach, the rotating spin polarization is caused by coherent spin precession of electrons drifting in an applied electric field. However, the present authors are not aware about any experimental studies of spin helixes excited by spin injection. In this Letter, we propose an alternative approach to induce spin helical configurations. Specifically, we demonstrate that in one-dimensional (1D) systems of fi- nite length with Bychkov-Rashba spin-orbit coupling the spin helical configurations emerge in the process of re- laxation of homogeneous spin polarization (see Fig. 1). Mathematically, such a strikingly unexpected transfor- mation of homogeneous spin polarization into the per- sistent spin helix occurs when we introduce boundary conditions on electron space motion to describe finite- length structures (in infinite systems the homogeneous spin polarization decays exponentially as predicted by D'yakonov-Perel' theory [18]). Using a novel approach that maps spin drift-diffusion equations into a heat trans- fer equation for a complex field we find the exact time- dependence of the spin polarization dynamics. It is inter- esting that the amplitude of the resulting spin helix has an oscillatory dependence on the system's length. Below, we provide an intuitive explanation of this result based on properties of solution of heat equation. Moreover, it is necessary to emphasize that our theory is generalized for the case of two-dimensional (2D) channels and can be straightforwardly verified experimentally. In partic- ular, experimentally, the homogeneous spin polarization can be easily created using the optical orientation by cir- cularly polarized light. Therefore, we believe that our approach would simplify tremendously the generation of long-living spin helical configurations in semiconductor structures and advance the field of spin storage in semi- conductors. FIG. 1: (Color online) Schematics of spontaneous transfor- mation of homogeneous spin polarization into persistent spin helix in a finite length system with Bychkov-Rashba spin-orbit coupling. 0=t∞→t2 FIG. 2: (Color online) Dynamics of formation of persistent spin helix from homogeneous spin polarization pointing in z direction at t = 0. These plots were obtained using Eq. (13) at ηL = 15.45. This value of the parameter ηL corresponds to the second local maximum of spin helix amplitude shown in Fig. 3. Let us consider dynamics of electron spin polariza- tion in a 1D system of a length L in x direction in the presence of Bychkov-Rashba spin-orbit coupling. In one- dimensional limit, spin drift-diffusion equations [2] can be written as ∂Sx ∂t ∂Sy ∂t ∂Sz ∂t = D∆Sx + C = D∆Sy, = D∆Sz − C ∂Sz ∂x − 2γSx, ∂Sx ∂x − 2γSz, (1) (2) (3) where D = (cid:96)2/τ is the coefficient of diffusion, ∆ = ∂2/∂x2, C = 2ηD is the constant describing spin rota- tions, γ = η2D/2 is the coefficient describing spin relax- ation, η = 2αm−1 is the spin precession angle per unit length, α is the spin-orbit coupling constant, m is the ef- fective electron mass, (cid:96) is the mean free path and τ is the momentum relaxation time. It follows from Eq. (2) that y component of spin polarization, Sy, is not coupled to any other component of spin polarization. Consequently, selecting Sy(x, t = 0) = 0 we can safely take out Sy from our consideration. Eqs. (1,3) are complimented by stan- dard boundary conditions [21] (cid:19) (cid:18) (cid:19) + CSz = 0, 2D Γ ∂Sz ∂x − CSx = 0. (4) Γ Here, Γ = [x = 0, x = L]. Mathematically, the bound- ary conditions (4) are so-called third-type boundary con- ditions. This specific form of boundary conditions con- serves the spin polarization of electrons that scatter from the sample edges. We assume that at the initial moment (cid:18) 2D ∂Sx ∂x of time the spin polarization is homogeneous and points in z direction, that is Sx(x, t = 0) = 0, Sz(x, t = 0) = S0. (5) Let us introduce a complex polarization S = Sx + iSz. It is straightforward to show that Eqs. (1,3) and bound- ary conditions (4) can be rewritten in a more compact form using S: ∂S ∂t = D (cid:18) ∂2S ∂x2 − iC ∂S ∂x − iCS 2D ∂S ∂x − 2γS, (cid:19) = 0. Γ Defining a complex field u(x, t) by the relation u(x, t) = e−iηxS(x, t), (6) (7) (8) we find that Eq. (6) transforms into the heat equation ∂u ∂t = D ∂2u ∂x2 , (9) supplemented by Neumann (or second-type) boundary conditions = 0. (10) (cid:18) ∂u (cid:19) ∂x Γ Moreover, it is worth noticing that the initial conditions for u(x, t) are related to the initial conditions for S as u(x, t = 0) = e−iηxS(x, t = 0). (11) Consequently, the initially homogeneous spin polariza- tion in z direction (Eq. (5)) corresponds to a spatially modulated complex field u(x, t = 0) = S0 sin (ηx) + iS0 cos (ηx) . (12) 0.020.040.060.080.100.20.40.60.81.0Sz/S0Time (in units of L2/D)x (in units of L)-0.250000.25000.50000.75001.000(a)0.020.040.060.080.100.00.20.40.60.81.0Sx/S0(b) Time (in units of L2/D)x (in units of L)-0.250000.25000.50000.75001.000The solution of Eq. (9) with the boundary condi- tions (10) and initial condition (12) was obtained by the method of separation of variables. It can be presented in the form 3 S(x, t) sin(ηL/2) = i S0 ηL/2 1 − (−1)ne−iηL (ηL)2 − (πn)2 e +∞(cid:88) n=1 eiη(x−L/2) + − π2 n2Dt L2 cos (cid:16) πnx (cid:17) . (13) L 2ηLeiηx This is our main analytical result describing dynamics of spin polarization in 1D finite-length structures. Note that Sx and Sz components of spin polarization are given (13), respectively. by real and imaginary parts of Eq. The first term in the right-hand side of Eq. (13) de- scribes the persistent profile of spin polarization (in the form of spin helix) emerging at long times. Concerning the second term in the right-hand side of Eq. (13), it governs the dynamics of transformation of the initially homogeneous spin polarization into the persistent spin helix. Fig. 2 demonstrates dynamics of Sz and Sx com- ponents of spin polarization given by Eq. It is clearly seen that the initially homogeneous spin polariza- tion in z directions transforms into the persistent spin helix with an (infinitely) long lifetime. (13). Explicitly, in the long time limit, the spin polarization is given by Sx(x, t = +∞) = −S0 Sz(x, t = +∞) = S0 sin(ηL/2) ηL/2 sin(ηL/2) ηL/2 sin(η(x − L/2)), (14) cos(η(x − L/2)). (15) In these equations the factor sin(ηL/2)/(ηL/2) defines reduction of the spin helix amplitude with the respect to the initial amplitude of homogeneous spin polarization S0. We plot this function in Fig. 3. It is interesting that the spin helix amplitude is an oscillating function of the parameter ηL and takes zero values when ηL = 2πn where n is a positive integer. The positions of local max- ima can be found numerically. In particular, positions of four local maxima shown in Fig. 3 are 8.987, 15.450, 21.808, 28.132. The heat equation is the best starting point to under- stand the oscillatory dependence of spin helix amplitude on ηL depicted in Fig. 3. Accordingly to Eq. (11), the initially homogeneous initial condition (Eq. (5) for spin diffusion equations transforms into a modulated initial condition for the heat equation. As the solution of heat equation in the given context represents simply the pro- cess of temperature equilibration along the system, an integer number of modulation periods results in zero av- erage "temperature" and, correspondingly in zero spin helix amplitude. Moreover, we would like to mention that the spin helix formation process is described by a series of exponentially decaying terms whose time con- stants are given by τn = L2/(π2n2D). The longest of FIG. 3: (Color online) Normalized amplitude of the persistent spin helix as a function of ηL. Insets show schematically distributions of Sx and Sz at several specific values of ηL as indicated by arrows. Positions of minima and maxima points of the amplitude are discussed in the text. these times τ1 = L2/(π2D) provides the time scale of the transformation process. The dependence of spin pro- cess on L is intuitively clear as electrons should "feel" the system's length before the transformation ends. It's also interesting that such a time can be longer or shorter then the relaxation time of homogeneous spin polariza- tion τh = 1/(Dη2). In particular, τ1/τh = (ηL/π)2 mean- ing that τ1 < τh when ηL < π, the times are the same when ηL = π, and τ1 > τh when ηL > π (see also Fig. 3). In order to obtain an additional insight on spin re- laxation of the radial spin helix, we have performed ex- tensive Monte Carlo simulations employing an approach described in Refs. [8] and [22]. This Monte Carlo sim- ulation method uses a semiclassical description of elec- tron space motion and quantum-mechanical description of spin dynamics (the later is based on the Bychkov- Rashba coupling term). All specific details of the Monte Carlo simulations program can be found in the references cited above and will not be repeated here. A spin con- servation condition was used for electrons scattering from system boundaries. Generally, all obtained Monte Carlo simulation results are in perfect quantitative agreement with our analytical predictions thus confirming the mech- anism of formation of persistent spin helix from homoge- neous spin polarization. A comparison of selected ana- lytical and numerical curves is given in Fig. 4. The results reported in this paper can be readily gen- eralized for the persistent spin helix in two dimensions [4]. Indeed, it can be easily seen that in the case of equal strength of Bychkov-Rashba and Dresselhaus spin- orbit interactions, α = β (where β is the Dresselhaus 0510152025300.00.20.40.60.81.0xxx Spin Helix Amplitudeη LxSxSz 4 trol of spin helix characteristics is achievable via appro- priate choice of the above mentioned parameters. This suggested technique facilitates generation of spin helical states and can be used in both one- and two-dimensional geometries. I. S. acknowledges PhD scholarship from the Repub- lic of Turkey Ministry of National Education, Grant No: MEB1416. ∗ Electronic address: [email protected] 1 Y. V. Pershin, Phys. Rev. B 71, 155317 (2005). 2 Y. V. Pershin and V. A. Slipko, Phys. Rev. B 82, 125325 (2010). 3 C. P. Weber, J. Orenstein, B. A. Bernevig, S.-C. Zhang, J. Stephens, and D. D. Awschalom, Phys. Rev. Lett. 98, 076604 (2007). 4 B. A. Bernevig, J. Orenstein, and S.-C. Zhang, Phys. Rev. Lett. 97, 236601 (2006). 5 J. D. Koralek, C. P. Weber, J. Orenstein, B. A. Bernevig, S.-C. Zhang, S. Mack, and D. D. Awschalom, Nature 458, 610 (2009). 6 Y. Bychkov and E. Rashba, JETP Lett. 39, 78 (1984). 7 G. Dresselhaus, Phys. Rev. 100, 580 (1955). 8 A. A. Kiselev and K. W. Kim, Phys. Rev. B 61, 13115 (2000). 125310 (2004). (2004). 98, 113702 (2005). 9 E. Y. Sherman, Appl. Phys. lett 82, 209 (2003). 10 M. Q. Weng, M. W. Wu, and Q. W. Shi, Phys. Rev. B 69, 11 Y. V. Pershin and V. Privman, Phys. Rev. B 69, 073310 12 L. Jiang, M. Weng, M. Wu, and J. Cheng, J. Appl. Phys. 13 P. Schwab, M. Dzierzawa, C. Gorini, and R. Raimondi, Phys. Rev. B 74, 155316 (2006). 14 M. Q. Weng, M. W. Wu, and H. L. Cui, J. Appl. Phys. 15 P. Kleinert and V. V. Bryksin, Phys. Rev. B 79, 045317 103, 063714 (2008). (2009). 80, 235327 (2009). (2010). 3023 (1972). 76, 4793 (1996). 90, 146801 (2003). 16 M. Duckheim, D. L. Maslov, and D. Loss, Phys. Rev. B 17 I. V. Tokatly and E. Y. Sherman, Ann. Phys. 325, 1104 18 M. I. Dyakonov and V. I. Perel', Sov. Phys. Solid State 13, 19 A. R. Cameron, P. Riblet, and A. Miller, Phys. Rev. Lett. 20 J. Schliemann, J. C. Egues, and D. Loss, Phys. Rev. Lett. 21 V. M. Galitski, A. A. Burkov, and S. Das Sarma, Phys. 22 S. Saikin, Y. Pershin, and V. Privman, IEE-Proc. Circ. Rev. B 74, 115331 (2006). Dev. Syst. 152, 366 (2005). FIG. 4: (Color online) Long-time distribution of Sz at ηL = 8.987 found employing Monte Carlo simulation approach. The analytical curve is obtained using Eq. (15). The Monte Carlo simulation was performed for 105 electrons in GaAs structure of 1.7µm length. This plot obtained using the pa- rameter values τ = 0.1ps, l = 10nm, α = 3·10−12eV m. Inset: orientation of 2D channel for two-dimensional spin helix ex- citation experiments. spin-orbit coupling constant), the equations of spin dif- fusion in 2D [4] take the general form of Eqs. (1-3). Therefore, introducing appropriate boundary conditions, namely, reducing the system into a 2D channel in [-110] direction (see the inset in Fig. 4), we obtain the situa- tion completely equivalent to that in 1D from the point of view of spin dynamics. Taking into account recent ex- perimental demonstration of persistent spin helix [5] the emergence of persistent spin helix from homogeneous spin polarization can be straightforwardly detected. Finally, we would like to note that the amplitude of persistent spin helix can be increased by a repetitive excitation of homogeneous polarization by a train of laser pulses. In summary, we have demonstrated that persistent spin helix forms in the process of relaxation of homo- geneous spin polarization in finite length systems. This observation can be used as a different technique for creat- ing spin helical structures in semiconductors. The solu- tion of spin drift-diffusion equations describing formation of persistent helix was derived analytically and numeri- cally using Monte Carlo simulation approach. The re- sults obtained in both ways are in perfect agreement. It is interesting that the persistent helix amplitude demon- strates an oscillatory dependence on the system length and strength of spin orbit interaction. Therefore, the con- 0.00.20.40.60.81.0-0.3-0.2-0.10.00.10.20.3[010] Sz/S0 Monte Carlo result Analytical resultx/L[100]

1509.03373

1

1509

Effective spin Hall properties of a mixture of materials with and without spin-orbit coupling: Tailoring the effective spin-diffusion length

We study theoretically the effective spin Hall properties of a composite consisting of two materials with and without spin-orbit (SO) coupling. In particular, we assume that SO material represents a system of grains in a matrix with no SO. We calculate the effective spin Hall angle and the effective spin diffusion length of the mixture. Our main qualitative finding is that, when the bare spin diffusion length is much smaller than the radius of the grain, the effective spin diffusion length is strongly enhanced, well beyond the "geometrical" factor. The physical origin of this additional enhancement is that, with small diffusion length, the spin current mostly flows around the grain without suffering much loss. We also demonstrate that the voltage, created by a spin current, is sensitive to a very weak magnetic field directed along the spin current, and even reverses sign in a certain domain of fields. The origin of this sensitivity is that the spin precession, caused by magnetic field, takes place outside the grains where SO is absent.

cond-mat.mes-hall

cond-mat

a Effective spin Hall properties of a mixture of materials with and without spin-orbit coupling: Tailoring the effective spin-diffusion length Z. Yue1, M. C. Prestgard2, A. Tiwari2, and M. E. Raikh1 1Department of Physics and Astronomy, University of Utah, Salt Lake City, UT 84112, USA 2Department of Materials Science and Engineering, University of Utah, Salt Lake City, Utah 84112, USA We study theoretically the effective spin Hall properties of a composite consisting of two materials with and without spin-orbit (SO) coupling. In particular, we assume that SO material represents a system of grains in a matrix with no SO. We calculate the effective spin Hall angle and the effective spin diffusion length of the mixture. Our main qualitative finding is that, when the bare spin diffusion length is much smaller than the radius of the grain, the effective spin diffusion length is strongly enhanced, well beyond the "geometrical" factor. The physical origin of this additional enhancement is that, with small diffusion length, the spin current mostly flows around the grain without suffering much loss. We also demonstrate that the voltage, created by a spin current, is sensitive to a very weak magnetic field directed along the spin current, and even reverses sign in a certain domain of fields. The origin of this sensitivity is that the spin precession, caused by magnetic field, takes place outside the grains where SO is absent. PACS numbers: 85.75.-d,72.25.Rb, 78.47.-p I. INTRODUCTION The spin Hall effect1 -- 3(SHE), predicted theoretically more than four decades ago1,2, is nowadays routinely observed in many materials,4 -- 17 which include tradi- tional and exotic metals, prominent semiconductors, and graphene. Moreover, the inverse spin Hall effect (ISHE), i.e. generation of voltage drop normal to the spin current, was recently "put to work". It serves as a tool to detect whether or not the spin current is injected into a non- magnetic material from an ac-driven ferromagnet in the course of spin pumping. Most recently18 -- 21 the pumped spin currents in certain polymers were registered via in- verse spin Hall voltage which they induced in Pt elec- trode located at some distance from the interface with ferromagnet. The latest focus22 -- 24 of the research on the spin physics in organics is the study of the properties of platinum- containing π-conjugated polymers. In these materials Pt atoms are embedded in the polymer backbone chains. While the SO coupling, which is the origin of the SHE, is very weak in polymers, adding of Pt creates the elements of the backbone where it is locally strong. These elements can be separated either by one or by three π-conjugated spacer unit lengths. In this regard, a general question arises: how the spin Hall effect is realized in composite materials where the strong SO and low SO domains are intermixed? Note that, by now, all theoretical studies of SO-related transport assumed that the SO coupling is homogeneous. The goal of the present paper is to develop an ele- mentary theory which addresses the question formulated above. Unlike Refs. 25 and 26, we will not specify a mechanism of SO on the microscopic level, but rather focus on purely "geometrical" aspects. Namely, we will consider the following minimal model: a system of SO grains is dissolved in a matrix with no SO. The question FIG. 1. (Color online) (a) Conventional geometry for the inverse spin Hall effect. Spin current flowing along y causes a buildup of the voltage, V ISH, between the edges z = ±L/2. The buildup takes place as long as y is smaller than the spin diffusion length, λ. (b) Schematic illustration of a "granular" geometry, where the SO-coupled material is dissolved in the matrix with no SO coupling. (c) microscopic scenario of ISHE on a single spherical granule of a radius, a. Spin current with polarization along x turns the sphere into an electrical dipole directed normally to the current. The magnitude of a dipole moment, Pc, depends on the ratio between a and λ, while the electric field inside the granule is homogeneous. we will be interested in is: what are the effective spin Hall characteristics of the mixture. Firstly, we address a mechanism of the formation of the inverse spin Hall voltage between the edges of the sample in the geometry of the mixture. Unlike the case of homogeneous SO, this formation happens as follows. The spin current turns each SO grain into an electric dipole. All dipole moments are oriented normal to the spin current. Thus the potentials they create at the up- per and the lower boundaries of the sample add up. The y✓azPcisisVISHVISHe↵⌧outs=1is(y)isdLzx(a)(b)(c)2 difference of these potentials is the effective ISH voltage, V SH eff , of the mixture, which can be related to the effective spin Hall angle, θSH eff . Naively, one would expect that, in a mixture of grains eff = (na3)θSH of density, n, and radius, a, the relation θSH holds within a numerical factor. Here θSH is the spin Hall angle of the bulk SO material. This is simply because na3 is the volume fraction of the SO material. Equally, one would expect that the effective spin relaxation time of the mixture is 1/(na3) times longer than in the SO material, so that spin diffusion length, λeff is related to the spin diffusion length, λ, of the SO material as λeff = (na3)−1/2λ. The above expectations are correct only in the limit when the grains are small enough, namely, a (cid:28) λ, so that the portion of spin polarization, which is lost within a single grain, is small. The opposite case of large grains, a (cid:29) λ, is much less trivial. As we show below, in this limit V SH In other words, at small λ, the effective spin-diffusion length sat- urates. This finding can be loosely interpreted from the perspective of diffusion in the presence of the absorbing traps. The stronger is the absorption, the smaller is the concentration of particles at the position of the trap. eff ∼ λna2V SH, while λeff ∼ (na)1/2 . 1 eff s Finally, we will demonstrate that V SH is sensitive to a very weak magnetic field. In a homogeneous material, the spin Hall effect gets suppressed in the field with Larmour frequency Ω ∼ τ−1 , where τs is the spin-relaxation time. For the mixture, the characteristic field is ∼ T −1, where T is the diffusion time between the sample edges. This is because spin precession takes place mostly outside the grains. The paper is organized as follows. II we solve an auxiliary problem of electric the polarization of a grain with a given radius, a, by the spin current. The solution is then employed to calculate the effective inverse spin Hall voltage in the mixture of grains with concentration, n. Sensitivity of this voltage to a weak longitudinal magnetic field is studied in Sect. III. In Sect. IV the effective diffusion length, λeff, of the mixture is expressed via λ, a, and the parameter na3. The physics of elongation of λeff for small λ (cid:28) a is discussed in Sect. V. Concluding remarks are presented in Sect. VI. In Sect. II. CALCULATION OF EFFECTIVE CHARACTERISTICS OF THE MIXTURE A. Single grain The simplest way to incorporate the spin Hall effect on a quantitative level27 is to add to the current density, j = σE, the term γD curl P where σ and D are the conductivity and the diffusion coefficient, respectively, P (r) is the coordinate-dependent spin polarization. The strength of the SO coupling is quantified by a dimension- less parameter γ. The system of coupled equations for FIG. 2. (Color online) (a) The cross section z = 0. Distri- bution of the spin current in the (x, y)-plane in the presence of a spherical grain, Eq. (38), is illustrated schematically for r > a. Inside the grain, r < a, this distribution is determined by Eq. (29). (b) Distribution of the spin polarization along the radius, r, is plotted for Dout/Din = 2 and three values of λ: a/λ = 0.2 (green), a/λ = 4 (blue), and a/λ = 12 (purple). Enhancement of the effective spin diffusion length for small λ/a is a result of a strong suppression of polarization near the boundary r = a. the spatial distribution of P (r) and j(r) reads27 j = σE + eγD curl P . qij = −D ∂Pj ∂xi + γ e σεijkEk. (1) (2) The second equation defines the component i of the flux of the j-projection of spin polarization. The system be- comes closed27 when it is complemented by the continuity equation ∂qij ∂xi + Pj τs = 0. (3) Consider an isolated spherical grain with radius, a, and with the strength of SO-coupling, γ, embedded into an infinite medium with γ = 0 and with no spin relaxation, τs = ∞, Fig. 1. Assume that the flux of spins, oriented along the x-axis and flowing along the y-axis, is incident on the grain. In application to the geometry, Fig. 1, the essence of the inverse spin Hall effect is that the incident spin current, is, induces an effective electric dipole on the sphere. The induced dipole moment is perpendicular to (a)(b)both, the current direction and polarization direction in the incident flux, i.e. it is directed along the z-axis. To calculate the magnitude, Pc, of the dipole moment it is natural to switch to spherical coordinates in which the incident polarization, Px = − is y, and the spin- Dout current density, iy = is, have the form is Dout r sin θ eφ, P = − where er, eθ, and eφ are the unit vectors along radial, polar, and azimuthal axes, respectively, see Fig. 2. is = is(sin θ er + cos θ eθ), (4) Induced dipole moment along z creates an electrostatic potential, 3 is(cid:0) 1 r (cid:1) ∂P φ reproduces Eq. (4). where the first term is ∂P φ out/∂r, while the second term out/∂θ. At large distances the current Eq. (11) There are two unknown constants, Pc and χs, in the expressions for electric field and spin polarization inside the sphere, and two unknown constants, Ein and P , in the corresponding expressions outside the sphere. These constants are determined from the four boundary condi- tions at r = a: (i) Continuity of the tangent component of electric field Ein = −Pc a3 . (12) (ii) Continuity of the normal component of the charge ϕout = Pc cos θ r2 , current (5) outside the sphere. From the form of ϕout we conclude that the θ- dependence of ϕ inside the sphere is also proportional to cos θ. This, together with Poisson's equation ∆ϕ = 0, suggests that the induced electric field, Ein, inside the sphere is homogeneous, so that ϕin = −Einr cos θ. (6) Substituting Eq. (2) into Eq.(3), and taking into ac- count that ∂Ein/∂xi = 0, we conclude that all the com- ponents of polarization inside the sphere satisfy the dif- fusion equation Din∆Pj + Pj τs = 0, (7) where Din is the diffusion coefficient inside the sphere. As we will see below, the polarization, P (r), has only φ-component inside the sphere and at all distances out- side the sphere. As in the incident flux, Eq. (4), the angular dependence of Pφ is ∝ sin θ. Outside the sphere, where ∆P = 0, the general form of Pφ is Pout = − is Dout (r + χs r2 ) sin θ eφ, (8) where the constant χs is the "spin polarizability". Inside the sphere, the solution of Eq. (7), proportional to sin θ, has the form Pin = P i1(r/λ) sin θ eφ, where P is a constant, and λ = (Dinτs)1/2 , (9) (10) is the diffusion length. The function i1(x) is a modified spherical Bessel function. We chose the function i1 be- cause it is finite at x = 0. While the polarization has only φ-component, the spin current, defined as a flow of the φ-component of spin, can be presented in the vector form iφ = is (1 − 2χs r3 ) sin θer + (1 + χs r3 ) cos θeθ , (11) (cid:104) (cid:105) a (cid:16) (cid:17) is Dout − a + χs a2 2eγDin P σinEin + i1(a/λ) = 2σoutPc a3 . (13) (iii) Continuity of the spin polarization = P i1(a/λ). (14) (iiii) Continuity of the spin flux though the boundary Din P λ i(cid:48)1(a/λ) + γ e σinEin = is a3 − 1 . (15) (cid:16) 2χs (cid:17) The system Eqs. (12)-(15) yields the sought expression for the spin-current-induced dipole moment Pc = − 6ea3γ (σin + 2σout)M is, (16) where M in denominator is the dimensionless combina- tion M = 2Dout Din − 2γ2σin σin + 2σout + ai(cid:48)1( a λ ) λi1( a λ ) (17) Naturally, the proportionality coefficient between Pc and the spin current contains the first power of the SO cou- pling strength, γ. The second term in Eq. (17) contains γ2, and can be safely neglected. The ratio Dout/Din can be replaced by σout/σin. It is seen from Eq.(17) that the factor M depends strongly on the relation between the radius of the sphere and the spin-diffusion length. For a (cid:28) λ the last term in Eq. (17) is 1, while for λ (cid:28) a it is big and equal to a/λ. In the latter case Eq. (16) yields Pc ∝ λa2. This dependence has a simple interpretation. Namely, for λ (cid:28) a the induced dipole is generated only inside a spherical layer of a thickness ∼ λ near the surface of the sphere, see Fig. 2. Description of a direct spin Hall effect for a sphere is completely similar to the case of the inverse spin Hall effect considered above. A charge current, ic, along the y direction generates a spin dipole moment, Ps, in the z-direction. Analytical expression for Ps is similar to Eq. (16) Ps = 3σina3γ e(σin + 2σout)DinM ic. (18) 4 FIG. 3. (Color online) The effective spin Hall voltage is the sum of contributions from individual SO-induced dipoles. With density of granules, n, the typical distance between the neighbors is n−1/3. It is much bigger than the radius, a, but much smaller than the sample width, L, which allows to re- place the sum by the integral Eq. (20). FIG. 4. (Color online) Dependence of the effective ISHE volt- age on a longitudinal magnetic field, ωL, is plotted from Eq. (25) for three different positions. y0, in the units of (dL)1/2, along the sample. Blue, violet, and green curves correspond to y0 = 2, y0 = 4, and y0 = 6 , respectively. B. Finite density of grains Consider a sample of a rectangular shape with a width, L, and thickness, d, (L (cid:29) d). As the injected spin cur- rent flows through the cross section, the voltage builds up between the edges z = ±L/2. The easiest way to calculate this voltage is to sum the contributions of in- dividual dipoles. If a grain is located at a point with coordinates (xi, yi, zi), see Fig. 3, then the potential dif- ference between the edges, created by an induced dipole reads V (xi, yi, zi) = ( L 2 − zi)Pc i + ( L i + y2 [x2 (cid:16) − − [x2 3 2 2 − zi)2] ( L 2 + zi)Pc i + ( L i + y2 2 + zi)2] (cid:17) , 3 2 (19) where Pc is given by Eq. (16). In calculating the effective inverse spin Hall voltage the summation over dipoles is replaced by integration (cid:90) d 2 (cid:90) ∞ (cid:90) L 2 reduces to multiplication by d. The final result reads V SH eff (y0) = 2ndPc (cid:34) (cid:16) L (cid:17) (cid:16) 2y0 (cid:17) ln × d + ln − ln d (cid:32)(cid:113) L2 (cid:113) L2 y2 0 y2 0 (cid:33)(cid:35) . (21) + 1 + L y0 + 1 + 1 + L y0 − 1 At small distances from the injection point, d (cid:28) y0 (cid:28) L, the first two terms in Eq. (21) dominate. The second logarithm describes a gradual increase of V SH eff with y0. At large distances, y0 (cid:29) L, the second and the third logarithms combine into ln(L/d) leading to the result V SH eff (∞) = 4ndPc ln 24e(na3)d ln( L = − (σin + 2σout) d )γ + ai(cid:48) 1( a λ ) λi1( a λ ) (cid:105) is. (22) Note, that for highly conducting grain, both factors in denominator do not depend on the characteristics, σout and Dout, of the matrix. eff depends strongly on the relation between λ and a. Overall, Eq. (21) describes the growth and subsequent saturation of the inverse spin Hall voltage. In this domain V SH (cid:16) L (cid:17) (cid:104) 2Dout d Din V SH eff (y0) = n dx − d 2 dzV (x, y, z), (20) dy − L 2 III. MAGNETIC-FIELD DEPENDENCE −y0 where y0 is the distance from the point at which voltage is measured to the point of spin-current injection. Natu- rally, the replacement of the sum by integral is justified when nL2d (cid:29) 1. The integration over y is straightfor- ward. Subsequent integral over z diverges logarithmically at z = L/2 and z = −L/2. This divergence should be cut off at (z ± L/2) ∼ d. Then the integration over x The behavior of V SH eff with position, y0, becomes non- trivial in the presence of magnetic field directed along the y-axis, a somewhat similar effect was pointed out in Ref. 27. If the magnetic field is weak, so that the Lar- mour frequency, ωL, is much smaller that τ−1 and much smaller than Dout/a2, which is the inverse diffusion time through the grain, then the effect of magnetic field on s isVISHe↵(y0)yz(xi,yi,zi)PcPcPcPcisLdLn1/3zx0.51.01.52.02.53.0-0.20.00.20.4VSHe↵(!L)[arb.units](!L/2Dout)1/2generation of electric dipole can be neglected. Instead, the field affects only the polarization in the spin current incident on the grain. This allows one to use the result Eq. (16) in calculation the ωL-dependence of V SH eff . and Pz, satisfy the system of equations: Dout Outside the grains, the polarization components, Px d2Px dy2 + d2Pz dy2 − ωLPx = 0. Assuming that ωLPz = 0 and Dout at the point of injection the polarization was along x, we find (cid:104)(cid:16) ωL (cid:104)(cid:16) ωL 2Dout (cid:17) 1 (cid:17) 1 2 2 (cid:105) (cid:105) y y 2Dout exp exp (cid:104) (cid:104) (cid:16) ωL (cid:16) ωL 2Dout (cid:17) 1 (cid:17) 1 2 2 (cid:105) (cid:105) y , y . 2Dout − − (23) Px(y) =Px(0) cos Pz(y) =Px(0) sin (cid:34) (cid:112) dy (cid:90) ∞ 0 1 (y − y0)2 + d2 − V SH eff (y0, ωL) = 2ndPc (cid:112) 1 (y − y0)2 + L2 5 Suppose that a grain is positioned at y = y0. Then the induced dipole moment will be a vector orthogonal to polarization with components (cid:18) Px(y0) (cid:19) (cid:18) Pz(y0) (cid:19) Pc, Px(0) Px(0) Pz(y0) = Px(y0) = − Pc, (24) where Pc, given by Eq. (16), is proportional to the mag- nitude of the spin current, is, which does not change in the presence of magnetic field. To proceed further, we notice that only Pz-component of the induced dipole moment contributes to the buildup of V SH eff and should be substituted into Eq. (19) instead of Pc. We first perform integration over z and x. The remaining integral over y takes the form (cid:35) (cid:104)(cid:16) ωL 2Dout (cid:17) 1 2 (cid:105) y cos exp (cid:104) (cid:16) ωL 2Dout (cid:17) 1 2 (cid:105) . y − (25) For ωL = 0 Eq. (25) reproduces the limiting cases of Eq. (21). With characteristic distance y0, being ∼ L, we conclude that characteristic magnetic field is ωL = 1 T = Dout L2 , (26) which is a natural scale at which the diffusion time through a square with a side L is equal to the Larmour period. Simple asymptotic expressions for V SH eff can be obtained in the domain ωL (cid:29) ωL, when the second term in the integrand can be neglected: (cid:16) (cid:17)1/2 Dout/ωL (i) d (cid:28) y0 (cid:28) In this limit, the log- divergence at large y is cut off at y ∼ (Dout/ωL)1/2, and we get . V SH eff (y0, ωL) = 2ndPc ln (cid:17)1/2 (cid:16) Dout/ωL (ii) y0 (cid:29) . We can now neglect y com- pared to y0 in the square brackets. Then the integration can be easily performed yielding (cid:16)(cid:114) Dout y0 d2 ωL (cid:17) (cid:16) Dout (cid:17) 1 2 FIG. 5. (Color online) The effective spin diffusion length in the units (3π2na)−1/2 is plotted from Eq. (36) for ratios Dout/Din: Dout/Din = 1 (green), Dout/Din = 5 (purple), and Dout/Din = 10 (blue). Note the saturation of λeff at small λ . (27) V SH eff (y0, ωL) = ndPc ωLy2 0 IV. EFFECTIVE SPIN DIFFUSION LENGTH . (28) The asymptotes Eq. (27), (28) do not cover the entire domain of ωL. At the crossover field ωL ∼ Dout/y2 0 . Eq. (27) exceeds Eq. (28) by a large factor ∼ ln(y0/d). As the magnetic-field dependence of voltage is plotted numeri- cally, see Fig. 4, it appears that in the intermediated do- main the ISHE voltage exhibits two sign reversals. This means that the oscillations in Eq. (23) do not average out completely after integration over the positions of the spheres. There are two reasons why the effective spin-diffusion length of the mixture exceeds λ. The first reason is obvi- ous: the grains are sparse and there is no spin relaxation in between the grains. The second reason is much more subtle and becomes important when λ is much smaller than the grain radius. Namely, the rate of the spin re- laxation at the grain surface is suppressed. Formally, this suppression, illustrated in Fig. 2, follows from the behavior of polarization inside the grain 05101520253001234a/e↵/(3⇡2na)1/26 Substituting Eq. (33) into Eq. (34), we readily find τeff = Dinτs 3π2na3DoutF( a λ ) . (35) Note that the product in the numerator is equal to λ2. We can now use the expression for the effective relaxation time to find the effective spin-diffusion length λeff = Doutτeff = (36) (cid:2)3π2na3F( a λ )(cid:3)1/2 . λ (cid:112) Din Let us trace the decrease of λeff as the spin-diffusion rate inside the sphere gradually decreases. For λ (cid:29) a, the function F(x) can be replaced by F(0) = 4(2Dout+Din) . Thus the enhancement of the spin-diffusion length due to patterning the SO material into granules is ∼ (na3)−1/2. In the opposite limit λ (cid:28) a we have F(x) ≈ 1/x2. This leads to the unexpected conclusion that in this limit λeff saturates at the value ∼ (na)1/2 . The origin of this sat- uration is suppression of polarization at the surface, the effect discussed above and further elaborated on in Ap- pendix. 1 V. DISCUSSION (21), (22) and Eq. (i). The two main results of the present paper are Eqs. (36) for the effective inverse spin Hall voltage and the effective spin diffusion length of the mixture. It is convenient to cast Eq. (22) in the form of the relation between the effective spin-Hall angle, θSH eff , of the mixture and the spin-Hall angle, θSH, of the material of the grain. The spin Hall angle is defined as the proportionality coefficient between the charge and spin current densities, more precisely, jc = θSH(2e/)js. Then Eq. (22) takes the form (cid:16) L (cid:17) 12(na3) ln + ai(cid:48) 2Dout d 1( a λ ) λi1( a λ ) Din (cid:16) i1( r (cid:17) λ ) i1( a λ ) 3ai1( r λ ) λ ) + Din Pin(r) = − 3a DinM is sin θ eφ = − 2Douti1( a is sin θ eφ. (29) a λ i(cid:48)1( a λ ) It is seen from Fig. 2 that, for λ = a/12, the radial distribution of Pin(r) not only falls off rapidly from the surface towards the center, but its value at the surface is small. Physical origin of this smallness is elucidated in the Appendix. While our goal is to find λeff, in order not to deal with boundaries we first calculate the effective spin relaxation time of the mixture. Spin relaxation takes place only inside the spheres. If at time t = 0 the polarization inside the sphere is distributed according to Eq. (29), then the rate of decay of this polarization is given by the integral over the volume of the sphere (cid:90) (cid:104) R = 1 τs (cid:105) φ dΩ Pin(r) . (30) the form, = Using explicit (x cosh(x) − sinh(x)) /x2, the modified spherical Bessel function, the integral can be evaluated, and the result can be cast in the form i1(x) of (cid:16) a (cid:17) λ R = 3π2a4is Dinτs F , (31) where the dimensionless function F(x) is defined as F(x) = x sinh(x) − 2 cosh(x) + 2 . 2( Dout Din − 1)[x cosh(x) − sinh(x)]x + x3 sinh(x) (32) The result Eq. (31) can be also expressed through the polarization outside the sphere by replacing is by PoutDout/a, see Eq. (8). One has (cid:16) a (cid:17) λ R = 3π2a3PoutDout Dinτs F . (33) θSH eff = θSH, (37) In the absence of spin current, the spin relaxation in- side the spheres causes the time decay of the spin po- larization in the medium between the spheres. This is because diffusing carriers eventually "hit" a sphere. Con- sider an interval (y0− δy 2 ), and assume that there are hard walls at the ends, so electrons do not flow in or out. Then the initial net polarization, Pout(y0)δy, in- side the interval will decay with some effective rate τ−1 eff . To find this rate, we substitute the the two-dimensional density of spheres in the interval, nδy, into the balance equation 2 , y0 + δy Pout(y0)δy τeff = nδyR. (34) where we assumed σin (cid:29) σout. Essentially, the propor- tionality between θSH eff and θSH is determined by a "volume factor", na3. Note, however, that θSH is the characteris- tics of a homogeneous SO-film, only as long as the film thickness, w, is much smaller than λ. For w (cid:29) λ, θSH falls off as λ/w. At the same time, the decay of θSH eff with y0 sets in only when y0 exceeds the effective spin diffusion length of the mixture. This length is much bigger than λ, as it was shown in Section IV. (ii) Note that, strictly speaking, Eq. (36) describes λeff only within a numerical factor. This factor was lost as we replaced is by PoutDout/a, assuming that the first term in Eq. (8) dominates. In fact, precisely at r = a, the two terms almost cancel each other. Indeed, substituting the expression for χs into Eq. (8), we can cast it in the form (cid:32) Pout = − is Dout a3 r2 + r − (cid:104) 3a3 2 + aDini(cid:48) 1( a λ ) λDouti1( a λ ) (cid:33) (cid:105) r2 sin θ eφ. (38) In the limit λ (cid:28) a and r = a, the expression in the brackets is equal to 3λDout/Din, and thus, is much smaller than a. However, for bigger r ∼ a the compensation of the first two terms does not take place, and the relation is ∼ PoutDout/a holds. The suppression of Pout near the surface of the sphere, expressed by Eq. (38), is the reason why λeff saturates when λ → 0, see Fig. 5. Loosely speaking, strong re- laxation "repels" the spins from the boundary, which, in turn, slows down the effective relaxation. The above physics is quite general. To illustrate it, in the Appendix we consider a model example of diffusion of particles in the presence of an absorbing trap and demonstrate that, with increasing the absorption rate, the concentration of particles vanishes at the position of the trap. (iii). It is instructive to compare our result Eq. (22) with the expression for the perturbation of spin cur- rent flowing in a normal metal around a ferromagnetic sphere28. Rather that the SO coupling in our case, the difference of spin-up and spin-down carriers in Ref. 28 is caused by the difference of their conductivities inside the ferromagnet. As a result, the induced dipole mo- ment in our case is normal to the spin current, while the induced "spin dipole moment"28 is along the spin cur- rent. Other than that, the two expressions resemble each other. There is, however, an important difference. If the conductivity of the ferromagnetic sphere28 is much higher than the conductivity of the surrounding normal medium, then the perturbation of the spin current is sup- pressed (resistance mismatch). On the contrary, for the inverse spin Hall effect, the bigger is the ratio Din/Dout, the stronger is the modification of the spin current out- side the sphere. (iiii). For a quantitative example of the effect of granu- larity on the effective parameters of the mixture, assume that the density of the SO granules is na3 = 10−2, while the spin diffusion length in the material of the granule is λ = 0.2a. Compared to the geometry in Fig. 1 with no granularity we "lose" 100 times in the inverse spin Hall voltage. At the same time, we gain in λeff. Substituting λ = 0.2a into Eq. (36), and assuming Din (cid:29) Dout, we find λeff = 10λ. VI. CONCLUDING REMARKS 1. For experimentally verifying our theoretical results, composites of SO and no SO materials can be prepared using a variety of widely available fabrication techniques. For example, in Ref. 29, authors used a pulsed laser de- position technique to prepare a composite comprising of gold nanoclusters embedded in ZnO matrix. In Ref. 30, 7 a self-assembly approach was used to fabricate a compos- ite comprising of nickel nanoclusters embedded in amor- phous Al2O3 matix. In Ref. 31, a nanofabrication ap- proach was employed to prepare a magnonic crystal com- prising of cobalt nanodots embedded in a permalloy film. Similar approaches can be used to prepare the desired composite structures of SO and no SO materials, say Pt or Au nanodots (with large θSH) embedded in films of low θSH materials (such as copper, molybdenum or even semiconductors like silicon). 3. 2. From device point-of-view, an obvious way to en- hance the spin diffusion length would be by creating a 1D structure of alternating SO and no SO layers To achieve this, however, the thickness of the SO-layer should be smaller than λ. Conversely, in granular system the en- hancement takes place when λ (cid:28) a. This is because the spin current can flow around the spheres. In numerous spin-pumping experiments, see e.g. Refs. 4 -- 17, the measured quantity, V SH, is proportional either to θSHλ, when the thickness of non-magnetic ma- terial is much bigger than λ, or simply to θSH in the op- posite limit. A comprehensive list of experimental values of θSH and λ for a number of heavy metals can be found in Ref. 32. This list indicates that, while, separately, θSH, and λ vary within wide ranges, the range of change of their product is much narrower, see also Ref. 33. Overall, there is still experimental ambiguity in ex- tracting the intrinsic SO parameters of materials from the experiment. In this regard, granularity can offer a help, by bringing a new spatial scale, the radius of the grain, a. As shown in Fig. 5, the value λeff depends very strongly on the relation between λ and a. 4. In a specific case of a semiconductor ZnO the inverse spin Hall effect was studied both in pumping experiment34 and directly by measuring the nonlocal voltage35. In both measurements the value θSH was found to be anomalously big, compared e.g. to Si14,15 . It has recently been shown36 that the value θSH in ZnO can be tuned very sensitively by changing the oxygen ambient under which it is grown37. Films prepared under high oxygen rich environment showed a large value for θSH (∼ 0.1), while the films prepared under low oxygen am- bient showed an order of magnitude lower value of θSH. VII. ACKNOWLEDGEMENTS The authors are grateful to E. G. Mishchenko for a use- ful discussion. The work was supported by NSF through MRSEC DMR-1121252. Appendix A Consider a diffusion in one dimension. If at time t = 0 the distribution of particles is a δ-peak, i.e. n(x, 0) = (cid:90) ∞ 0 8 The normalized solutions, ψk(x), which satisfy Eq. (A4) have the form ψk(x) = 1 π1/2 cos(kx + ϕk), where the phase ϕk is found from the condition tan ϕk = − a 2Dτsk , (A7) (A8) imposed by Eq. (A6). Upon substituting Eq. (A5) into Eq. (A3) and using the initial condition, we arrive at the final result (cid:90) ∞ 0 n(x, t) = 1 π cos kx + a dk 1 + ( 2Dτsk sin kx 2Dτsk )2 a e−Dk2t. (A9) It is now convenient to introduce a dimensionless coor- dinate x = x/(Dt)1/2 and the dimensionless time depen- dent parameter t = a2t/4Dτ 2 s . In new variables Eq. (A9) assumes the form n(x, t) = a 2πDτs√t q2 cosxq + √tq sinxq q2 + t dq e−q2 . (A10) We see that the characteristics of the trap, a and τs, enter only into rescaling of time. In Fig. 6 we plot Eq. (A10) for four different t. It is seen from Fig. 6 that, with time, the density n(0, t) at the origin develops a dip. The smaller is τs, i.e. the more absorbing is the trap, the sharper is the dip. This conclusion also follows from the long-time asymptote of n(x, t), when we neglect q2 in denominator compared to t. Then the integration yields n(x, t) (cid:12)(cid:12)(cid:12)t(cid:29)1 (cid:17) (cid:16)(cid:112) (cid:17) (cid:16) ax tx + 1 + 1 3 2 2Dτs = = a 8√πDτst √Dτ 2 s√πa2t 3 2 e− x2 4 e− x2 4Dt . (A11) This asymptote indicates that the ratio of concentrations at half-width, x ≈ (Dt)1/2, and at the origin is ∼ t1/2, i.e. the dip is deep. The next question we ask ourselves is how the to- tal number of particles(cid:82) dx n(x, t) decreases with time. Upon integration Eq. (A10), we get (cid:90) ∞ Then the expression for G(x,x',t) reads N (t) = dx n(x, t) G(x, x(cid:48), t) = ψk(x)ψk(x(cid:48)) exp(−Dk2t). (A5) −∞ =Erfc(t 2 ) exp(t) = 1 The second term in Eq. (A4) plays the role of delta- potential barrier, and causes the discontinuity of the derivative of ψk (cid:12)(cid:12)(cid:12)x=0+ − ∂ψk ∂x (cid:12)(cid:12)(cid:12)x=0− = ∂ψk ∂x a Dτs ψk(0). (A6) where Erfc(s) is the complementary error function. It is seen from Eq. (A12) that the change of the decay rate 1 − a (cid:16) t (cid:16) πD (cid:17) 1 τs 2τs a t 2 πD 2 (cid:17) 1 , t (cid:29) 1, , t (cid:28) 1, (A12) FIG. 6. Diffusive spreading of the initial particle distribution n(x, 0) = δ(x) in the presence of an absorbing trap located at x = 0 is described by Eq. (A10). Shown is n(x, t) at a fixed time, t0, for different absorption efficiencies, κ, Eq. (A13). The more absorbing the trap is, the deeper is the dip at the origin, and the slower is the decay of the net number of particles at long times, as follows from Eq. (A12). δ(x), then it spreads with time as n(x, t) = 1 2√πDt exp (cid:16) (cid:17) x2 4Dt − (A1) where D is the diffusion coefficient. Suppose now, that an absorbing trap is placed at the coordinate origin. Then the spreading is governed by the equation ∂n ∂t − = −D ∂2n ∂x2 + aδ(x)n τs (A2) where a is the size of the trap, and τs is the absorption rate. Then the time-dependent concentration, n(x, t), can be expressed through the Green function of Eq. (A2) n(x, t) = dx(cid:48)G(x, x(cid:48), t)n(x(cid:48), 0) (A3) It is convenient to present the Green function in terms of eigenfunctions, ψk of Eq. (A2), which satisfy the Schodinger-like equation ∂2ψk ∂x2 + − D aδ(x) τs ψk = k2ψk. (A4) (cid:90) (cid:88) k -6-4-202460246-6-4-202460.000.050.100.150.200.250.30-6-4-202460.0000.0010.0020.003n(x,to)n(x,to)n(x,to)(a)(b)(c)(x/Dto)1/2(x/Dto)1/2(x/Dto)1/2(a)=0.2(b)=2(c)=20to=⌧s[arb.units][arb.units][arb.units]9 ∂N/∂t takes place at t ∼ 1. This change is caused by the development of the dip. Indeed, for t (cid:29) 1 the decay rate falls off with time as t−3/2. Overall, we conclude that the spreading of the parti- cle density in the presence of a trap is governed by a dimensionless parameter then, n(x, t) will develop a dip at x = 0 and the decay of the net number of particles will proceed even slower. For large efficiency, the dip will developed early, namely at t ∼ τs/κ2 (cid:28) τs, after which time the decay of N (t) will be governed by the value of n(0, t) of the concentration at the dip. κ = a (Dτs)1/2 , (A13) which is the dimensionless efficiency of absorption by the trap. If this efficiency is small, the spreading will proceed as in the absence of the trap for most of the time, until the concentration at x = 0 becomes really small. Only Formation of a dip in our model problem puts into a general perspective the behavior of the effective spin diffusion length in the system of the SO grains. In the limit λ (cid:28) a, see Eq. (36), the value λeff saturates because the polarization near the boundary gets suppressed as a result of the development of a local minimum. 1 M. I. Dyakonov and V. I. Perel, Pis'ma Zh. Eksp. Teor. Fiz. 13, 657 (1971) [Sov. Phys. JETP Lett. 13, 467 (1971)]. 2 M. I. Dyakonov and V. I. Perel, Phys. Lett. A 35, 459 (1971). 3 J. E. Hirsch, Phys. Rev. Lett. 83, 1834 (1999). 4 E. Saitoh, M. Ueda, H. Miyajima, and G. Tatara, Appl. Phys. Lett. 88, 182509 (2006). ringhaus, Nat. Mater. 12, 622 (2013). 19 S. Watanabe, K. Ando, K. Kang, S. Mooser, Y. Vaynzof, H. Kurebayashi, E. Saitoh, and H. Sirringhaus, Nat. Phys. 10, 308 (2014). 20 Z. Qiu, M. Uruichi, D. Hou, K. Uchida, H. M. Yamamoto, E. Saitoh, AIP Advances 5, 057167 (2015). 21 M. Kimata, D. Nozaki, Y. Niimi, H. Tajima, and Y. Otani, 5 H. Y. Inoue, K. Harii, K. Ando, K. Sasage, and E. Saitoh, Phys. Rev. B 91, 224422 (2015). J. Appl. Phys. 102, 083915 (2007). 6 K. Ando, S. Takahashi, K. Harii, K. Sasage, J. Ieda, S. Maekawa, and E. Saitoh, Phys. Rev. Lett. 101, 036601 (2008). 7 K. Ando, T. Yoshino, and E. Saitoh, Appl. Phys. Lett. 94, 152509 (2009). 8 J.-C. Rojas-S´anchez, N. Reyren, P. Laczkowski, W. Savero, J.-P. Attan´e, C. Deranlot, M. Jamet, J.-M. George, L. Vila, and H. Jaffr`es, Phys. Rev. Lett. 112, 106602 (2014). 9 K. Ando, S. Takahashi, J. Ieda, Y. Kajiwara, H. Nakayama, T. Yoshino, K. Harii, Y. Fujikawa, M. Matsuo, S. Maekawa, and E. Saitoh, J. Appl. Phys. 109, 103913 (2011). 10 O. Mosendz, J. E. Pearson, F. Y. Fradin, G. E. W. Bauer, S. D. Bader, and A. Hoffmann, Phys. Rev. Lett. 104, 046601 (2010). 11 B. Heinrich, C. Burrowes, E. Montoya, B. Kardasz, E. Girt, Y.-Y. Song, Y. Sun, and M. Wu, Phys. Rev. Lett. 107, 066604 (2011). 12 J. E. G´omez, B. Z. Tedlla, N. R. ´Alvarez, G. Alejandro, E. Goovaerts, and A. Butera, Phys. Rev. B 90, 184401 (2014). 13 A. Yamamoto, Y. Ando, T. Shinjo, T. Uemura, and M. Shiraishi, Phys. Rev. B 91, 024417 (2015). 14 K. Ando and E. Saitoh, Nat. Commun. 3, 629 (2012). 15 E. Shikoh, K. Ando, K. Kubo, E. Saitoh, T. Shinjo, and M. Shiraishi, Phys. Rev. Lett. 110, 127201 (2013). 16 S. Dushenko, M. Koike, Y. Ando, T. Shinjo, M. Myronov, and M. Shiraishi, Phys. Rev. Lett. 114, 196602 (2015). 17 S. Singh, A. Ahmadi, C. T. Cherian, E. R. Mucciolo, E. del Barco, and B. Ozyilmaz, Appl. Phys. Lett. 106, 032411 (2015). 18 K. Ando, S. Watanabe, S. Mooser, E. Saitoh, and H. Sir- 22 B. Khachatryan, T. D. Nguyen, Z. V. Vardeny, and E. Ehrenfreund Phys. Rev. B 86, 195203 (2012). 23 C.-X. Sheng, S. Singh, A. Gambetta, T. Drori, M. Tong, S. Tretiak, and Z. V. Vardeny Sci. Rep. 3 (2013). 24 D. Sun, K. J. van Schooten, H. Malissa, M. Kavand, C. Zhang, C. Boehme, and Z. V. Vardeny, unpublished. 25 Z. G. Yu, Phys. Rev. Lett. 106, 106602 (2011). 26 Z. G. Yu, Phys. Rev. B 85, 115201 (2012). 27 M. I. Dyakonov, Phys. Rev. Lett. 99, 126601 (2007). 28 R. C. Roundy and M. E. Raikh Phys. Rev. B 91, 045202 (2015). 29 A. Tiwari, A. Chugh, C. Jin, and J. Narayan, J. Nanosci. Nanotechnol. 3, 368 (2003). 30 J. Narayan and A. Tiwari, J. Nanosci. Nanotechnol. 4, 726 (2004). 31 see e.g. S. Tacchi, G. Duerr, J. W. Klos, M. Madami, S. Neusser, G. Gubbiotti, G. Carlotti, M. Krawczyk, and D. Grundler Phys. Rev. Lett. 109, 137202 (2012), and refer- ences therein. 32 H. L. Wang, C. H. Du, Y. Pu, R. Adur, P. C. Hammel, and F. Y. Yang, Phys. Rev. Lett. 112, 197201 (2014). 33 P. Laczkowski, J.-C. Rojas-S´anchez, W. Savero-Torres, H. Jaffr`es, N. Reyren, C. Deranlot, L. Notin, C. Beign`e, A. Marty, J.-P. Attan`e, L. Vila, J.-M. George, and A. Fert, Appl. Phys. Lett. 104, 142403 (2014). 34 J.-C. Lee, L.-W. Huang, D.-S. Hung, T.-H. Chiang, J. C. A. Huang, J.-Z. Liang, and S.-F. Lee, Appl. Phys. Lett. 104, 052401 (2014). 35 M. C. Prestgard and A. Tiwari, Appl. Phys. Lett. 104, 122402 (2014). 36 M. C. Prestgard, Z. Yue, M. E. Raikh and A. Tiwari, in preparation. 37 M. C. Prestgard, Ph.D thesis, University of Utah, 2015. 

1806.04486

1

1806

Anomalously low dielectric constant of confined water

The dielectric constant of interfacial water has been predicted to be smaller than that of bulk water (= 80) because the rotational freedom of water dipoles is expected to decrease near surfaces, yet experimental evidence is lacking. We report local capacitance measurements for water confined between two atomically-flat walls separated by various distances down to 1 nm. Our experiments reveal the presence of an interfacial layer with vanishingly small polarization such that its out-of-plane dielectric constant is only approximately 2. The electrically dead layer is found to be two to three molecules thick. These results provide much needed feedback for theories describing water-mediated surface interactions and behavior of interfacial water, and show a way to investigate the dielectric properties of other fluids and solids under extreme confinement.

cond-mat.mes-hall

cond-mat

Anomalously low dielectric constant of confined water L. Fumagalli1,2*, A. Esfandiar3, R. Fabregas4,5, S. Hu1,2, P. Ares1,2, A. Janardanan1,2, Q. Yang1,2, B. Radha1,2, T. Taniguchi6, K. Watanabe6, G. Gomila4,5, K. S. Novoselov1,2, A. K. Geim1,2* 1 School of Physics & Astronomy, University of Manchester, Manchester M13 9PL, UK. 2 National Graphene Institute, University of Manchester, Manchester M13 9PL, UK. 3 Department of Physics, Sharif University of Technology, P.O. Box 11155-9161, Tehran, Iran. 4 Departament d'Electrònica, Universitat de Barcelona, C/ Martí i Franquès 1, 08028 Barcelona, Spain. 5 Institut de Bioenginyeria de Catalunya, C/ Baldiri i Reixac 15-21, 08028 Barcelona, Spain. 6 National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan. * Email: [email protected]; [email protected] is lacking. We report for water confined between The dielectric constant ε ε ε ε of interfacial water has been predicted to be smaller than that of bulk water (ε ε ε ε ≈≈≈≈ 80) because the rotational freedom of water dipoles is expected to decrease near surfaces, yet experimental evidence local capacitance measurements two atomically-flat walls separated by various distances down to 1 nm. Our experiments reveal the presence of an interfacial layer with vanishingly small polarization such that its out-of-plane εεεε is only ~ 2. The electrically dead layer is found to be two to three molecules thick. These results provide much needed feedback for theories describing water-mediated surface interactions and behavior of interfacial water, and show a way to investigate the dielectric properties of other fluids and solids under extreme confinement. ion solvation, molecular Electric polarizability of interfacial water determines the strength of water-mediated intermolecular forces, which in turn impacts a variety of phenomena including surface hydration, transport through reactions and macromolecular nanopores, chemical assembly, to name but a few1-3 . The dielectric properties of interfacial water have therefore attracted intense interest for many decades4-7 and, yet, no clear understanding has been reached8-11. Theoretical12-14 and experimental studies15-17 have shown that water exhibits layered structuring near surfaces, suggesting that it may form ordered (ice-like) phases under ambient conditions. Such ordered water is generally expected to exhibit small polarizability because of surface-induced alignment of water molecular dipoles which are then difficult to reorient by applying an electric field7-10 . Despite a massive amount of literature dedicated to the subject (see, for example, refs 4-11), the dielectric its depth remain constant of essentially are challenging. interfacial water and unknown because measurements The previous experiments to assess ε of interfacial water mostly relied on broadband dielectric spectroscopy applied to large-scale naturally-occurring systems such as nanoporous and dispersions4,5,10,18,19. These systems allow sufficient amount powders crystals, zeolite in interfacial water for carrying out capacitance of measurements but the involved complex geometries require adjustable parameters and extensive modelling, which result large and poorly controlled experimental uncertainties. For example, the extracted values of ε are strongly dependent on assumptions about the interfacial layer thickness. For the lack of direct probes to measure the polarizability of interfacial water, most evidence has come so far from molecular dynamics (MD) simulations, which also involve certain assumptions. These studies generally predict that the polarizability should be reduced by approximately an order of magnitude7-9 but the quantitative accuracy of these predictions is unclear because the same simulation approach struggles to reproduce the known ε for bulk water phases20. In this work, we used slit-like channels of various heights h which could be controllably filled with water. The channels were incorporated into a capacitance circuit with exceptionally high sensitivity to local changes in dielectric properties, which allowed us to determine the out-of-plane dielectric constant ε⊥ of the water confined inside. The studied devices were fabricated by van der Waals assembly21 using three atomically flat crystals of graphite and hexagonal boron nitride (hBN) following a recipe reported previously22,23 (see Supplementary Method S1 and Fig. S1). Here graphite serves as a bottom layer for the assembly as well as the ground electrode in capacitance measurements (Fig. 1a). Next a spacer layer was placed on top of graphite. It was an hBN crystal patterned into parallel stripes. The assembly was completed by placing another hBN crystal on top (Figs 1b,c). The spacer determined the channels' height h, and the other two crystals served as top and bottom walls. The reported channels were usually ∼ 200 nm wide and several micrometers long. Each of our devices for a given h contained several channels in parallel (Fig. 1), which ensured high reproducibility of our measurements and reduced statistical errors. When required, the channels could be filled with water through a micrometer-size inlet etched in graphite from the back22,23 (Fig. 1a). 1  Fig. 1 Experimental setup for dielectric imaging. a Its schematic. The top layer and side walls made of hBN are shown in blue; graphite serving as the ground electrode is in black. The three-layer assembly covers an opening in a silicon nitride membrane (light brown). The channels are filled with water from the back. The AFM tip, kept always in a dry nitrogen atmosphere, served as the top electrode. b,c Cross-sectional schematics before (b) and after (c) filling the channels with water (not to scale). d Three-dimensional topography image of one of the devices. e-g AFM topography of the sagged top hBN for devices with different h before filling them with water. Scale bars: 500 nm. h-j Topography profiles for the top layer (black) and the part not covered by hBN (cyan) as indicated by color- coded lines in (d). Red curves: Same devices after filling with water. To probe ε of water inside the channels, we employed scanning dielectric microscopy based on electrostatic force detection with an atomic force microscope (AFM), adapting the approach described in ref. 24. Briefly, by applying a low-frequency ac voltage between the AFM tip and the bottom electrode, we could detect the tip-substrate electrostatic force, which translates into the first derivative of the local capacitance dC/dz in the out-of-plane direction z. By raster-scanning the tip, a dC/dz (or "dielectric") image was acquired, from which local dielectric properties could be reconstructed (see Supplementary Methods S3 and S4). Note these measurements. First, hBN is highly insulating, which allows the electric field generated by the AFM tip to reach the subsurface water without being screened. It is also highly beneficial to have hBN as the side walls (spacers) because this provides a simple reference for comparison between the dielectric properties of hBN (ε⊥ ≈ 3.5)25 and the nearby water of the same thickness (Fig. 1c). As shown below, the latter arrangement yielded a clear dielectric contrast proving that the dielectric constant of confined water strongly changes with decreasing h, independently of the modelling. the use of hBN that is essential for Unlike the previous reports22,23, we chose to use relatively thin (30-80 nm) top crystals. This not only allowed us to reach closer to the subsurface water but also to control that the channels were fully filled during the capacitance measurements (see below) (Supplementary Method S2 and Fig. S2). If there was no water inside, the top hBN exhibited notable sagging22 as illustrated in Fig. 1b. Figures 1e-g show AFM topographic images for representative devices with h ≈ 10, 3.8 and 1.4 nm under dry conditions. All of them exhibit some sagging, and its extent depends on thickness of the top hBN22 (black curves in Figs 1h-j). The channel heights h could also be determined from the same images using the areas that were not covered by the top hBN layer (cyan curves). Such initial imaging as well as dielectric imaging after filling the channels was carried out at room temperature and the whole AFM chamber was filled with dry nitrogen. Figures 2a-c show AFM topographic images for the same three devices and the same scan areas as in Figs 1e-g but after filling the channels with water, which was done by exposing the backside of our devices to deionized water22 (Fig. 1a). As the channels became filled through the inlet in the bottom graphite, this lessened adhesion between the side and the sagging top walls and, consequently, 2  Fig. 2 Dielectric imaging of confined water. a-c Topographic images of the three devices in Fig. 1 after filling them with water. Scale bars: 500 nm. d-f Corresponding dC/dz. The shown images were obtained by applying a tip voltage of 4 V at 1 kHz (other voltages and frequencies down to 300 Hz yielded similar images). Commercial cantilevers with tips of 100-200 nm in radius were used to maximize the imaging sensitivity. g Averaged dielectric profiles across the channels in (d-f). h Simulated dC/dz curves as a function of ε⊥ for the known geometries of the three shown devices (Shown are the peak values in the middle of the channels). Symbols are the measured values of dC/dz from (g). Their positions along the x-axis are adjusted to match the calculated curves. Bars and light-shaded regions: Standard errors as defined in Supplementary Information. diminished (Fig. 1c). The top hBN covering water-filled channels became practically straight with little topographic contrast left, independently of h (red curves in Figs 1h-j). The corresponding dielectric images for the discussed devices after their filling are shown in Figs 2d-f. One can see very strong contrast which moreover reverses with h. For the case of the 10 nm-channels, the red regions containing subsurface water indicate ε⊥ larger than that of hBN, as expected (Figs 2d and 2g, red). On the contrary, for the dielectric contrast practically disappeared (Figs 2e and 2g, cyan) whereas the 1.4 nm-thick water exhibited the opposite, negative contrast (Figs 2f and 2g, blue). The images show that the polarizability of confined water strongly depends on its thickness h and can reach values smaller than that of hBN with its already modest ε⊥ ≈ 3.5. As mentioned above, a reduction in ε⊥ for strongly confined water is generally expected on the basis of atomistic simulations7-9 but the observed decrease is much stronger than predicted (ε⊥ ≈ 10) or commonly assumed in the literature. Figure 2h shows the resulting curves for the discussed three devices in Figs 2a-c. By projecting the measured capacitive signals (symbols on the y-axis of Fig. 2h) onto the x-axis, we find ε⊥ ≈ 15.5, 4.4 and 2.3 for h ≈ 10, 3.8 and 1.4 nm, respectively. We emphasize that ε⊥ is the only unknown in our model as all the other parameters were determined experimentally. Also, note that some devices exhibited small (few Å) residual sagging in the filled state (see, e.g., Figs 2a,c). If not taken into account, this effect can lead to systematic albeit small errors in determining ε⊥ (by effectively shifting the calculated curves in the y-direction). Our calculations included this residual sagging, too (Fig. S5). the 3.8 nm-thick water, We repeated such experiments and their analysis for more than 40 devices with h ranging from ∼ 1 to 300 nm. The results are summarized in Fig. 3 which shows the found ε⊥ as a function of h. The bulk behavior (ε⊥ ≈ 80) recovers only for water as thick as ∼ 100 nm, showing that the confinement can affect the dielectric properties of even relatively layers (Fig. S6). At smaller thicknesses, ε⊥ evolves approximately linearly with h and approaches a limiting value of ∼ 2.1 ± 0.2 at h < 2 nm where only a few layers of water can fit inside the channels. Note is independent of varying details of our experimental geometries such as, e.g., thickness of the top hBN layer and the AFM tip radius (Fig. S7). the functional dependence thick water To quantify the measured local capacitance and find ε⊥ for different water thicknesses, we use a three-dimensional electrostatic model that takes into account the specific geometry of the measured devices as well as of the used AFM tips (see Supplementary Method S5 and Figs S3, S4). The model allows numerical calculation of dC/dz as a function of ε⊥ for a dielectric material inside the channels. that in Fig. 3 3 Although the found ε⊥ remains anomalously small (< 20) over a wide range of h up to 20 nm (Fig. 3), this does not actually mean that the polarization suppression extends over the entire volume of the confined water. Indeed, the capacitance response comes from both interfacial and inner molecules, effectively averaging their contributions over the channel thickness. To this end, we recall that water near solid surfaces is believed to have a pronounced layered structure which extends approximately 10 Å into the bulk12- 17. Accordingly, the found dependence ε⊥(h) can be attributed to a cumulative effect from the thin near-surface layer with the low dielectric constant εi whereas the rest of the water has the normal, bulk polarizability, εbulk ≈ 80. The overall effect can be described by three capacitors in series as shown in the inset of Fig. 3. This model yields the effective ε⊥ = h/[2hi /εi + (h - 2hi)/εbulk] where hi is the thickness of the near-surface layer. Its εi can be taken as ≈ 2.1 in the limit of small h if we assume that the layered 13) structure does not change much with increasing h (ref. and is similar at both graphite and hBN surfaces, as the MD simulations predict14. Figure 3 shows that the proposed simple model describes well the experimental data, allowing an estimate for the thickness hi of interfacial water with the suppressed polarization (see Supplementary Method S9 and Fig. S8). Within the experimental error, our data yield hi ≈ 7.5 ± 1.5 Å, in agreement with the expected layered structure of water14-17. In other words, the electrically dead two-three molecular diameters away from the surface. This is also consistent with the thickness h = 1.5-2 nm – the double of hi – where the limiting value ε⊥ ≈ 2.1 is reached (see Fig. 3), which can be understood as the distance at which the near-surface layers originating from top and bottom walls begin to merge. layer extends To conclude, we have succeeded in the long-lasting quest to measure the dielectric constant of confined water at the nanoscale. Our results are important for better understanding of long-range interactions in biological systems, including those responsible for the stability of macromolecules such as DNA and proteins, and of the electric double in electrochemistry, energy storage, etc. The results can also be used to fine tune parameters in future atomistic simulations of confined water. that plays a critical role layer References 1 Israelachvili, J. N. Intermolecular and surface forces (Academic Press, 2011). 2 Leikin, S., Parsegian, V. A., Rau, D. C. & Rand, R. P. Hydration forces. Annu. Rev. Phys. Chem. 44, 369-395 (1993). 3 Honig, B. & Nicholls, A. Classical electrostatics in biology and chemistry. Science 268, 1144-1148 (1995). 4 Stern, O. Zur theorie der elektrolytischen doppelschicht. Elektrochem. Angew. Phys. Chem. 30, 508-516 (1924). 5 Conway, B. E., Bockrais, J. O'M. & Ammar, I. A. The dielectric constant of the solution in the diffuse and Helmholtz double layers at a 4 Fig. 3 Dielectric constant of water under strong confinement. Symbols: ε⊥ for water channels with different h. The y-axis error is the uncertainty in ε⊥ that follows from the analysis such as in Fig. 2h. The x-error bars show the uncertainty in the water thickness including the residual sagging. Red curves: Calculated ε⊥(h) behavior for the model sketched in the inset. It assumes the presence of near-surface layer with εi = 2.1 and thickness hi whereas the rest of the channel contains the ordinary bulk water. Solid curve: Best fit yielding hi = 7.4 Å. The dotted, dashed and dashed-dotted curves are for hi = 3, 6 and 9 Å, respectively. Horizontal lines: Dielectric constants of bulk water (solid) and hBN (dashed). The dielectric constant of water at optical frequencies (square of its refractive index) is shown by the dotted line. The dielectric constant ε⊥ ≈ 2.1 measured for few-layer water is exceptionally small. Not only it is much smaller than that of bulk water (ε ≈ 80) and proton-disordered ice phases such as ordinary ice Ih (ε ≈ 99)26,27 but the value is also smaller than that in low-temperature proton-ordered ices (ε ≈ 3-4)27. Moreover, the found ε⊥ is small even in comparison with the high-frequency dielectric constant ε∞ due to dipolar relaxation (ε∞ ≈ 4-6 for liquid water28,29 and ε∞ ≈ 3.2 for ice Ιh26,27). Nonetheless, ε⊥ ≈ 2.1 lies – as it should – above ε ≈ 1.8 for water at optical frequencies26,29, which is the contribution due to the electronic polarization. The above comparison implies that the dipole rotational contribution is completely suppressed, at least in the direction perpendicular to the atomic planes of the confining channels. This agrees with the MD simulations that find water dipoles to be oriented preferentially parallel to moderately hydrophobic surfaces such as hBN and graphite12-14. The small ε⊥ also suggests that the hydrogen- bond contribution which accounts for the unusually large ε∞ ≈ 4-6 in bulk water28,29 is suppressed, too. The remaining polarizability can be attributed mostly to the electronic contribution the confinement) plus a small contribution from atomic dipoles, similar to the case of non-associated liquids29. change under (not expected to charged interface in aqueous solution. Trans. Faraday Soc. 47, 756- 766 (1951). 6 Hubbard, J. B. & Onsager, L. J. Dielectric dispersion and dielectric friction in electrolyte solutions. I. J. Chem. Phys. 67, 4850-4857 (1977). 7 Chandra, A. Static dielectric constant of aqueous electrolyte solutions: Is there any dynamic contribution? J. Chem. Phys. 113, 903-305 (2000). 8 Zhang, C., Gygi, F. & Galli, G. Strongly anisotropic dielectric relaxation of water at the nanoscale. J. Phys. Chem. Lett. 4, 2477-2481 (2013). 9 Schlaich, A., Knapp, E. W. & Netz, R. R. Water dielectric effects in planar confinement. Phys. Rev. Lett. 117, 048001 (2016). 10 Ben-Yaakov, D., Andelman, D. & Podgornik, R. Dielectric decrement as a source of ion-specific effects. J. Chem. Phys. 134, 074705 (2011). 11 Bonthuis, D. J. & Nets, R. R. Unraveling the combined effects of dielectric and viscosity profiles on surface capacitance, electro-osmotic mobility, and electric surface conductivity. Langmuir 28, 16049-16059 (2012). 12 Lee, C. Y., McCammon, J. A. & Rossky, P. J. The structure of liquid water at an extended hydrophobic surface. J. Chem. Phys. 80, 4448- 4455 (1984). 13 Cicero, G., Grossman, J. C., Schwegler, E., Gygi, & F., Galli, G. Water confined in nanotubes and between graphene sheets: A first principle study. J. Am. Chem. Soc. 130, 1871-1878 (2008). 14 Tocci, G., Joly, L. & Michaelides, A. Friction of water on graphene and hexagonal boron nitride from ab initio methods: very different slippage despite very similar interface structures. Nano Lett. 14, 6872- 6877 (2014). 15 Israelachvili, J. N. & Pashley, R. M. Molecular layering of water at surfaces and origin of repulsive hydration forces. Nature 306, 249-250 (1983). 16 Toney, M. F., Howard, J. N., Richer, J., Borges, G. L., Gordon, J. G. et al. Voltage-dependent ordering of water molecules at an electrode– electrolyte interface. Nature 368, 444-446 (1994). 17 Velasco-Velez, J-J., Wu, C. H., Pascal, T. A. , Wan, L. F., Guo, J. et al. The structure of interfacial water on gold electrodes studied by x-ray absorption spectroscopy. Science 346, 831-834 (2014). 18 Cui, H-B., Takahashi, K., Okano, Y., Kobayashi, H., Wang, Z. et al. Dielectric properties of porous molecular crystals that contain polar molecules. Angew. Chem. Int. Ed. 44, 6508-6512 (2005). 19 Haidar, A. R. & Jonscher, A. K. The dielectric properties of zeolites in variable temperature and humidity. J. Chem. Soc., Faraday Trans. 1, 82, 3535-3551 (1986). 20 Aragones, J. L., MacDowell, L. G. & Vega, C. Dielectric constant of ices and water: a lesson about water interactions. J. Phys. Chem. A, 115, 5745-5758 (2011). 21 Geim, A. K. & Grigorieva, I. V. Van der Waals heterostructures. Nature 499, 419-425 (2013). 22 Radha, B., Esfandiar, A., Wang, F. C., Rooney, A. P., Gopinadhan, K. et al. Molecular transport through capillaries made with atomic-scale precision. Nature 538, 222-225 (2016). 23 Esfandiar, A., Radha, B., Wang, F. C., Yang, Q., Hu, S. et al. Size effect in ion transport through angstrom-scale slits. Science 358, 511- 513 (2017). 24 Fumagalli, L., Esteban-Ferrer, D., Cuervo, A., Carrascosa, J. L. & Gomila, G. Label-free identification of single dielectric nanoparticles and viruses with ultraweak polarization forces. Nat. Mater. 11, 808- 816 (2012). 25 Kim, K. K., Hsu, A., Jia, X., Kim, S. M., Shi, Y. et al. Synthesis and characterization of hexagonal boron nitride film as a dielectric layer for graphene devices. ACS Nano 6, 8583-8590 (2012). 26 Petrenko, V. F. & Whitworth, R. W. Physics of Ice (Oxford University Press, 1999). 27 Johari, G. P. The spectrum of ice. Contemp. Phys. 22, 613-642 (1981). 28 Hill, N. E. Interpretation of the dielectric properties of water. Trans. Faraday Soc. 59, 344-346 (1963). 29 Hill, N. E., Vaughan, W. E., Price, A. H. & Davies, M. Dielectric properties and molecular behaviour (Van Nostrand Reinhold Company Ltd, London, 1969). 5  Supplementary Information S1. Device fabrication We made our devices following fabrication procedures similar to those reported in refs 22,23 . In brief, a free-standing SiN membrane (500 nm in thickness) was made from a commercial Si/SiN wafer and used as a substrate for the van der Waals assembly (see Fig. S1; purple). A rectangular aperture of ≈ 3×25 µm2 in size was then etched in the membrane (Fig. S1). This aperture served later as an inlet to fill the nanochannels with water from a reservoir connected to the back of the wafer (Fig. 1a of the main text). Next, we transferred a large cleaved graphite crystal (thickness of ∼ 10-50 nm) to seal the aperture. Separately, an hBN crystal referred to as spacer was prepared on another substrate and patterned into parallel stripes using e-beam lithography and reactive ion etching. The hBN spacer had thickness h chosen in the range of ∼ 1-300 nm. The stripes were spaced apart by ∼ 200 nm and had widths of 0.5-1.5 µm. The spacer stripes were then transferred onto the bottom graphite and aligned perpendicular to the long-axis of the aperture in the SiN membrane (Fig. S1b). As the next step, reactive ion etching was used again from the back of the Si/SiN wafer to project the aperture onto the hBN-on-graphite assembly. The second hBN crystal referred to as top-hBN was prepared with a thickness of 30-80 nm and transferred on top of the assembly. As a result, we obtained an array of channels with the height h and width ∼ 200 nm. The top hBN crystal sealed the etched opening so that the only path from the back side of the SiN membrane to its top was through the resulting nanochannels. After each transfer, we annealed our assembly in Ar/H2 at 400°C for 3 hours to remove any polymer residue and other contamination. Finally, we made an electrical contact to the bottom graphite using photolithography and e-beam evaporation of Au. Optical images of a representative device with the channel height h ≈ 4 nm are shown in Fig. S1. Fig. S1 Devices for local dielectric imaging of confined water. a Optical micrograph of one of our devices. The top hBN layer is ∼ 45 nm thick and h ≈ 4 nm. The free-standing SiN membrane appears in purple; the Si/SiN wafer in green. The graphite layer is contacted with gold pads to serve as the ground electrode. Scale bar: 10 µm. b Zoom into the central region of (a). The areas with nanochannels are shown by the two dashed rectangles. Regions with the hBN spacers not covered by the top hBN and used to measure h are outlined by black dashes. 6 S2. Filling nanochannels with water It was essential to verify that there was water inside nanochannels probed by our scanning probe approach. In particular, we needed to ensure that individual channels under investigation were neither empty nor contained another material because in principle they could be, for example, blocked by contamination or filled with a polymer residue. Global measurements such as those reported previously22,23, in which a water flow through hundreds of channels was detected, were insufficient for the purpose of our study. Note that we could see the water inside individual channels with h > 100 nm using optical microscopy but water was invisible for h < 20 nm as illustrated by Fig. S2. Here one can clearly see that water filled all the large channels connected to the water inlet, except for one that is probably blocked by contamination (see Fig. S2a). On the contrary, Fig. S2b shows that the optical contrast was insufficient to detect water inside channels with small h or actually even see such small channels. To verify the presence of water in the latter case, we adopted the following strategy. The thickness of the top hBN layer was chosen deliberately in the range of typically 30 to 50 nm, which allowed the hBN cover to sag inside the channels if they were empty (in dry air) as sketched in Fig. 1b of the main text. Upon filling them with water from the backside inlet, the top layer straightened (Fig. 1c of the main text). Accordingly, by monitoring topographic changes in the top hBN position before and after (see Figs 1e-g and Figs 2a-c, respectively), we could ensure that individual channels under investigation were first empty and then filled with water for their dielectric imaging. Importantly, topographic and dielectric AFM images could be acquired one after another without perturbing the experimental setup. Note that if the top hBN sagged completely and touched the bottom graphite or if the channels were blocked by contamination, no straightening of the top hBN occurred. Such channels were obviously excluded from our investigation. This monitoring procedure was working well even for devices with h < 2 nm, which required Å-scale topographical imaging to detect sagging and straightening (Fig. 1j of the main text). For such channels, we typically used a slightly thicker top hBN (50 to 80 nm) to avoid its excessive sagging. We studied more than 40 devices in which the top layer was partially sagged as required for monitoring of the water filling. Our success rate was roughly 50% with the rest of the devices being blocked, most probably because of sagging of a very thin (few nm) part of the top hBN, which could be often found near cleaved edges23. For unblocked channels that allowed water inside, our dielectric Fig. S2 Optical images of our devices after filling them with water. a Thick device (h ≈ 242 nm). Channels with water appear darker than the empty channels that are seen to the right of the image and not connected to the inlet (grey rectangle). b Thin (h ≈ 3 nm) device filled with water. Individual channels cannot be resolved on the micrograph. Scale bars: 10 µm. 7 measurements discussed below were highly reproducible. Note that both topographical and dielectric measurements were always carried out at very low (few %) humidity because the water reservoir attached to the back side of the Si/SiN was completely isolated from the AFM chamber whereas a water flux through the channels themselves was so small22 that it could not possibly change the humidity even locally. S3. Local dielectric imaging Dielectric images of the water-filled channels were obtained by a scanning probe technique24 here referred to as scanning dielectric microscopy. It is based on local electrostatic force detection30,31 by using an atomic force microscope. Images were taken at room temperature and in a dry atmosphere (relative humidity of few %) using a commercial AFM (Nanotec Electronica). After locating a region of interest and taking its topographic image, we scanned the AFM tip at the constant height zscan from the top hBN surface. The dielectric images were acquired at 1 sec per line with an applied ac voltage of typically Vac = 4 V and the frequency ν = 1,000 Hz, unless stated otherwise. We recorded mechanical oscillations of the AFM cantilever induced by the electrostatic force between the tip and the surface at the double frequency (2ν) using a lock-in amplifier. The first derivative of the tip-substrate capacitance 2 where D2ν is the cantilever dC/dz in the out-of-plane direction z is given by dC/dz = D2ν (z)·4k/vac oscillation amplitude at 2ν, and k the spring constant of the cantilever. The expression is valid for frequencies well below the resonance frequency of the cantilever. Images obtained in this mode depend only on the dielectric properties of probed devices, their geometry and the AFM tip geometry. We determined the scan height zscan by recording the tip deflection in the dc mode, and the dC/dz signal was also recorded as a function of the tip-surface distance at image edges, as previously reported24. While the deflection-distance curve allowed us to determine zscan, we used the dC/dz signal to detect any vertical drift and corrected zscan for it. Typically, zscan was larger than 15 nm to avoid short-range interactions. Note that this approach is different from scanning polarization force microscopy32 in that the measured force variations at 2ν are not used as a feedback signal in our case. Instead, we turn off the feedback, retract the tip at height zscan from the surface and scan it in a straight line, which minimizes stray capacitance variations and simplifies data analysis. A representative example of dielectric imaging is shown in Fig. S6 for a device with large channels, in which the bulk-water dielectric behavior was recorded. Before and after taking the dielectric image, we also took dC/dz-approach curves over distances of 0-600 nm from the substrate. These curves were used to calibrate the AFM tip geometry in situ (see Fig. S7 and refs 24,33-36). The approach curves were taken directly above top hBN near the scanned area, after verifying that we recovered the same geometrical parameters as measured above the bottom graphite and gold contacts. We used commercial doped-diamond coated probes (CDT-CONTR, Nanosensors) with spring constants in the range 0.3 - 1.0 N/m, nominal radii of 100 to 200 nm and the cone half-angle of ∼ 30º. S4. Dielectric image analysis 37 software and custom- All topographic, dielectric and other AFM data were analyzed using WSxM made Matlab and Mathcad routines. To extract the dielectric constant of water, we analyzed changes in dC/dz over filled channels as compared to the derivative measured over hBN spacer regions, that is, not the absolute value of dC/dz. For brevity, we below redefine dC/dz as dC/dz = dC(zscan, ε⊥)/dz – dC(zscan, εhΒΝ)/dz, where εhBN is the out-of-plane dielectric constant of hBN ~ 3.5 (ref. 25). We then compared the dC/dz detected over the center of the nanochannel (peak value) with the calculated value 8 using our numerical model discussed in the next chapter. ε⊥ was the only fitting parameter to match the experimental and numerical data. All the necessary geometrical parameters of our samples were experimentally determined using AFM and scanning electron microscopy. We found the geometric parameters of our AFM probes in situ by fitting the experimental dC/dz approach curves with their numerical model as described previously24,33-36. This allowed us to obtain the effective tip radius R and the cone half-angle θ , which are responsible24,36 of the local electrostatic interaction, with a high accuracy of ± 3 nm and ± 0.25º, respectively. The spring constants of our cantilevers were given by the manufacturer but we verified that the use of probes with different spring constants did not affect the extracted dielectric constants, in agreement with the result of ref. 24 (see supplementary information therein). Some dielectric constants ε⊥ that were found experimentally and are summarized in Fig. 3 of the main text have asymmetric error bars. This is a result of the logarithmic-like dependence of the tip- surface capacitance on ε⊥ (see the simulated curves in Fig. 2h and Fig. S4e). The feature is typical for local dielectric measurements (see, e.g., ref. 24) and caused by the use of a sharp tip as the top electrode instead of a planar electrode. The logarithmic dependence is also responsible here for the higher sensitivity of our technique to negative capacitive variations in dC/dz (ε⊥ ≤ 3.5) as compared to large positive changes (ε⊥ > 10), as it can be seen in Figs S4b,d,e. This explains why the error bars in Fig. 3 of the main text are large for thick water (h > 10 nm), despite the dC/dz signal is large in this case. The experimental parameters used in the calculations for the three devices of Figs 1-2 of the main text are the following. Sample dimensions in Fig. 1e and Figs 2a,d: h = 10 nm, top hBN layer thickness H = 51 nm, channel width w = 200 nm, hBN spacer width ws = 800 nm. For Fig. 1f and Figs 2b,e: h = 3.8 nm, H = 46 nm, w = 170 nm, ws = 800 nm. For Fig. 1g and Figs 2c,f: h = 1.4 nm, H = 39 nm, w = 200 nm, ws = 800 nm. The dielectric images in Figs 2d-f (h = 10, 3.8 and 1.4 nm) were measured at scan heights zscan = 30, 25 and 17 nm and with tip radii R = 165, 137 and 101 nm (half-angle θ = 29.0, 31.5 and 30.5º), respectively. Note that the observed suppression in ε⊥ is independent of the scan height and the tip radius. We carefully verified this by taking dielectric images at different scan heights (not shown here but see ref. 24) and with different AFM probes (see Fig. S7). Only the capacitive contrast (dC/dz) changes with zscan and R, increasing for smaller scan heights and larger radii24. S5. Finite-element numerical simulations Three-dimensional finite-element numerical calculations were implemented using COMSOL Multiphysics 5.2a (AC/DC electrostatic module) linked to Matlab. The AFM probe was modelled as a truncated cone with half-angle θ and height Hcone terminated with a tangent hemispherical apex of radius R as shown in Fig. S3a. To reduce computational time, the cone height was reduced to 6 µm (half its nominal value) and the cantilever was modeled as a disk of height Hcantilever = 3 µm and zero length Lcantilever, thus omitting the cantilever length. We have checked that these approximations have no impact on the extracted dielectric constants for the geometry analyzed here36,38 . We simulated the probed heterostructure as three buried nanochannels (Fig. S3a). They were modeled as rectangular parallelepipeds of length l = 2.5 µm, height h, width w, spacing ws (found experimentally as discussed above) and the out-of-plane dielectric constant ε⊥. The water channels were surrounded from above by a dielectric matrix with εhΒΝ = 3.5 of a rectangular shape (length l = 2.5 µm, width W and height H + h). Note that, for our thickest channels (h > 100 nm), we modeled them with trapezoidal rather rectangular cross-sections in order to take into account the ∼ 55º angle of the lateral walls, which appeared during etching of thick hBN spacer crystals by reactive ion plasma. 9 For each device, we numerically solved the Poisson's equation for the specific dimensions of the device and the probe with the nanochannel dielectric constant ε⊥ as the only varying parameter. We calculated the electrostatic force acting on the probe and, therefore, the capacitance first-derivative dC/dz as a function of ε⊥ by integrating the built-in Maxwell stress tensor on the probe surface. To avoid size effects related to the simulation box, we used a cylindrical box with infinite lateral extension at the top and lateral boundaries by using the built-in infinite-element transformation. The boundary conditions were set as follows: applied voltage of 1 V at the tip surface; zero voltage at the bottom electrode; zero charge at the top and side boundaries. We validated these simulations against analytical formulas for thin films as previously reported34,38 . Optimization and numerical noise reduction were carried out to meet the accuracy required here. Note that our simulations involved 3D structures with sizes spanning over more than three orders of magnitude - from the micrometer-sized matrix and probe down to the atomically-thin channels. To this end, a mesh of ~ 106 elements was typically required. An example of the electrostatic potential generated around a representative device is shown in Fig. S3b. Furthermore, we implemented Matlab routines to simulate the tip scanning at a constant height zscan from the top hBN surface as in the experiments. This allowed us to compute dielectric images dC(x,y)/dz where (x,y) is the in-plane tip position. Examples of the calculated images and corresponding profiles along the x-axis are plotted in Fig. S4 for representative devices with ε⊥ = 2 and 80. In addition, we also computed fixed-position "spectroscopic" curves, in which the tip was held fixed over the center of a channel and dC/dz was calculated as a function of ε⊥ with respect to the value computed over the center of the hBN spacer. We used such spectroscopic curves to fit our experimental data and obtain ε⊥, as shown in Fig. 2h using the real data and in Fig. S4e for simulated ones. Fig. S3 Numerical simulations. a Simplified schematics of our 3D model, including the AFM tip and three nanochannels (not to scale). b Example of calculated potential distributions. For clarity, only the potential distribution inside the device is shown. In this case, we used H = 40 nm, h = 10 nm, w = 150 nm, ws = 800 nm; W = 3 µm; εhΒΝ = 3.5 and ε⊥ = 2. AFM tip: R = 100 nm, θ = 25º, Hcone = 6 µm, Hcantilever = 3 µm, Lcantilever = 0 µm, zscan = 20 nm. 10  Fig. S4 Simulated dielectric images. a,c Dielectric constant ε⊥ = 2 for (a) and ε⊥ = 80 for (c). Scan height zscan = 20 nm from the top hBN. b,d Corresponding profiles for three heights zscan = 15, 20 and 25 nm. Relative dC/dz are shown, taken with respect to their values over the hBN spacer. Used parameters: h = 3 nm, H = 40 nm, w = 150 nm, ws = 800 nm, R = 100 nm, θ = 30º. e Simulated dC/dz as functions of ε⊥ with the tip fixed at the channel center. Symbols indicate ε⊥ = 2 and 80 for parameters as in (b) and (d). Note that such "spectroscopic" curves show the contrast inversion at ε⊥ = εhBN = 3.5 as well as decrease in sensitivity with increasing zscan, as expected. S6. Subsurface sensitivity The ability of electrical scanning probe techniques such as electrostatic force microscopy (EFM), Kelvin probe force microscopy (KPFM) and scanning microwave microscopy (SMM) to obtain subsurface images on the nanoscale is widely acknowledged (see, for example, refs 39-42). By exploiting the long-range nature of the electrostatic interaction between the tip and a conductive substrate, these techniques are able to detect nanoscale objects buried inside a dielectric matrix. In particular, non- destructive visualization of conductive objects such as carbon nanotubes embedded in dielectrics of hundreds of nm in thickness has previously been reported using EFM39,40 and KPFM41. Our work uses a similar approach based on electrostatic-force detection, which allows detection of water as thin as 1 nm buried ∼ 100 nm below. The subsurface sensitivity in our case depends on several parameters. It obviously decreases with the thickness H of the top hBN layer and the scan height. Also, the sensitivity increases with the width and the height of the nanochannels and the AFM tip radius. Accordingly, we used AFM tips with large radii (100-200 nm) rather than probes with small few-nm radii as in ref. 24. This was intentional to enhance our sensitivity and reach to the water below our relatively thick (40-80 nm) top hBN. The latter thickness was required to avoid the collapse of our nanochannels (see above). Wider channels with w > 200 nm would increase sensitivity but, unfortunately, were also nonviable because of the same collapse22. 11 S7. Effect of residual sagging After filling water inside the studied nanochannels, they often exhibited small residual sagging (≤ 3 Å) for h < 20 nm, where our thick channel devices (h > 100 nm) usually swelled slightly (by 1-2 nm). We verified that these topographic features had no major impact on our results. Moreover, to achieve highest possible accuracy in our experiments, we corrected our numerical modelling by including this residual sagging/swelling for each individual device. The simulation setup is sketched in the inset of Fig. S5a. As an example, Fig. S5a shows the simulated profiles with and without residual sagging by 3 Å for the device of h = 1.4 nm and ε⊥ = 2 of Fig. 2 of the main text. In either case the resulting dC/dz variation remains clearly negative, and the profile is slightly higher if the sagging is not included in the model (open symbols). Accordingly, Fig. S5b shows the simulated "spectroscopic" curves used to extract the water's dielectric constant for all three devices of Fig. 2 of the main text. Without including the sagging into our model, the resulting curves for dC/dz would go slightly higher (dashed) than those that take into account the sagging and are shown in Fig. 2h (solid). This would lead to a slight underestimate for the dielectric constants of confined water. Instead of the correct values ε⊥ = 15.5, 4.4 and 2.3 (filled symbols), ignoring the sagging effect (s = 3, 1.5 and 3 Å as measured in Fig. 2a-c) could have resulted in ε⊥ = 10.2, 3.8 and 1.65 (open symbols) for h = 10, 3.8 and 1.4 nm, respectively. Note that the relative impact of sagging is practically the same for all the three devices independently of their h. This behavior can be traced back to the logarithmic decrease in the dC/dz signal with increasing ε⊥. On one hand, the impact of any small topography artifact is expected to decrease with increasing h. One the other hand, this is counterbalanced by the larger uncertainty with increasing water's ε⊥ due to the logarithmic sensitivity discussed above. Hence, the effect of the residual sagging turns out to be roughly the same for all the channels. Fig. S5 Sagging effect. a Simulated dC/dz profiles with and without 3 Å-sagging (filled and open symbols, respectively) for the device with h = 1.4 nm in Fig. 2 of the main text and ε⊥ = 2. Parameters: h = 1.1 nm and the sagging depth s = 3 Å (filled symbols); same h and no sagging s = 0 (open); the other parameters are as in Fig. 2. Inset: Sketch of the model to include sagging (not to scale). b Simulated dC/dz as functions of ε⊥ with (solid) and without (dashed curves) taking into account the residual sagging for the three specific devices in Fig. 2 of the main text, s = 3, 1.5 and 3 Å for h = 10, 3.8 and 1.4 nm, respectively, as measured in Figs 2a-c. Device parameters are as in Supplementary Method S4. Open (filled) symbols are the measured dC/dz and their projections onto the ε⊥ axis without (with) including the sagging. 12 Fig. S6 Dielectric imaging of large channels. a Topographic image and b Corresponding profile of a device with h ≈ 242 nm after filling it with water (blue curve). The topography profile of the hBN spacer is shown in cyan. c Corresponding dielectric image and d Its averaged profile. Scale bar: 1 µm. S8. Effect of the tip radius We verified that the measured dielectric constant of confined water (Fig. 3 of the main text) was independent of geometry and dimensions of our AFM probes. To this end, we repeated the dielectric measurements using different probes. Their effective tip radii R were measured in situ before and after each dielectric imaging experiment because values of R are required in our simulations. Examples of the approach curves used to extract R are shown in Fig. S7a for the three specific AFM probes employed in the experiments of Fig. 2 of the main text. Fig. S7b plots the measured R of all the probes used in our experiments to extract the dielectric constants in Fig. 3. In this case, R is plotted against the water thickness h. One can see that the used R are randomly scattered over the expected range 100-200 nm indicated by the manufacturer and there is no correlation with h or other geometrical parameters of our devices. This shows that the observed reduction in water's ε⊥ was independent of the used AFM tips. We also confirmed that for the known tip radii and using hBN crystals as test structures, our approach yielded the correct value of εhBN ≈ 3.5 (not shown). Fig. S7 Impact of the tip radius. a Experimental approach curves (symbols) and their fitting (solid curves) to find tip radii. The data correspond to the experiments of Fig. 2. b Measured R for the AFM probes used in our experiments for all h shown in Fig. 3. The blue-shaded region indicates the nominal range expected for these probes. 13 S9. Intermediate water thickness As described in the main text, the strong suppression of ε observed for water of intermediate thickness can be readily explained by having a thin near-surface layer that is not polarizable in series with normally polarizable water further into the bulk. The experimental data presented in Fig. 3 of the main text were fitted for h > 2 nm using the tri-capacitor model shown in the figure's inset. In the analysis where we used the weighted nonlinear least-squares method, we assumed constant ε⊥ = 2.1 for the interfacial layer (as in our thinnest channels) and ε⊥ = 80 for the bulk water. The best fit yielded the interfacial water thickness hi = 7.4 Å (Fig. 3 and Fig. S8, red solid curve; 95% confidence interval of 7.0–7.8 Å). This estimate agrees well with the widely accepted model of the layered structure of water near surfaces, which extends 2-3 water diameters (∼ 3 Å) into the bulk. This model is also consistent with the minimum ε⊥ ≈ 2.1 found for h < 2 nm. It is instructive to compare the observed dependence ε⊥(h) with that predicted by the above model for other values of the dielectric constant, in particular for those often assumed in the literature. As an example, Fig. S8 shows the calculated curves for εi = 6 (roughly one order of magnitude smaller than in bulk water7-9 and representative of water high-frequency behavior26-29). The resulting curves for the interfacial thickness hi = 3, 6 and 9 Å lie well above our experimental results and do not intersect the value ≈ 3.5 corresponding to hBN's dielectric constant. This illustrates again that it is impossible to explain the obtained dielectric images showing zero and negative contrast without much stronger suppression of ε⊥ than routinely assumed in the literature for interfacial water. Fig. S8 Expected and measured suppression of the dielectric constant in interfacial water. Data shown in red are same as in Fig. 3 of the main text. Blue curves: ε⊥(h) predicted by the same model but using εi = 6 instead of 2.1 and thickness hi = 3, 6 and 9 Å (dotted, dashed and dashed-dotted curves, respectively). 14 Supplementary references: 30. S. V. Kalinin, A. Gruverman, Scanning Probe Microscopy, electrical and electromechanical phenomena at the nanoscale (Springer, New York, 2007). 31. B. D. Terris, J. E. Stern, D. Rugar, H. J. Mamin, Contact electrification using force microscopy. Phys. Rev. Lett. 63, 2669 (1989). 32. J. Hu, X. D. Xiao, D. F. Ogletree, M. Salmeron, Imaging the condensation and evaporation of molecularly thin films of water with nanometer resolution. Science 268, 267-269 (1995). 33. S. Gómez-Moñivas, L. S. Froufe-Pérez, A. Caamaño, J. J. Sáenz, Electrostatic forces between sharp tips and metallic and dielectric samples. Appl. Phys. Lett. 79, 4048 (2001). 34. L. Fumagalli, G. Ferrari, M. Sampietro, G. Gomila, Dielectric-constant measurement of thin insulating films at low frequency by nanoscale capacitance microscopy. Appl. Phys. Lett. 91, 243110 (2007). 35. L. Fumagalli, G. Ferrari, M. Sampietro, G. Gomila, Quantitative nanoscale dielectric microscopy of single-layer supported biomembranes. Nano Lett. 9, 1604-1608 (2009). 36. L. Fumagalli, G. Gramse, D. Esteban-Ferrer, M. A. Edwards, G. Gomila, Quantifying the dielectric constant of thick insulators using electrostatic force microscopy. Appl. Phys. Lett. 96, 183107 (2010). 37. I. Horcas, R. Fernández, J. M. Gómez-Rodriguez, J. Colchero, J. Gómez-Herrero et al., WSxM a software for scanning probe microscopy and a tool for nanotechnology. Rev. Sci. Instrum. 78, 013705 (2007). 38. G. Gomila, G. Gramse, L. Fumagalli, Finite-size effects and analytical modeling of electrostatic force microscopy applied to dielectric films. Nanotechnology 25, 255702 (2014). 39. T. S. Jespersen, J. Nygård, Mapping of individual carbon nanotubes in polymer/nanotube composites using electrostatic force microscopy. Appl. Phys. Lett. 90, 183108 (2007). 40. M. Zhao, X. Gu, S. E. Lowther, C. Park, Y. C. Jean et al., Subsurface characterization of carbon nanotubes in polymer composites via quantitative electric force microscopy. Nanotechnology 21, 225702 (2010). 41. O. A. Castañeda-Uribe, R. Reifenberger, A. Raman, A. Avila, Depth-sensitive subsurface imaging of polymer nanocomposites using second harmonic Kelvin probe force microscopy. ACS Nano 9, 2938–2947 (2015). 42. A. Tselev, J. Velmurugan, A. V. Ievlev, S. V. Kalinin, A. Kolmakov, Seeing through walls at the nanoscale: microwave microscopy of enclosed objects and processes in liquids. ACS Nano 10, 3562–3570 (2016). 15 

1212.3358

1

1212

Stabilization of a linear nanomechanical oscillator to its ultimate thermodynamic limit

The rapid development of micro- and nanooscillators in the past decade has led to the emergence of novel sensors that are opening new frontiers in both applied and fundamental science. The potential of these novel devices is, however, strongly limited by their increased sensitivity to external perturbations. We report a non-invasive optomechanical nano-stabilization technique and apply the method to stabilize a linear nanomechanical beam at its ultimate thermodynamic limit at room temperature. The reported ability to stabilize a mechanical oscillator to the thermodynamic limit can be extended to a variety of systems and increases the sensitivity range of nanosensors in both fundamental and applied studies.

cond-mat.mes-hall

cond-mat

Stabilization of a linear nanomechanical oscillator to its ultimate thermodynamic limit E. Gavartin1, P. Verlot1∗and T.J. Kippenberg1,2† 1Ecole Polytechnique F´ed´erale de Lausanne, EPFL, 1015 Lausanne, Switzerland 2Max-Planck-Institut fur Quantenoptik, Hans-Kopfermann-Strasse 1, 85748 Garching, Germany November 11, 2018 Abstract The rapid development of micro- and nanooscillators in the past decade has led to the emergence of novel sensors that are opening new frontiers in both applied and fundamental science. The potential of these novel devices is, however, strongly limited by their increased sensitivity to external perturbations. We report a non- invasive optomechanical nano-stabilization technique and apply the method to stabilize a linear nanomechanical beam at its ultimate thermodynamic limit at room temperature. The reported ability to stabilize a mechanical oscillator to the thermodynamic limit can be extended to a variety of systems and increases the sensitivity range of nanosensors in both fundamental and applied studies. Thermal noise is well known to be a ubiquitous and fundamental limit for the sensitivity of micro- and nanome- chanical oscillators [1, 2, 3]. This noise arises from the unavoidable coupling of the mechanical device to a thermal bath [4], which is responsible for an incoherent motion with random amplitude and phase [5]. A number of precision measurements based on using nanomechanical resonators, such as gradient force [6, 7], mass [8, 9] and charge [10] detection, or time and frequency control [11], consist in detecting linear changes induced by the measured system on the phase of the coherently driven nanomechanical sensor [1, 12]. These applications are thus theoretically limited by the random phase fluctuations related to thermal motion. Decreasing the effect of thermal noise can be accomplished by increasing the relative contribution of the coherent drive [1]. However, the latter has to be kept below a certain threshold, above which the mechanical oscillator becomes nonlinear [3, 13]: At this threshold, thermal noise constitutes an irreducible sensing limit for linear nanomechanical measurements. In practice, even when driven close to their nonlinear onset, nanomechanical resonators presently operate far away from the thermal limit [8, 14, 15, 16]. This is due to a plethora of external perturbation sources acting on the oscillator [2], such as temperature fluctuations [17], adsorption-desorption noise [18] or molecular diffusion along the oscillator [19]. While several proposals have been made [20, 21] and implemented [22, 23] to decrease excess phase noise, they are based on oscillators being driven into the nonlinear regime and require specific operating conditions, thus making the schemes difficult to adapt for a general class of resonators. In contrast, here we introduce an approach that generally applies to linear nanomechanical resonators. This approach employs an auxiliary mechanical mode as a frequency noise detector, whose output is used to stabilize the frequency of the fundamental out-of-plane mode of a nanomechanical beam. Its experimental implementation enables quasi-optimal linear operation at room tempera- ture, with an almost perfect external phase noise cancellation, as demonstrated for the nanomechanical beam driven just below its nonlinear onset. Hence, our system operates close to its ultimate thermodynamic linear sensing limit. Our scheme can be readily incorporated into a variety of other systems used for applications in [7, 8, 9], frequency control [11] or fundamental studies [24, 25]. Our system consists of an integrated hybrid optomechanical transducer described elsewhere [26]. Briefly, it consists of a 90 µm × 700 nm × 100 nm low dissipation nanomechanical beam made of high-stress silicon nitride (Si3N4) placed in the near-field of a high-Q whispering gallery mode (WGM) confined in a disk-shaped microcavity (cf. Fig. 1(a)). Laser light is coupled in and out of the oscillator with a tapered fiber. Mechanical motion is read out using the optomechanical interaction by placing the laser frequency at mid-transmission of the optical mode, such that the phase fluctuations of light induced by the mechanical motion are directly transformed into amplitude fluctuations readily detected with a photodiode [27]. ∗Corresponding author; Email: [email protected] †Corresponding author; Email: [email protected] 1 Our stabilization scheme is based on the intuition that most of the observed external frequency noise in our system arises from perturbations affecting the entire beam geometry (e.g. temperature fluctuations causing causing changes in length, strain among others). Under such conditions, the induced frequency fluctuations of the various eigenmodes are expected to be highly correlated. One of these modes can thus be used as a "noise detector" whose output serves as an error signal in a feedback mechanism applied to the beam, enabling a cancellation of the geometry variations and thereby a stabilization of all eigenmodes (cf. Fig. 2(a)). In the following our analysis will focus more specifically on two modes, which are the lowest-order out-of-plane mode, which we seek to stabilize, and the lowest-order in-plane mode, which is used as the noise detector. These two modes (referred to as mode 1 and 2, respectively, in the following) have resonance frequencies νM,1 = ΩM,1/2π = 2.84 MHz and νM,2 = ΩM,2/2π = 3.105 MHz as well as linewidths of ΓM,1/2π = 7.5 Hz and ΓM,2/2π = 48 Hz. Note that using the in-plane mode as the noise detector has several advantages: First, the accurate position resolution of the beam's intrinsic motion enables to achieve a large bandwidth (νD (cid:39) 20 × ΓM,2/2π (cid:39) 1 kHz, cf. Fig. 1(d)) over which frequency fluctuations can be detected with thermal limited imprecision. Second, this geometry enables a selective exposure of the perpendicular out-of-plane mode to external signals, thus preserving its sensing capabilities when feedback is turned on. Experimentally, the frequency fluctuations of each mode are detected via its out-of-phase response to an external radiation pressure coherent drive with a frequency of ΩF,i (i ∈ {1, 2}), similar to a phase- locked-loop detection. We choose the driving strengths such that both modes have an amplitude of just below the respective nonlinear thresholds. Both measured out-of-phase quadratures for mode 1 and 2 are shown in Figure 2(d) with the demodulated response plotted versus the detuning ωi/2π = (ΩF,i − ΩM,i)/2π between the driving and the resonance frequency. By putting the external drive close to the resonance frequency (ΩF,i (cid:39) ΩM,i), we can monitor the resonance frequency as small changes on the order of δΩi (cid:46) ΓM,i that are proportional to the out-of-phase quadrature signals. Our feedback mechanism relies on DC-photothermal actuation applied to the beam by sending a third amplitude modulated laser tone (control beam) with input power in the mW range (see Fig. 1(e)). The frequency shift follows linearly the optical input power as shown in Figure 2(b) for mode 1. Figure 2(c) shows the frequency response of modes 1 and 2 to a triangular wave function applied to the control beam. One can already note the similarities of the fluctuations superimposed to the response triangles in both cases, indicating the common origin of the frequency noise over the various eigenmodes. These noise correlations can be quantitatively confirmed by determining the frequency correlations C1−2[ω] = 2 [ω](cid:105)/ (Sν1 [ω]Sν2[ω])1/2 (with δνi[ω] denoting the Fourier transform of the frequency fluctuations of mode (cid:104)δν1[ω]δν∗ i, Sνi its spectrum and (cid:104)...(cid:105) statistical average), which are represented in Figure 3(c) (inset). It should be emphasized that in absence of external frequency noise, the frequency thermal imprecisions over modes 1 and 2 are expected to be uncorrelated: the high level of correlations is therefore a signature of the presence of external frequency noise. Experimentally, this excess of frequency noise is determined by comparing the phase noise spectrum Sφφ,i of mode i to its thermal equivalent imprecision, which in the strong drive limit is given by [2, 5] φφ,i [ω] (cid:39) Sth osc,i(cid:105) × 1 (cid:104)X 2 (cid:16) ΓM,ikBT meff,iΩ2 M,i ω2 + Γ2 M,i/4 (cid:17) , (1) with (cid:104)X 2 osc,i(cid:105) as the variance of the driven amplitude, meff,i as the effective mass, kB as Boltzmann's constant, T as the temperature and ω as the Fourier frequency. Equation 1 simply reflects the fact that phase noise can be interpreted as the "angle" under which the thermal distribution appears from the center of the phase space (see Fig. 3(b)). The time evolution of the frequency drift δνM,i, as determined from the mechanical out-of-phase response, is shown in Figure 3(a) for mode 1 and mode 2. It is clearly evident that the frequency noise is strongly decreased for the stabilized case. These time traces are used to obtain the phase noise spectrum via a Fourier transform. The results are shown in Figure 3(c), both in the non-stabilized (red) and stabilized (blue) cases. Dark colored curves correspond to mode 1 and light colored to mode 2. The results are calibrated using the inferred driven root-mean-square amplitudes (cid:104)X 2 2 = 4.1 nm, just below their respective nonlinear thresholds. When the system is not stabilized, the phase noise of both modes is clearly above the thermal limit. With the stabilization scheme turned on, the phase noise of mode 1 is reduced to the thermal limit in the frequency range of ∼10-50 Hz. The spectrum of mode 2 implies that its phase noise was reduced to even below the thermal limit, which is clearly unphysical. This artifact arises from squashing, which is generally observed in an in-loop measurement where the feedback gain is so large that the reinjected background - in our case the thermally induced frequency fluctuations - is the dominating part of the error signal. As the spectrum of mode 1 is obtained in an out-of-loop measurement, reinjection of background would manifest itself correctly as an increase in the phase noise √ spectrum. Since the thermal frequency noise background Sth Hz) in Hz), the reinjection of noise is limited for mode 1 - the mode of interest - to a comparison to mode 1 (41 mHz/ √ φφ,i is smaller for mode 2 (28 mHz/ ωω,i = ω2Sth osc,1(cid:105) 1 2 = 1.7 nm and (cid:104)X 2 osc,2(cid:105) 1 2 level of 1.5 dB with respect to the thermal limit. We also observe the effect of stabilization on the correlations between the frequency noise of both modes. As noted above, noise suppression is expected to correspond to a significant decrease of the frequency noise corre- lations, as the noises resulting from fundamental thermodynamic fluctuations are uncorrelated. We confirm this experimentally as shown in the inset in Figure 3(c). To access the long-term stability for mode 1, we determine its Allan deviation. In order to avoid measurement artifacts, the mechanical motion is no longer driven using an external frequency reference, but is maintained in self- oscillation using a feedback circuit [28, 26] (see Fig. 1(e)). The readout signal is fed into a frequency counter (Agi- lent 53230), which determines the Allan deviation as (cid:104)δνM,1/νM,1(cid:105)τ = with ¯νM,k as the averaged frequency in the kth discrete time interval over the gate time τ . k=1 ((¯νM,k+1 − ¯νM,k) /νM )2(cid:105)1/2 (cid:80)N 2(N−1) (cid:104) 1 , The obtained results are shown in Figure 3(d), where the red and blue dots represent the values of the fractional Allan deviation of mode 1 as a function of gate time in the non-stabilized and stabilized cases, respectively. The slow fluctuations are significantly reduced in presence of stabilization, with the noise suppression even exceeding 20 dB at the second scale. The achieved long-term stability is further compared to the theoretical limits of our system. For a self-sustained mechanical resonator, the thermo-mechanically limited Allan variance σ1(τ ) = (cid:104)δνth M,1/νM,1(cid:105)τ can be shown to be given by (see Supplementary section S2.2): (cid:32) (cid:104)X 2 th,1(cid:105) osc,1(cid:105) × (cid:104)X 2 1 QM,1ΩM,1τ (cid:33)1/2 σ1(τ ) = , (2) th,1(cid:105)1/2. The data of Figure 3(d) were acquired in presence of a self-sustained displacement (cid:104)X 2 This analysis shows that our stabilization scheme enables reducing external frequency noise close to the thermal limit, but not closer than a factor of 2 (that is 6 dB in power), which disagrees with the spectral data of Figure 3(c). This can be explained by a reduced sensitivity of noise detection: Mode 2 being externally driven, the expression of its thermal limited Allan deviation differs from mode 1 and is described by (see Supplementary section 2.2): osc,1(cid:105)1/2 (cid:39) 100(cid:104)X 2 (cid:32) (cid:104)X 2 σ2(τ ) = th,2(cid:105) osc,2(cid:105) × 3 + e−ΓM,2τ − 4e− ΓM,2τ (cid:104)X 2 (ΩM,2τ )2 2 (cid:33) 1 2 . osc,2(cid:105)1/2 (cid:39) 102(cid:104)X 2 With (cid:104)X 2 (cid:104)δνth limit, which is given by: (3) M,2(cid:105)τ(cid:39)0.1 s (cid:39) 10−8νM,2 is comparable to th,2(cid:105)1/2, the noise detection imprecision (cid:104)δνth M,1(cid:105)τ(cid:39)0.1 s (cid:39) 10−8νM,1 (Eq. 2), and therefore has to be taken into account when determining the stabilization (cid:1)1/2 (cid:104)δνlim (4) M,2(cid:105)τ as well as the overall stabilization limit given by Eq. 4 are represented in Figure 3(d), showing very good agreement with the experimental data up to gate times in the 100 ms range. On longer timescales, the frequency fluctuations are stabilized to a flat level in the 10 ppb range, which we attribute to insufficient gain. M,1(cid:105)τ = (cid:0)(cid:104)δνth The thermal limit for mode 2 (cid:104)δνth τ + (cid:104)δνth M,2(cid:105)2 M,1(cid:105)2 . τ Finally, we analyze the frequency stability and its improvement with the stabilization scheme by using a third independent method based on a recent proposal [29]. Briefly, this method is based on the simultaneous measurement of both in-phase and out-of-phase quadratures X1(t) and X2(t) of the coherently driven mechanical motion. These quadratures are used to reconstruct the complex motion trajectory u(t) = X1(t)+iX2(t) whose moments (cid:104)un(cid:105) can be shown to be insensitive to thermal noise, enabling the extraction and quantitative study of external frequency noise. Thus, the presence of external frequency fluctuations δΩM(t) is revealed by an anomaly of the ratio r = (cid:104)u2(cid:105)/(cid:104)u(cid:105)2, which falls below unity by an amount proportional to the frequency noise power. Indeed, the quadratures of the coherently driven motion write as the sum of a steady and a fluctuating part X1,2(t) = (cid:104)X1,2(cid:105) + δX1,2(t). A straight forward expansion gives (cid:104)u2(cid:105) = (cid:104)u(cid:105)2 + (cid:104)(δX1)2(cid:105) − (cid:104)(δX2)2(cid:105) − 2i(cid:104)δX1δX2(cid:105). With the thermal contribution to the quadrature fluctuations δX th 1,2 being uncorrelated and with identical variances, the previous expression becomes (cid:104)u2(cid:105) = (cid:104)u(cid:105)2 +(cid:104)(δX ext 1,2 denoting the quadratures fluctuations due to the presence of external frequency noise. For a resonantly driven system, the in-phase quadrature X1 is maximum, and (cid:104)u(cid:105)2. δX ext 1 = (∂X1/∂ω) δΩM(t) = 0, (cid:104)u2(cid:105) is thus reduced to the quantity (cid:104)(δX ext )2(cid:105) (cid:39) (∂X2/∂ω)2 (cid:104)δΩ2 )2(cid:105)− 2i(cid:104)δX ext )2(cid:105)−(cid:104)(δX ext M(cid:105) (cid:39) 4(cid:104)δΩ2 M(cid:105) (cid:105), with X ext 1 δX ext 1 2 2 2 Γ2 M 3 At the mechanical resonance frequency, r gives a direct access to the frequency noise power, with no calibration being required. In a more sophisticated approach [29], it can be shown that for a weak Gaussian frequency noise, the ratio r[ω] varies along the driving frequency as r[ω] = 1 − Sωω ΓM + 2iω , (5) with Sωω denoting the frequency noise spectral density. Figure 4 shows the experimental determination of r[ω] we performed with our system, both in the non-stabilized (Fig. 4(c)) and in the stabilized cases (Fig. 4(c,d)). Figures 4(a) and 4(b) show the realizations of the phase-space trajectory used to determined r[ω (cid:39) 0] in the non-stabilized and stabilized configurations respectively. As expected, without stabilization we obtain a discernible dip around the mechanical resonance frequency as shown in Figure 4(c). The dip depth is about 1 − r[ω = 0] (cid:39) 4 × 10−2 with the deviations from the expected shape resulting from long-term frequency drifts that change δω. Following Eq. 5 such a dip is expected to Hz)2, that is about 15 dB above the thermal limit correspond to a frequency noise Sνν = Sωω/(2π)2 = (218 mHz/ νν[ω] = ω2 Hz)2, in very good agreement with the results presented in Fig. 3(c). When Sth the stabilization is switched on, the dip is greatly reduced to a level of 1 − r[ω = 0] (cid:39) 6 × 10−4, corresponding to Hz)2 that is 1.5 dB above the expected thermal limit. This value confirms the a residual noise Sνν = (27 mHz/ origin of this remaining imprecision as resulting from squashing, which is expected to generate the very same noise excess as explained above. √ 4π2 Sφφ,1[ω] (cid:39) (41 mHz/ √ √ In conclusion, we have proposed and implemented a scheme allowing a non-invasive stabilization of a mechanical mode. This technique proves to be remarkably efficient, as shown by the almost perfect noise cancellation for the lowest order out-of-plane mode of a high-Q nanomechanical resonator operated just below its nonlinear threshold, which places our system close to the optimum noise limit at room temperature. Our method is very general and can be readily applied to a number of ultra-sensitive nanomechanical systems: First, our scheme allows using any mechanical mode for detection of frequency noise and we successfully demonstrated noise cancellation using the third-order out-of-plane mode. To apply an error signal one needs a DC restoring force, which is widely available either via photothermal forces in optomechanical systems such as in our work or as electrostatic forces in electrical systems [30, 31, 32]. Importantly, our scheme fully preserves the detection capabilities of the stabilized system, which we have experimentally verified by showing efficient radiation-pressure induced thermomechanical squeezing [33] of the fundamental out-of-plane mode [34] that was impossible to observe without stabilization due to long-term frequency drifts. The fabrication was carried out at the Center of MicroNanotechnology (CMi) at EPFL. Funding for this work was provided by the NCCR Quantum Photonics, the SNF, DARPA Orchid and an ERC. References [1] Albrecht, T. R., Grutter, P., Horne, D. & Rugar, D. Frequency modulation detection using high-q cantilevers for enhanced force microscope sensitivity. J. Appl. Phys. 69, 668 -- 673 (1991). [2] Cleland, A. N. & Roukes, M. L. Noise processes in nanomechanical resonators. J. App. Phys. 92, 2758 -- 2769 (2002). [3] Ekinci, K. L. & Roukes, M. L. Nanoelectromechanical systems. Rev. Sci. Instrum. 76, 061101 (2005). [4] Saulson, P. Thermal noise in mechanical experiments. Physical Review D 42, 2437 (1990). [5] Briant, T., Cohadon, P., Pinard, M. & Heidmann, A. Optical phase-space reconstruction of mirror motion at the attometer level. The European Physical Journal D - Atomic, Molecular, Optical and Plasma Physics 22, 131 -- 140 (2003). [6] Giessibl, F. et al. Atomic resolution of the silicon (111)-(7x7) surface by atomic force microscopy. Science (New York, NY) 267, 68 (1995). [7] Rugar, D., Budakian, R., Mamin, H. & Chui, B. Single spin detection by magnetic resonance force microscopy. Nature 430, 329 -- 332 (2004). 4 [8] Chaste, J. et al. A nanomechanical mass sensor with yoctogram resolution. Nature Nanotech. 7, 300 -- 303 (2012). [9] Jensen, K., Kim, K. & Zettl, A. An atomic-resolution nanomechanical mass sensor. Nature Nanotech. 3, 533 -- 537 (2008). [10] Cleland, A. & Roukes, M. A nanometre-scale mechanical electrometer. Nature 392, 160 -- 162 (1998). [11] Nguyen, C.-C. Mems technology for timing and frequency control. Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on 54, 251 -- 270 (2007). [12] Binnig, G., Quate, C. & Gerber, C. Atomic force microscope. Phys. Rev. Lett. 56, 930 -- 933 (1986). [13] Nayfeh, A. & Mook, D. Nonlinear Oscillations (Wiley, New York, 1979). [14] Feng, X. L., White, C. J., Hajimiri, A. & Roukes, M. L. A self-sustaining ultrahigh-frequency nanoelectrome- chanical oscillator. Nature Nanotech. 3, 342 -- 346 (2008). [15] Lee, J., Shen, W., Payer, K., Burg, T. P. & Manalis, S. R. Toward attogram mass measurements in solution with suspended nanochannel resonators. Nano Lett. 10, 2537 -- 2542 (2010). [16] Fong, K. Y., Pernice, W. H. P. & Tang, H. X. Frequency and phase noise of ultrahigh Q silicon nitride nanomechanical resonators. Phys. Rev. B 85, 161410 (2012). [17] Ekinci, K., Yang, Y. & Roukes, M. Ultimate limits to inertial mass sensing based upon nanoelectromechanical systems. J. Appl. Phys. 95, 2682 -- 2689 (2004). [18] Yang, Y. T., Callegari, C., Feng, X. L. & Roukes, M. L. Surface adsorbate fluctuations and noise in nanoelec- tromechanical systems. Nano Lett. 11, 1753 -- 1759 (2011). [19] Atalaya, J., Isacsson, A. & Dykman, M. I. Diffusion-induced dephasing in nanomechanical resonators. Phys. Rev. B 83, 045419 (2011). [20] Yurke, B., Greywall, D. S., Pargellis, A. N. & Busch, P. A. Theory of amplifier-noise evasion in an oscillator employing a nonlinear resonator. Phys. Rev. A 51, 4211 -- 4229 (1995). [21] Kenig, E. et al. Passive phase noise cancellation scheme. Phys. Rev. Lett. 108, 264102 (2012). [22] Antonio, D., Zanette, D. H. & Lopez, D. Frequency stabilization in nonlinear micromechanical oscillators. Nature Commun. 3 (2012). [23] Villanueva, L. et al. A nanoscale parametric feedback oscillator. Nano Lett. 11, 5054 -- 5059 (2011). [24] Verlot, P., Tavernarakis, A., Briant, T., Cohadon, P. & Heidmann, A. Scheme to probe optomechanical correlations between two optical beams down to the quantum level. Phys. Rev. Lett. 102, 103601 (2009). [25] Arcizet, O. et al. A single nitrogen-vacancy defect coupled to a nanomechanical oscillator. Nature Phys. 7, 879 -- 883 (2011). [26] Gavartin, E., Verlot, P. & Kippenberg, T. A hybrid on-chip optomechanical transducer for ultrasensitive force measurements. Nature Nanotech. 7, 509 -- 514 (2012). [27] Cai, M., Painter, O. & Vahala, K. J. Observation of critical coupling in a fiber taper to a silica-microsphere whispering-gallery mode system. Phys. Rev. Lett. 85, 74 -- 77 (2000). [28] Cohadon, P. F., Heidmann, A. & Pinard, M. Cooling of a mirror by radiation pressure. Phys. Rev. Lett. 83, 3174 -- 3177 (1999). [29] Maizelis, Z. A., Roukes, M. L. & Dykman, M. I. Detecting and characterizing frequency fluctuations of vibrational modes. Phys. Rev. B 84, 144301 (2011). [30] Kozinsky, I., Postma, H., Bargatin, I. & Roukes, M. Tuning nonlinearity, dynamic range, and frequency of nanomechanical resonators. Appl. Phys. Lett. 88, 253101 -- 253101 (2006). [31] Faust, T., Krenn, P., Manus, S., Kotthaus, J. P. & Weig, E. M. Microwave cavity-enhanced transduction for plug and play nanomechanics at room temperature. Nature Commun. 3 (2012). 5 [32] Poggio, M. et al. An off-board quantum point contact as a sensitive detector of cantilever motion. Nature Phys. 4, 635 -- 638 (2008). [33] Rugar, D. & Grutter, P. Mechanical parametric amplification and thermomechanical noise squeezing. Phys. Rev. Lett. 67, 699 -- 702 (1991). [34] Gavartin, E., Verlot, P. & Kippenberg, T. Efficient thermomechanical squeezing of a high-qm nanomechanical oscillator via radiation pressure. In preparation . 6 (a) Optical micrograph Figure 1: The hybrid optonanomechanical system and experimental setup. (b) Magnified view of (a) showing 4 of the 35 hybrid structures present on the chip used in the experiment. showing the details of the hybrid structure (false colors). (c) and (d) Thermal displacement spectra S1/2 xx [Ω] of the fundamental out-of-plane (mode 1) and in-plane (mode 2) modes, respectively. The green curves denote the thermal motion of the mechanical modes. The insets show finite-element simulations of the respective mechanical modes with the red arrows representing the direction of displacement. The gray area in (d) represents the bandwidth ∆νD in which the stabilization scheme enables noise cancellation at the fundamental thermodynamical limit. (e) Experimental setup for the stabilization mechanism, with the readout and actuation of the mechanical oscillator being obtained via optical means with external cavity diode lasers (ECDL). The readout signal from the photodiode (PD) is split into a demodulation circuit for the directly driven noise detector mode 2 and a circuit for mode 1, which is either a directly driven (straight line) or self-driven (gray line) circuit. Two amplitude modulators provide the means to drive the mechanical modes (AM2) and to control the mechanical frequency by changing the optical power coupled into the cavity (AM1). The frequency fluctuations δνM,1 and δνM,2 are recorded either with an oscilloscope, when the system is operated in the direct driving mode or δνM,1 is recorded with a frequency counter when the mode is self-driven. 7 Figure 2: Detecting and stabilizing frequency noise with an optonanomechanical system. (a) Scheme describing the stabilization principle. Phase fluctuations of the in-plane mode (mode 2) are detected by determining its out-of-phase response to a coherent drive and serve as an error signal. By applying a feedback control based on these fluctuations, the out-of-plane mode (mode 1) is stabilized. (b) Resonance frequency shift ∆νM,1 of mode 1 in terms of its mechanical linewidth ΓM,1 versus input optical power. (c) Resonance frequency shifts ∆νM,1 and ∆νM,2 of modes 1 and 2, respectively, for a saw-tooth modulation of the input optical power P . The correlation of the response of both modes is clearly visible. (d) The out-of-phase mechanical response of mode 2 (upper graph) and mode 1 (lower graph) to a coherent drive versus the detuning (ΩF − ΩM)/2π. 8 Figure 3: Frequency stabilization at the ultimate thermodynamic limit. (a) Time evolution of the fre- quency drift δνM for mode 1 (upper graph) and mode 2 (lower graph). The non-stabilized case is given by the red curve and the stabilized case by the grey one. (b) Thermal equivalent phase noise in phase-space: X1 and X2 are the in-phase and out-of-phase components of mechanical motion, respectively, and form the phase-space. In this representation, the phase of mechanical motion identifies to the azimuthal coordinate φ. The phase noise δφ simply corresponds to the angle under which the thermal distribution appears from the center. (c) Frequency noise excess of modes 1 (dark-colored) and 2 (light-colored) over the Fourier frequency components. The red and blue curves show the non-stabilized and stabilized case, respectively. The stabilization scheme reduces the frequency noise by over 10 dB up to 90 Hz and allows to stabilize mode 1 to the thermal limit for a wide bandwidth. Inset: The correlation of frequency response for both modes for the non-stabilized (red) and the stabilized (blue) case. (d) The fractional Allan deviation for the non-stabilized (red) and stabilized (blue) case. The dashed line shows the thermal limit associated with mode 1 for a self-driven oscillator. The dot-dashed line shows the thermal limit for mode 2, which was externally driven. The full line shows the thermal limit for mode 1 in presence of feedback (Eq.4) by taking into account the reinjected imprecision from mode 2. 9 Figure 4: Frequency noise self-characterization. (a) A realization of the phase-space trajectory acquired in the non-stabilized case when driving the nanomechanical beam at the mechanical resonance frequency (ω (cid:39) 0). The acquisition duration is on the order of one minute. The red Gaussian disk represents the thermal distribution of variance (cid:104)X 2 th(cid:105), responsible for an equivalent phase noise δφth.(b) Same as in (a), but for the stabilized case. The remaining phase noise clearly appears as being substantially reduced compared to non-stabilized case. (c) The ratio (cid:104)u2(cid:105)/(cid:104)u(cid:105)2 for mode 1, which signifies the presence of frequency fluctuations due to non-thermal contributions, versus detuning from resonance frequency (red: nonstabilized; blue: stabilized). The red dashed-line corresponds to the model given by Eq. 5 for a Gaussian noise with frequency power spectral density 15 dB above the thermal limit. Stabilization greatly reduces the dip due to excessive non-thermal noise. (d) Detailed view of (cid:104)u2(cid:105)/(cid:104)u(cid:105)2 for the stabilized case. The blue dasehd-line correspond to the model given by Eq. 5 for a Gaussian noise with frequency power spectral density 1.5 dB above the thermal limit. 10 Supplementary Material - Stabilization of a strongly driven nanomechanical oscillator to the thermal limit 1 Phase noise spectrum of an oscillator at thermal equilibrium In this section, we present the calculation of a thermal equivalent phase noise for a mechanical resonator at thermal equilibrium. Both cases of a free running and externally driven oscillator are treated successively. In the following, we assume the oscillator to be well described within the weakly damped harmonic approximation, with mechanical susceptibility χ[Ω] given by: χ[Ω] = 1 M (Ω2 M − Ω2 − iΓMΩ) , (S 1) with M , ΩM and ΓM as the mass, mechanical resonance frequency and damping rate respectively. 1.1 Free-running thermally driven oscillator Figure S1: Phase-space representation of a mechanical state at thermal equilibrium. (a) Phase-space trajectory of a mechanical oscillator left at thermal equilibrium. The trajectory describes in cartesian coordinates (X1(t), X2(t)), or equivalently in polar coordinates (
(t), ϕ(t)). (b) Phase-space trajectory of a mechanical resonator driven at the mechanical resonance. The thermal state is "displaced" of a quantity (cid:104)X 3 osc(cid:105)1/2 along the abscise. We start this calculation with noting that the phase ϕ(t) of a thermal state x(t) can be expressed as a function of its slowly varying in-phase and out-of-phase motion components X1(t) and X2(t) (x(t) = X1(t) cos ΩMt + X2(t) sin ΩMt, ΩM/2π denoting the mechanical resonance frequency) as follows: (cid:18) X2(t) (cid:19) X1(t) ϕ(t) = 2 arctan . (S 2) Here the factor of 2 preceding the arctan function is to allow the phase to vary over the interval [0, 2π]. In order to obtain a suitable expression for deriving the spectral properties of ϕ, we subsequently take the derivative of Eq.S 2, which gives:
(t) ϕ(t) = X1(t)X2(t) − X2(t)X1(t), 1 2 (S 3) with
(t) = X 2 the autocorrelation of the quantity
ϕ can be factorized, which leads to: 2 (t) being proportional to the energy operator. The phase and the energy being independent, 1 (t) + X 2 1 4 (cid:104)
(t)
(t(cid:48))(cid:105)(cid:104) ϕ(t) ϕ(t(cid:48))(cid:105) = (cid:104) X1(t) X1(t(cid:48))(cid:105)(cid:104)X2(t)X2(t(cid:48))(cid:105) + (cid:104) X2(t) X2(t(cid:48))(cid:105)(cid:104)X1(t)X1(t(cid:48))(cid:105) −(cid:104) X1(t)X1(t(cid:48))(cid:105)(cid:104)X2(t) X2(t(cid:48))(cid:105) − (cid:104) X1(t(cid:48))X1(t)(cid:105)(cid:104)X2(t(cid:48)) X2(t)(cid:105). (S 4) 11 In the above equation, (cid:104)...(cid:105) is for statistical average, t and t(cid:48) for two arbitrary moments in time, and the factorization on the right side arises from the well known statistical independence of the motion quadratures X1 and X2. For evident reasons of symmetry, the two terms on the right side of Eq. S 4 on the first line (resp. on the second line) are identical such that it can be written: 1 8 (cid:104)
(t)
(t(cid:48))(cid:105)(cid:104) ϕ(t) ϕ(t(cid:48))(cid:105) = (cid:104) X1(t) X1(t(cid:48))(cid:105)(cid:104)X1(t)X1(t(cid:48))(cid:105) − (cid:104) X1(t)X1(t(cid:48))(cid:105)(cid:104)X1(t) X1(t(cid:48))(cid:105). (S 5) We now determine each factor appearing on the ride side of Eq. S 5. We start with the simpler one (cid:104)X1(t)X1(t(cid:48))(cid:105), which identifies with the autocorrelation function of the quadrature X1. Assuming the stationarity of the problem, we have (cid:104)X1(t)X1(t(cid:48))(cid:105) = (cid:104)X1(0)X1(τ = t(cid:48) − t)(cid:105), and the analytic expression of this autocorrelation function can be obtained via the Fourier transform of the spectrum of X1 SX,1[ω] = χ[ΩM + ω]2ST F [ω] = 4M ΓMkBT is the thermal force spectral density): F [ω] (where ST (cid:104)X1(0)X1(τ )(cid:105) = dωeiωτ SX,1[ω] = ST F (2M ΩM)2ΓM e− ΓM 2 t. (cid:90) +∞ −∞ (cid:18) X1: (S 6) (S 7) 2 t(cid:19) , Similarly, noting that S X,1[ω] = ω2SX,1[ω], one obtains the autocorrelation function of (cid:104) X1(0) X1(τ )(cid:105) = ST F (2M ΩM)2 δ(t) − ΓM 4 e− ΓM ture X1 as the convolution between its impulse response χ(t) =(cid:82) +∞ where δ is for the Dirac delta function. To determine the cross correlations (cid:104) X1X1(cid:105) and (cid:104)X1 X1(cid:105), we write the quadra- 2 t (Θ F δ(t − t(cid:48))) [5]. denoting the Heaviside step function) and an effective thermal force Fth (with (cid:104)Fth(t)Fth(t(cid:48))(cid:105) = ST After a tedious but simple calculation, we obtain: −∞ dωeiωtχ[ΩM + ω] = 1/(2M ΩM)Θ(t)e− ΓM F )2 (cid:104) X1(t)X1(t(cid:48))(cid:105)(cid:104)X1(t) X1(t(cid:48))(cid:105) = − (ST (2M ΩM)2 × e−ΓMt 4 . (S 8) Combining Eqs. S 5, S 6, S 7, and S 8, we finally find: The autocorrelation of
remains to be calculated. To do so, we first note that (cid:104)
(0)
(τ )(cid:105) = 2(cid:0)(cid:104)X 2 (2M ΩM)2 × δ(t)e− ΓM (cid:104)
(t)
(t(cid:48))(cid:105)(cid:104) ϕ(t) ϕ(t(cid:48))(cid:105) = 2 t. 1 8 (ST F )2 The equipartition of the energy gives immediately the second term of the previous parenthesis, (cid:104)X 2 (kBT /M Ω2 ing with expanding it as: 1 (0)(cid:105)2 = 1 , remains to be determined, which we do start- M)2. The first one, that is the autocorrelation of X 2 (S 9) 1 (0)(cid:105)2(cid:1). 1 (0)X 2 1 (τ )(cid:105) + (cid:104)X 2 (cid:90)(cid:90)(cid:90)(cid:90) (cid:104)X 2 1 (0)X 2 1 (τ )(cid:105) = dt1dt2dt3dt4χ(0 − t1)χ(0 − t2)χ(τ − t3)χ(τ − t4)(cid:104)Fth(t1)Fth(t2)Fth(t3)Fth(t4)(cid:105), [−∞,+∞]4 (S 10) (Fth(t1), Fth(t2), Fth(t3), Fth(t4)) being a multivariate Gaussian distribution, Wick's theorem applies, and the sta- tistical average of Eq.S 10 can be rewritten as: (cid:104)Fth(t1)Fth(t2)Fth(t3)Fth(t4)(cid:105) = (cid:104)Fth(t1)Fth(t2)(cid:105)(cid:104)Fth(t3)Fth(t4)(cid:105) + (cid:104)Fth(t1)Fth(t3)(cid:105)(cid:104)Fth(t2)Fth(t4)(cid:105) +(cid:104)Fth(t1)Fth(t4)(cid:105)(cid:104)Fth(t2)Fth(t3)(cid:105). (S 11) The previous equation shows that the fourth moment of the thermal force is a combination of products of its autocorrelation functions (cid:104)Fth(ti)Fth(tj)(cid:105) = ST F [Ω (cid:39) Ωm]δ(ti − tj): (cid:104)Fth(t1)Fth(t2)Fth(t3)Fth(t4)(cid:105) (cid:39) Sth FF[Ωm]2 {δ(t1 − t2)δ(t3 − t4) + δ(t1 − t3)δ(t2 − t4) +δ(t1 − t4)δ(t2 − t3)} . (S 12) 12 Replacing Eq.S 12 in Eq.S 10, one obtains: (cid:104)X 2 1 (0)X 2 1 (τ )(cid:105) = (ST F )2 (cid:40)(cid:18)(cid:90) +∞ −∞ (cid:19)2 (cid:18)(cid:90) +∞ −∞ + 2 dt1χ2(t1) (cid:19)2(cid:41) . dt1χ(t1)χ(τ − t1) (S 13) Using the expression of the impulse response χ(t) given above, one finally obtains for the autocorrelation of
: (cid:104)
(0)
(τ )(cid:105) = 4 (cid:18) kBT (cid:19)2(cid:16) 1 + e−ΓMt(cid:17) M Ω2 M . (S 14) Last, using that S ϕ ϕ[ω] = ω2Sϕϕ[ω], Eqs. S 9 and S 14 lead to the very simple expression for the phase noise: Sϕϕ[ω] = ΓM ω2 . (S 15) The phase noise of a free-running, thermally driven mechanical resonator is hence entirely determined by and proportional to its dissipation rate, which is a well known result. Note however that importantly, this does not depend on the temperature. 1.2 Feedback driven oscillator at thermal equilibrium It is straight to show that the expression S 15 extends to the case of a feedback driven oscillator at thermal equi- librium: Indeed, the effect of linear dissipative feedback is to change the susceptibility to an effective susceptibility χfb[ω] with the linewidth ΓM being replaced by an effective linewidth Γfb which is an affine function of feedback gain. Noting that the thermal force is unchanged by feedback, and assuming that the feedback noise is negligible, the phase noise of the feedback driven resonator is obtained by deriving the above analysis with replacing ΓM by Γfb, which leads to: Sfb (S 16) Thereby, positive dissipative strong feedback (Γeff (cid:28) ΓM) enables considerable reduction of the phase noise of the resonator. While this result is somehow similar to the situation of an externally driven resonator as we will see in the next section, the phase noise decrease is limited to the stability of the external reference, while feedback is self-referenced. The experimental performance of the feedback technique will therefore mostly rely on the ability to design a low-noise feedback circuit. ϕϕ[ω] = Γfb ω2 . 1.3 Externally driven resonator To do so, we start back with Eq. S 2 We now turn to the case of an externally driven mechanical resonator. We further assume that the oscillator is resonantly driven using a coherent amplitude (cid:104)X 2 M)1/2. As a consequence, since X1 has been arbitrarily defined as being the phase reference (see Fig. S1), its (random) thermal component can be neglected osc(cid:105)1/2). The motion phase imprecision resulting from the thermal Xth,1 (X1(t) = (cid:104)X 2 fluctuations Xth,2 of the quadrature X2 is thereby given by: osc(cid:105)1/2 + Xth,1(t) (cid:39) (cid:104)X 2 osc(cid:105)1/2 (cid:29) (kBT /M Ω2 The phase noise associated to an externally driven state at thermal equilibrium therefore follows immediately from Eq. S 17: ϕ(t) (cid:39) Xth,2(t) osc(cid:105)1/2 (cid:104)X 2 , (S 17) Sosc ϕϕ [ω] = Sth XX[ω] (cid:104)X 2 osc(cid:105) , (S 18) which is a well known result [1, 2]: As increasing the external drive amplitude1, the thermal induced phase noise decreases. However, as noted above, the frequency stability limit then relies on that of the external reference. Reaching better stability performance will be thereby preferably possible using the positive feedback technique as announced above. 1We assume the mechanical motion to remain linear. 13 osc(cid:105) = (cid:104)X 2 Figure S2: Phase and frequency stability of a mechanical resonator at thermodynamic equilibrium. (a) Schematic of the displacement spectrum of a thermally-driven oscillator (blue), feedback driven resonator with (cid:104)X 2 th(cid:105), th(cid:105). (b) Corresponding phase noise spectra. (c) Phase and externally driven resonator (brown) with (cid:104)X 2 noise of an externally driven high-Q mechanical oscillator as a function of the Fourier frequency and mechanical damping-rate (normalized to an arbitrary damping unit Γ0). (d) Phase noise of an externally driven high-Q mechanical oscillator as a function of the Fourier frequency and driving power. In both (c) and (d), the phase noise decreases towards darker regions. (c) and (d) show that increasing the external drive or mechanical quality factor lead to similar improvement for the phase noise performance. (e) Allan deviation of a thermally-driven oscillator (blue), feedback driven resonator with (cid:104)X 2 th(cid:105), and externally driven resonator (brown) with th(cid:105). (f) Allan deviation for an externally driven resonator as a function of gate time and mechanical (cid:104)X 2 damping rate (normalized to an arbitrary damping unit Γ0). The Allan deviation decreases towards the darker regions. Note the contours which tend to be parallel at higher damping rates, which reflects the fact that the frequency stability becomes independent of damping rate ΓM at higher gate times. fb(cid:105) = (cid:104)X 2 osc(cid:105) = (cid:104)X 2 fb(cid:105) = (cid:104)X 2 2 Thermodynamic limit of frequency stability In this section, we first establish the mathematical connection between phase noise and frequency stability. We then derive the analytical expression of the Allan deviation corresponding to the three cases treated above. 2.1 Allan variance as a function of phase noise Here we establish the mathematical relationship between the Allan variance and the phase noise. For a given gate time τ , the Allan variance σ(τ ) defines as: σ2(τ ) = = 1 2f 2 M 1 2f 2 M lim n→∞ lim n→∞ (cid:17)2(cid:41) (cid:16) ¯fk − ¯fk−1 (cid:32) (k+1)τ(cid:90) n − 1 1 (cid:40)  1 n − 1 n(cid:88) n(cid:88) k=2 k=2 (cid:33)2  , dtf (t) kτ(cid:90) (k−1)τ (S 19) dtf (t) − 1 τ 1 τ kτ 14  1  1 n − 1 (cid:32) (cid:90)(cid:90) n(cid:88) n(cid:88) k=2 (cid:32) (cid:90)(cid:90) (cid:90)(cid:90) [0,τ ]2 −2 [0,τ ]2 σ2(τ ) = 1 f 2 Mτ 2 [0,τ ]2 (cid:90)(cid:90) dtdt(cid:48)f (t + kτ )f (t(cid:48) + (k − 1)τ ) [0,τ ]2 (cid:33) . dtdt(cid:48)(cid:16)(cid:104)f (t)f (t(cid:48))(cid:105) − (cid:104)f (t)f (t(cid:48) − τ )(cid:105)(cid:17) (cid:90)(cid:90) where fM = ΩM/2π denotes the resonator's mechanical resonance frequency. expanding the squared parenthesis in the above equation, one obtains: (cid:90)(cid:90) (cid:90)(cid:90) σ2(τ ) = 1 Mτ 2 lim n→∞ 2f 2 n − 1 dtdt(cid:48)f (t)f (t(cid:48)) + dtdt(cid:48)f (t)f (t(cid:48)) − 2 dtdt(cid:48)f (t)f (t(cid:48)) k=2 Dk+1,k+1 Dk,k Dk+1,k (S 20) where Dj1,j2 = [(j1 − 1)τ, j1τ ] × [(j2 − 1), j2τ ]. The latter equation can be rewritten using fixed integration's boundaries as follows: σ2(τ ) = 1 Mτ 2 lim n→∞ 2f 2 dtdt(cid:48)f (t + kτ )f (t(cid:48) + kτ ) + dtdt(cid:48)f (t + (k − 1)τ )f (t(cid:48) + (k − 1)τ ) (cid:33) , (S 21) Swapping the statistical and temporal summations and using the ergodic assumption, one obtains: . (S 22) To express the Allan variance in terms of frequency noise Sff [Ω], we use the Wiener-Khintchin theorem which enables expressing the autocorrelation functions featured in the above expression as the inverse Fourier Transform of the spectrum, (cid:104)f (t)f (t(cid:48))(cid:105) = 1 dΩSff [Ω]eiΩ(t−t(cid:48)). The latter equation reads therefore: +∞(cid:82) 2π −∞ +∞(cid:90) dΩSff [Ω]eiΩ(t(cid:48)−t)(cid:16) 1 − e−iΩτ(cid:17) . (S 23) σ2(τ ) = 1 2πf 2 Mτ 2 dtdt(cid:48) [0,τ ]2 −∞ Fubini's theorem enables swapping the time and frequency integration to finally have: σ2(τ ) = 1 πf 2 Mτ 2 dΩ Sff [Ω] Ω2 (cid:16) 1 − e−iΩτ(cid:17)(cid:16) (cid:17) 1 − cos(Ωτ ) . (S 24) (cid:90)(cid:90) +∞(cid:90) −∞ Using that the imaginary part of the latter equation cancels, that the frequency spectrum is a pair function Sff [Ω] = Sff [−Ω], and that the frequency noise Sφφ[Ω] = Sff [Ω]/Ω2, the Allan variance finally reads: (cid:18) 2 ΩMτ (cid:19)2 +∞(cid:90) dΩSϕϕ[Ω] sin4(cid:16) Ωτ (cid:17) 2 0 σ2(τ ) = 2 π . (S 25) Though being similar to what can be found in Cleland et al [?, ?], this expression is a factor of π lower, which is of great importance: as an example, when the first one leads to the conclusion that a system is at the thermal limit, we find that it is actually still a factor > 3 above. 2.2 Frequency stability of a mechanical resonator at thermodynamic equilibrium Using the result given by Eq. S 25, we are now able to determine the frequency stability associated to the cases treated in sections 1.1, 1.2 and section 1.3. 15 Free-running thermally driven oscillator Using Eqs. S 15 and S 25, one obtains for the frequency Allan deviation: σ2(τ ) = 1 QMΩMτ , (S 26) with QM = ΩM/ΓM denoting the mechanical quality factor of the resonator. Eq. S 26 shows that the stability increases with mechanical quality factor, gate time, and frequency: Therefore, besides being of high interest for purposes such as mass sensing [14, 8], high-frequency, high-Q mechanical resonators are extremely useful for absolute stability applications such as time and frequency control [11]. Feedback driven oscillator at thermal equilibrium As suggested by the very similar mathematical expres- sion of the phase noise in the free-running and feedback driven configurations, the expression of the thermal limited frequency stability in the latter case takes the same form as Eq. S 26: σ2(τ ) = 1 QfbΩMτ , (S 27) with Qfb = ΩM/Γfb the effective mechanical quality factor2. We note that Eq. S 27 can be expressed differently, as a function of the feedback driven motion variance (cid:104)X 2 th,fb(cid:105): (cid:104)X 2 th(cid:105) th,fb(cid:105) × (cid:104)X 2 σ2 fb(τ ) = 1 QMΩMτ , (S 28) th(cid:105)/(cid:104)X 2 where we used that Γfb/ΓM = (cid:104)X 2 th,fb(cid:105) [26]. This expression identifies to the one we derived in the next paragraph in the case of an externally driven oscillator at short gate times (τ ≤ 1/ΓM), with the important difference that it applies at all gate times (in the limit τ (cid:29) 1/ΩM). Again, the important difference with external driving is that this stability is no longer limited to the one of the external reference, which is found to be routinely on the order of a few ppb (10−9) at second-scale for quartz oscillators. Reaching higher stability that become comparable to atomic clocks' (i.e. below 10−10) will therefore rather require the use of positive dissipative feedback. Externally driven resonator driven resonator at thermal equilibrium is obtained using Eqs. S 18 and S 25, with Sth In the limit of a coherent drive (cid:104)X 2 osc(cid:105) (cid:29) (cid:104)X 2 th(cid:105), frequency stability of an externally XX[ω] being given by [?]: XX[ω] (cid:39) χ[ΩM + ω]2ST Sth F , (S 29) We note that in MEMS-NEMS literature, this calculation is mostly achieved using a truncated expression of the XX[ω (cid:29) ΓM]). This is probably a phase noise, with the slowest noise component being neglected (Sth technical legacy: MEMS technology first provided high-Q low-frequency oscillators, with very long mechanical coherence times 1/ΓM, easily on the order of a few tens of seconds, along which technical limitations (e.g. thermal drifts, pressure fluctuations etc...) are the prominent sources of frequency noise. However, the recent development of high-frequency NEMS has led to low-noise, fast decay rate oscillators, the former truncation being henceforth obsolete. We thereby compute the integral in Eq. S 25 without further assumption, and find: XX[ω] (cid:39) Sth σ2(τ ) = = osc(cid:105) × 3 + exp(−ΓMτ ) − 4 exp(−ΓMτ /2) (ΩMτ )2 kBT M(cid:104)X 2 M Ω2 (cid:104)X 2 th(cid:105) osc(cid:105) × H(t). (cid:104)X 2 (S 30) Eq.S 30 shows that the apparent frequency stability depends of the inverse ratio between the coherent drive and the thermal variance, along the thermal equivalent phase noise dependance. The function H(t) describes the gate- time dependance, and features essentially 2 regimes. For short gate-times (τ (cid:28) 1/ΓM), we have H(τ ) (cid:39) 1/QMΩMτ , and the Allan variance identifies to the result obtained when neglecting the noise's slow components. In this case, one sees that a high QM is favorable to lower measurement imprecision, along the usual optimization discussions. As already mentioned above, we also note that this asymptotic expression is identical to the effective frequency stability previously determined in the case of the feedback driven resonator, with (cid:104)X 2 osc(cid:105) being replaced by (cid:104)X 2 fb(cid:105), the use of feedback being though preferable as explained before. 2We assume the feedback mechanism to be purely dissipative, such that it does not induce any change in the mechanical resonance frequency. 16 On the other hand, for long gate times, Eq.S 30 leads to H(t) (cid:39) 3/Ω2 Mτ 2: The thermally-limited Allan variance no longer depends on the mechanical quality factor, and decreases with the square power of τ and with the fourth power of the mechanical resonance frequency. This 1/τ 2 convergence behavior can be understood since averaging over gate-time greater than τ (cid:29) 1/ΓM means that two consecutive averaged realizations of the frequency measurement are independent due to the random nature of thermal noise, in contrast to the short gate-time case where these two measurements are related via the mechanical coherence time. This result also pleads for developing high frequency oscillators [8, 14] as highly accurate frequency references, even disregarding their mechanical quality factor, which should be kept high enough only for readout constraints3. 3The mechanical quality factor has to be such that the thermal noise can be resolved, since the frequency measurement imprecision will be due to detection background otherwise 17 

1610.03742

1

1610

Helicoidal Graphene Nanoribbons: Chiraltronics

We present a calculation of the effective geometry-induced quantum potential for the carriers in graphene shaped as a helicoidal nanoribbon. In this geometry the twist of the nanoribbon plays the role of an effective transverse electric field in graphene and this is reminiscent of the Hall effect. However, this effective electric field has a different sign for the two iso-spin states and translates into a mechanism to separate the two chiral species on the opposing rims of the nanoribbon. Iso-spin transitions are expected with the emission or absorption of microwave radiation which could be adjusted to be in the THz region.

cond-mat.mes-hall

cond-mat

Helicoidal Graphene Nanoribbons: Chiraltronics Victor Atanasov Department of Condensed Matter Physics, Sofia University, 5 blvd. J. Bourchier, 1164 Sofia, Bulgaria∗ Theoretical Division and Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, NM 87545 USA† Avadh Saxena We present a calculation of the effective geometry-induced quantum potential for the carriers in graphene shaped as a helicoidal nanoribbon. In this geometry the twist of the nanoribbon plays the role of an effective transverse electric field in graphene and this is reminiscent of the Hall effect. However, this effective electric field has a different sign for the two iso-spin states and translates into a mechanism to separate the two chiral species on the opposing rims of the nanoribbon. Iso-spin transitions are expected with the emission or absorption of microwave radiation which could be adjusted to be in the THz region. PACS numbers: 02.40.-k, 03.65.Pm, 73.22.Pr, 73.43.Cd Introduction. The synergy of geometry, topology and electronic, magnetic or optical properties of materials is a prevalent theme in physics, especially when its manifes- tations are unusual and lead to unexpected effects. Note that helical nanoribbons provide a fertile ground for such effects. Both the helicoid (a minimal surface) and heli- cal nanoribbons are ubiquitous in nature; biomolecules in particular1–4. A helicoid has two chiralities (Fig. 1). Solid state examples include screw dislocations in smectic A liquid crystals5, certain ferroelectric liquid crystals6, recently synthesized graphene nanoribbons7–9, helicoids10 and spirals11,12. Various physical effects such as electromechanics in graphene nanoribbons and spirals including geometric ones can be found in [13–16]. Novel electronic phenomena in graphene nanoribbons are the main focus here. In this context, our goal is to an- swer the following question: what kind of effective quan- tum potential do the carriers experience on a graphene helicoid or a helical nanoribbon due to its geometry (i.e., curvature and twist)? Our main finding is that the twist ω serves as an effective electric field acting on the chiral electrons of graphene with a non-vanishing angular mo- mentum state. This is reminiscent of the quantum Hall effect; only here it is geometrically induced. Further- more, this electric field reverses polarity when the iso- spin (defined below with regard to the two components of a Dirac spinor) is changed leading to a separation of the iso-spin states of the carriers on the opposing rims of the nanoribbon. The helicoid geometry creates a pseudo-electric field and this unexpected result is intriguing in view of the typ- ical effect distortion has on graphene honeycomb lattice, that is to induce a pseudo-magnetic field, which leads to the valley-dependent edge states17. One possible reason for not observing pseudo-magnetic fields here is that the helicoid is a minimal surface (the mean curvature is zero everywhere), that is, it is curved but at the same time minimizes the surface energy, therefore not straining the FIG. 1: Two helicoidal nanoribbons with different chirali- ties: (a) ω > 0 and (b) ω < 0. Vertical axis is along x and the transverse direction ξ is across the nanoribbons. Here ξ0 is the inner radius and D is the outer radius. The two graphene iso-spin states (color coded as red and blue) collect on op- posing rims (separated in space). The respective rims are exchanged when the chirality of the helicoid is reversed. The same exchange takes place when the direction of propagation along the helicoid changes, that is m → −m. underlying lattice. We expect our results to lead to new experiments on graphene nanoribbons and other related Dirac twisted materials where the predicted effect can be verified and explored in the light of spintronics, literally in the case of graphene: "chiraltronics" ([18] and references therein). Note that we treat the nanoribbon as a continuum object without taking into account any discreteness of the underlying honeycomb lattice, i.e, we consider a Dirac equation rather than a tight-binding model. Thus are discussion is independent of whether the underlying graphene lattice is parallel or perpendicular to the chi- ral axis, keeping in view the experimental observations of Ref. 10. We also assume that the helicoid remains a 6 1 0 2 t c O 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 4 7 3 0 . 0 1 6 1 : v i X r a minimal surface without any distortion or strain. More- over, we assume the stability of the helicoid geometry and do not consider any instability issues that may arise experimentally. Helicoid geometry. To elaborate on the geometry of the helicoidal graphene nanoribbon we consider a strip whose inner and outer edges follow a helix around the x-axis (see Fig. 1 with ξ0 = 0). The represented surface is a helicoid and is described by the following equation: (cid:126)r = x (cid:126)ex + ξ [cos(ωx) (cid:126)ey + sin(ωx) (cid:126)ez], (1) 2 where ω = 2πn L , L is the total length of the strip and n is the number of 2π-twists. Here ((cid:126)ex, (cid:126)ey, (cid:126)ez) is the usual orthonormal triad in R3 and ξ ∈ [0, D], where D is the width of the strip. Let d(cid:126)r be the line element and the metric is encoded in 2dξ2, 1dx2 + h2 d(cid:126)r2 = (1 + ω2ξ2)dx2 + dξ2 = h2 where h1 = h1(ξ) =(cid:112)1 + ω2ξ2 and h2 = 1 are the Lam´e coefficients of the induced metric (from R3) on the strip. Next, we add a comment on the helicoidal nanoribbon, that is a strip defined for ξ ∈ [ξ0, D] (see Fig. 1). All the conclusions still hold true and all of the results can be translated using the change of variables ξ = ξ0 + s(D − ξ0), s ∈ [0, 1]. Here s is a dimensionless variable and one easily sees that for ξ0 → 0 we again obtain the helicoid. Effective geometric potential. In order to answer the question posed above, here we study the helicoidal sur- face to gain a broader understanding of the interaction between Dirac particles and curvature and the resulting possible physical effects. The properties of free electrons on this geometry have been considered before19–21 in the case of Schrodinger materials. The results of this paper are based on the Dirac equation for a confined quantum particle on a sub-manifold of R3. Following Refs. [22– 24] an effective potential appears in the two dimensional Dirac equation which in this geometry has the following form: −k+ ikx√ 1+ω2ξ2 + i∂ξ ikx√ − i∂ξ 1+ω2ξ2 −k− (cid:18) χA χB (cid:19) k± = ±E/vF , (3) where kx is the partial momentum in x-direction. For more information on the derivation, refer to the Ap- pendix as well as Ref. [25]. Let us consider here the azimuthal angle around the x axis: ωx and the angular momentum along this axis (cylindrical symmetry): Lx = − i ω ∂ ∂x . (4) This operator has the same eigenfunctions Lxφ(x) = mφ(x) as the Hamiltonian since they commute. The ξextr = 1 ω FIG. 2: The potential acting on each of the iso-spin states as a function of the width of the helicoid ξ. Here ω > 0. Note that the potentials have a maximum and then fall off ∝ 1/ξ2. The extremum for m = 1 state is reached for √ 8). For ξ (cid:29) ξextr the iso-spin separation scales ξextr = 1/(ω as ∆U (ξ (cid:29) ξextr) ≈ 2m ξ2 . corresponding eigenvalues are m. We conclude that the momentum kx is quantized kx = mω, m ∈ Z. (5) This is not surprising because of the periodicity of the wave function along x. Note that the value of the angu- lar momentum quantum number determines the direction the carriers take along the x axis either upward m > 0 or downward m < 0. This situation is reversed for a helicoid with opposite chirality (Fig. 1). Now we obtain for the first and second components of the spinor, that is the iso-spin states, the following governing effective Schrodinger equations: − ∂2 −∂2 ξ χA + UA(x)χA = −k2 ξ χB + UB(x)χB = −k2 ξ χA, ξ χB, ξ = k+k− = −E2/(vF )2, k2 (6) (7) (8) k2 x 1 + ω2ξ2 + 1 + ω2ξ2 − k2 x UA = W 2 m − W (cid:48) m = kxω2 (1 + ω2ξ2)3/2 ξ, (9) kxω2 m + W (cid:48) UB = W 2 m = Here Wm = kx/(cid:112)1 + ω2ξ2. These potentials are pseudo- (1 + ω2ξ2)3/2 ξ. (10) binding and are depicted in Fig. 2. Note the qualitative behavior after the extremal point is reached for (cid:113) 1 + m2 −(cid:112)m4 − 3m2 (cid:112)2(1 − m2) , (11) = 0,(2) where the corresponding potentials are provided m (cid:54)= 1. In the case m = 1 the extremum is reached for ξextr = 1/(ω √ 8), that is W < L/(4 Suppose the width of the nanoribbon W is smaller than 1/(ω 2πn), then we can approxi- mate the potential and restrict the expansion to the first order terms 8). √ √ UA ≈ k2 x + kxω2ξ, UB ≈ k2 x − kxω2ξ, then the governing effective equations become ξ χA +(cid:0)k2 + kxω2ξ(cid:1) χA = 0, ξ χB +(cid:0)k2 − kxω2ξ(cid:1) χB = 0, − ∂2 −∂2 k2 x + k2 ξ = k2. (12) (13) (14) (15) Note that the geometry induced potential acting on the two different iso-spin states is similar to the application of a constant electric field E, thus reminiscent of the Hall effect: UA ∝ eEξ, UB ∝ −eEξ, (16) where E = kxω2/e, with its sign being different for the different chiral states. Here e is the electron charge. Therefore, E separates them on the opposing rims of the helicoidal nanoribbon. It is exactly this observation that motivates us to assume a mechanism of separation of chi- ral states in graphene as the basis for a potential new branch of spintronics, namely chiraltronics. k2 x kxω2 These potentials are a sum of two contributions, an almost constant repulsive part (which pushes the carriers to the outer rim): 1+ω2ξ2 ≈ m2ω2 and a variable part (1+ω2ξ2)3/2 ξ ≈ ω3mξ which is repulsive or attractive as a function of the angular momentum quantum number m but more importantly, given m ≥ 0 attractive for iso-spin A (collects on the inner edge) and repulsive for iso-spin B (collects on the outer edge), see (12). The action of the first part ∝ m2ω2 qualifies it as a cen- trifugal potential. It pushes a particle to the boundary of the strip. Physically, one may understand the behav- ior described above using the uncertainty principle: for greater ξ a particle on the strip will have more available space along the corresponding helix and therefore the cor- responding momentum (energy) will be smaller than for a particle closer to the central axis. Since the behavior of the variable part of the poten- tial UB(ξ) for a particle with m ≥ 0 [UA(ξ) for m ≤ 0] qualifies it as a quantum anti-centrifugal one, it con- centrates the corresponding iso-spin carriers around the central axis for a helicoid (or the inner rim for a heli- coidal nanoribbon). Such anti-centrifugal quantum po- tentials have been considered for Schrodinger materials previously26. We note that the separability of the quantum dynamics along x and ξ directions with different potentials points to the existence of an effective mass anisotropy for the chiral electrons on the graphene helicoidal surface. 3 FIG. 3: Provided the nanoribbon is small enough, so that ξ < ξextr, the potential acting on each of the iso-spin states as a function of the width of the helicoid scales linearly with ξ. Note that the difference between the potentials acting on the two iso-spin states is ∆U (ξ < ξextr) ≈ 2mω3ξ. The frequency of the expected transition is in the THz region (for micron-sized ribbons). See text for further details. √ Experimental implications. A number of experimental consequences can be expected. We begin with the "thin strip" case, literally the case in which the width W < 2πn). The pseudo-binding potential (see Fig. 3) L/(4 would lead to a two-metastable-states problem and an os- cillation between the iso-spin states should be expected. The helicoidal graphene nanoribbon should exhibit an absorption line at frequency ν ≈ vF connected with the change (positive chirality helicoid ω > 0) of iso-spin from B to A. Using the restriction on the width of the nanoribbon the frequency turns out to be (cid:113)mn32πW/L3 (cid:115) m √ 2 2 ν ≈ n vF L , (17) which is determined by the geometric and material prop- erties only. In an attempt to evaluate its order of magni- tude we put L ≈ 10−6 m (∼micron) and vF ≈ 106 m/s to obtain ν ≈ 1012 Hz well into the THz region. The reverse process is also possible, that is emission in the THz. The change of iso-spin is in this case from A to B. Therefore we might expect a continuous emission, provided we feed the positive chirality helicoid with a current in the inner rim and extract the current (drain it) from the outer rim on the other end. The iso-spin current has to change and therefore emit THz radiation via a standard QED vertex. See the plot of the potential in Fig. 3. Another experimental effect stems from the form of the geometric potential along the width ξ of the helicoid. The potential in (2) is V = ikxσ1/(cid:112)1 + ω2ξ2. Here we follow the formalism in Ref. [27]. The matrix element of this potential in the Born approximation gives non- are defined in terms of the metric on the strip 0 0 0 −(1 + ω2ξ2) 0 −1 0 0 gµν = (22) Note, ηab = ηab = diag(1,−1,−1) is the choice of the Minkowski metric. Now we define the trei-bein fields eµ a : F  v2  ,  1 0 0 vF 0 0 0 0 0 0 et a =  .  ,  0 0 0  0 0 0 0 0 −1 0 1√ 1+ω2ξ2 0 0 0 0 eξ a =  0 0 − 0 ex a = 4 (23) (24) and eµa = gµνeν a γa matrices algebra fulfills γaγb + γbγa = 2ηabI and trγa = 0. Upon a straightfor- ward check, the following choice is found to be correct a. The γµ = eµ γt = σ3, γx = iσ1, γξ = iσ2, (25) where σj are the Pauli spin matrices. The curvilinear γµ's (19) then are γt = 1 vF σ3, γx = − iσ1(cid:112)1 + ω2ξ2 γξ = −iσ2. , (26) x = ω2ξ 1+ω2ξ2 and Γxx x = The non-zero Christoffel symbols components are Γxξ ξ = −ω2ξ. As a result, the spin Γξx connection Γµ can be computed from (20) which turns out to be vanishing: Γt = 0, Γx = 0 and Γξ = 0. Putting the corresponding terms in the Dirac equation (18) and looking for stationary states with energy E, Ψ = e− i Etψ, we obtain (cid:32) vF(cid:112)1 + ω2ξ2 (cid:33) σ1∂x + vF σ2∂ξ ψ = Eσ3ψ(x, ξ). (27) vanishing probability w(θ) ∝ sin2(θ/2), where θ is the scattering angle, for backward scattering. We conclude that the conductivity of the nanoribbon along the width, that is along the rim-to-rim channel is hindered. We believe, this is an additional confirmation of the iso-spin transition the carriers necessarily undertake to populate the opposing rim. Conclusion. Our main findings can be summarized as follows: the twist ω pushes the graphene carriers with iso-spin A and m ≥ 0 (m ≤ 0) towards the outer (inner) edge of the nanoribbon, respectively iso-spin B for m ≥ 0 (m ≤ 0) towards the inner (outer) edge of the nanorib- bon, and effectively separates chiral species on the oppos- ing rims of the helicoid and induces transitions at THz frequencies. These results are quite distinct from the ones in the case of twisted Schrodinger materials with a scalar wave function and a different geometry induced effective potential21. We also predicted an effective mass anisotropy for chiral electrons on the helicoid. We expect our results to motivate new low temperature experiments (in order to restrict to low m, that is non-dominant ac- tion of the repulsive part of the potential) on twisted graphene nanoribbons in light of the emerging opportu- nity to separate chiral states, explore chiraltronic appli- cations and possibly create new microwave devices. If the helicoid were elastically deformable then the coupling of chiral electrons to the strain field would possibly lead to a pseudo-magnetic field (in addition to a pseudo-electric field) among other interesting effects. In our analysis we have neglected any effects that may arise due to the underlying lattice discreteness and dis- tortion (strain) in a real graphene helicoidal nanoribbon. It would be worthwhile to study these effects numerically along with the potential stability of the considered geom- etry including the effects of van der Waals adhesion, etc. This work was supported in part by the U.S. Depart- ment of Energy. Appendix. The covariant approach for writing the Dirac equation on the curved surface of graphene is the following (cid:16) (cid:17) ivF γµ Dµ Ψ = 0, (18) where the curvilinear matrices are γµ = eµ a γa and Dµ = ∂µ − Γµ. Here (cid:0)∂µeν b + Γµλ (cid:1) γaγb νeλ b Γµ = 1 4 eνa (19) (20) is the spin connection. The Christoffel symbols are de- fined as: Γµλ trei-bein fields28 2 (∂µgλξ + ∂λgµξ − ∂ξgµλ) gξν. The ν = 1 (cid:19) (cid:18) ψA ψB The equations for the iso-spin components after the ansatz ψ(x, ξ) = , ψA,B(x, ξ) = eikx1,x2 x χA,B(ξ) (28) are(cid:18) where Wm(ξ) = kx/(cid:112)1 + ω2ξ2 with the additional con- (cid:19)(cid:18) χA −i∂ξ − iWm(ξ) i∂ξ − iWm(ξ) = 0, (29) (cid:19) k− χB k+ gµνeµ a eν b = ηab, ηabeµ a eν b = gµν (21) dition kx1 = kx2 = kx. 5 ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 C. W. G. Fishwick, A. J. Beevers, L. M. Carrick, C. D. Whitehouse, A. Aggeli, and N. Boden, Nano Lett. 3, 1475 (2003). 2 J. Crusats, J. Claret, I. Diez-Perez, Z. El-Hachemi, H. Garcia-Ortega, R. Rubires, F. Sagues, and J. M. Ribo, Chem. Commun. 13, 1588 (2003). DOI: 10.1039/b303273f 3 O-Y. Zhong-can and L. Ji-xing, Phys. Rev. Lett. 65, 1679 (1990); Phys. Rev. A 43, 6826 (1991). 4 J. M. Garcia Ruiz, A. Carnerup, A. G. Christy, N. J. Wel- Galvo, and F. Sato, J. Chem. Phys. 128, 164719 (2008). 15 Z. Qiao, S. A. Yang, B. Wang, Y. Yao, and Q. Niu, Phys. Rev. B 84, 035431 (2011). 16 J. Klinovaja, G. J. Ferreira, and D. Loss, Phys. Rev. B 86, 235416 (2012). 17 D.-B. Zhang, G. Seifert, and K. Chang, Phys. Rev. Lett. 112, 096805 (2014). 18 W. Han, R. K. Kawakami, M. Gmitra, and J. Fabian, Na- ture Nanotech. 9, 794 (2014). 19 R.C.T. da Costa, Phys. Rev. A 23, 1982 (1981). 20 R. Dandoloff and T.T. Truong, Phys.Lett. A 325, 233 ham, and S. T. Hyde, Astrobiology 2, 353 (2002). (2004). 5 R. D. Kamien and T. C. Lubensky, Phys. Rev. Lett. 82, 21 V. Atanasov, R. Dandoloff, and A. Saxena, Phys. Rev. B 2892 (1999). 6 D. M. Walba, E. Korblova, R. Shao, J. E. Maclennan, D. R. Link, M. A. Glaser, and N. A. Clark, Science 288, 2181 (2000). 7 N. Mohanty, D. Moore, Z. Xu, T. S. Sreeprasad, A. Na- garaja, A. A. Rodriguez, and V. Berry, Nature Commun. 3, 844 (2012). 8 D. V. Kosynkin, A. L. Higginbotham, A. Sinitskii, J. R. Lomeda, A. Dimiev, B. K. Price, and J. M. Tour, Nature 458, 872 (2009); 9 L. Jiao, L. Zhang, X. Wang, G. Diankov, and H. Dai, Na- ture 458, 877 (2009). 10 T.W. Chamberlain, J. Biskupek, G.A. Rance, A. Chuvilin, T.J. Alexander, E. Bichoutskaia, U. Kaiser, and A. N. Khlobystov, ACS Nano 6, 3943 (2012). 11 X. Zhang and M. Zhao, Scientific Reports 4, No. 5699 (2014). DOI: 10.1038/srep05699 12 S. M. Avdoshenko, P. Koskinen, H. Sevincli, A. A. Popov, and C. G. Rocha, Scientific Reports 3, No. 1632 (2013). DOI: 10.1038/srep01632 13 P. Koskinen, Appl. Phys. Lett. 99, 013105 (2011); T. Ko- rhonen and P. Koskinen, AIP Advances 4, 127125 (2014). 14 E. W. S. Caetano, V. N. Freire, S. G. dos Santos, D. S. 79, 033404 (2009). 22 M. Burgess and B. Jensen, Phys. Rev. A 48, 1861 (1993). 23 V. Atanasov and A. Saxena, Phys. Rev. B 81, 205409 (2010); Y. N. Joglekar and A. Saxena, Phys. Rev. B 80, 153405 (2009). 24 V. Atanasov and A. Saxena, J. Phys.: Condens. Matter 23, 175301 (2011). 25 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature 438, 197 (2005); Maria A. H. Vozmediano, Nature Phys. 7, 671 (2011); F. Guinea, M.A.H. Vozmedi- ano, M.P. Lopez-Sancho, and J. Gonzalez, Adv. Mater., 23 5324 (2011); P. San-Jose, J. Gonzalez, and F. Guinea Phys. Rev. Lett. 106, 045502 (2011). 26 M. A. Cirone, K. Rzazewski , W. P. Schleich , F. Straub, and J. A. Wheeler, Phys. Rev. A 65, 022101 (2001); V. Atanasov and R. Dandoloff, Phys. Lett. A 371, 118 (2007). 27 T. Ando, T. Nakanishi and R. Saito, J. Phys. Soc. Jpn. 67, 2857 (1998). 28 D. J. Struik, Lectures on Classical Differential Geometry, Second Ed. (Dover, Reading, MA, 1988). 

1710.01486

1

1710

Imaging resonant dissipation from individual atomic defects in graphene

Conversion of electric current into heat involves microscopic processes that operate on nanometer length-scales and release minute amounts of power. While central to our understanding of the electrical properties of materials, individual mediators of energy dissipation have so far eluded direct observation. Using scanning nano-thermometry with sub-micro K sensitivity we visualize and control phonon emission from individual atomic defects in graphene. The inferred electron-phonon 'cooling power spectrum' exhibits sharp peaks when the Fermi level comes into resonance with electronic quasi-bound states at such defects, a hitherto uncharted process. Rare in the bulk but abundant at graphene's edges, switchable atomic-scale phonon emitters define the dominant dissipation mechanism. Our work offers new insights for addressing key materials challenges in modern electronics and engineering dissipation at the nanoscale.

cond-mat.mes-hall

cond-mat

Title: Imaging resonant dissipation from individual atomic defects in graphene Authors: Dorri Halbertal1*, Moshe Ben Shalom2*, Aviram Uri1, Kousik Bagani1, Alexander Y. Meltzer1, Ido Marcus1, Yuri Myasoedov1, John Birkbeck2, Leonid S. Levitov3, Andre K. Geim2, Eli Zeldov1* Affiliations: 1Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot 7610001, Israel. 2National Graphene Institute, The University of Manchester, Booth St. E, Manchester M13 9PL, UK and the School of Physics and Astronomy, The University of Manchester, Manchester M13 9PL, UK. 3Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. *Correspondence to: [email protected] (D.H.); [email protected] (M.B.); [email protected] (E.Z.). Abstract Conversion of electric current into heat involves microscopic processes that operate on nanometer length- scales and release minute amounts of power. While central to our understanding of the electrical properties of materials, individual mediators of energy dissipation have so far eluded direct observation. Using scanning nano-thermometry with sub-µK sensitivity we visualize and control phonon emission from individual atomic defects in graphene. The inferred electron-phonon "cooling power spectrum" exhibits sharp peaks when the Fermi level comes into resonance with electronic quasi-bound states at such defects, a hitherto uncharted process. Rare in the bulk but abundant at graphene's edges, switchable atomic-scale phonon emitters define the dominant dissipation mechanism. Our work offers new insights for addressing key materials challenges in modern electronics and engineering dissipation at the nanoscale. One sentence summary: We visualize and control conversion of electric current into local heat at individual atomic defects and identify the dominant dissipation pathway in clean graphene. 1  Understanding microscopic mechanisms of momentum and energy dissipation is a central problem in the fields ranging from condensed matter to particle physics. It is also of keen interest for designing new approaches to handle, convert, and utilize energy in a bid to address key technological challenges. Dissipation pathways are particularly intriguing in ultra-pure materials of current interest, such as graphene (1), because of tight restrictions on the phase space for electron-phonon scattering (2–4). Furthermore, unlike particle physics where a particular decay process of interest can be "staged" in a collider, in condensed matter physics we are interested in the processes concealed microscopically within the material. This, along with the minute power released in such processes, poses a key challenge for experimentally probing dissipation at the nanoscale. Here we employ a recently-developed ultrasensitive scanning nanothermometer with sub-µK sensitivity (5), achieved with a superconducting quantum interference device placed on an extremely sharp tip-SQUID-on-tip-to probe these subtle effects in high-mobility graphene. Owing to its exceptional cleanness and the two-dimensional nature of its electrons and phonons (1), graphene is an excellent platform to study electron-phonon relaxation. Our measurements were performed on exfoliated graphene encapsulated between hexagonal Boron Nitride (hBN) layers and designed to inject electrons from narrow constrictions into a micron-scale "electron chamber" (Figs. 1A,B and Section 1 (6)). Transport measurements in such samples routinely show ballistic signatures over a wide range of temperatures and carrier densities (7–9). Our SQUID-on-tip (10) acts as an extremely sensitive thermometer (tSOT) (5) with an effective diameter of 33 nm and thermal sensitivity of 510 nK/Hz1/2 at 4.2 K, and provides an image of the local temperature variations 𝛿𝑇(𝑥, 𝑦) upon scanning at a height ℎ of 10 to 40 nm above the sample surface (as specified). The tip is mounted on a quartz tuning fork (11) which allows the tSOT to vibrate parallel to the sample surface with a controlled amplitude 𝑥𝑎𝑐~ 2.7 nm at a frequency of 37 kHz (Sections 2 and 3 (6)). The resulting ac signal, 𝑇𝑎𝑐(𝑥, 𝑦) = 𝑥𝑎𝑐𝜕𝛿𝑇(𝑥, 𝑦)/𝜕𝑥, renders higher sensitivity imaging (Fig. 1C) by avoiding the low-frequency 1/𝑓 noise of the tSOT. We control the carrier density in graphene globally by a back-gate bias 𝑉𝑏𝑔 on the Si/SiO2 and locally by applying 𝑉𝑡𝑔 potential to the tSOT (Fig. 1A). Description of sample fabrication and thermal imaging is provided in Methods (6). Figure 1C shows the thermal signal 𝑇𝑎𝑐(𝑥, 𝑦) measured while applying a fixed current 𝐼𝑑𝑐 = 3 µA through two of the constrictions as shown in Figure 1B. The image reveals a complex array of fine rings along the edges of the heterostructure (5). In addition, three isolated rings are observed in the bulk of graphene labeled 'A', 'B', and 'C'. The bulk rings are rare and have comparable diameters, in sharp contrast to the rings at the edges which are dense and display widely varying sizes (Movies S1-2 (6)). We show below that the rings mark dissipation from single atomic defects positioned at their centers. Electron-phonon cooling pathways in graphene are uniquely interesting for several reasons. Owing to the exceptional stiffness of the carbon bonds, scattering by optical phonons in graphene is inefficient below room temperature. Moreover, the small size of the electron's Fermi surface restricts the phase volume for scattering by acoustic phonons, blocking the intrinsic electron-lattice relaxation pathway (2–4). However, theory predict that disorder can significantly ease the electron-phonon scattering (4). Our key finding is that hot electrons, generated by the applied current, dissipate their energy through a very specific disorder- induced mechanism: resonant inelastic scattering by local electronic resonances due to individual defects. 2  Each such localized state (LS) mediates cooling through resonant electron scattering, creating an atomic- scale thermal link between the electronic bath, at an effective hot-electron temperature 𝑇𝑒, and the phonon bath, at a base temperature 𝑇𝑝 (Section 4 (6)). To characterize this novel thermal link we define the "electron-phonon heat conductivity of a defect" 𝜅𝑒𝑝(𝜀) which describes the power transferred between the baths 𝑃𝑒𝑝(𝜀) = 𝜅𝑒𝑝(𝜀)Δ𝑇, where Δ𝑇 = 𝑇𝑒 − 𝑇𝑝 and 𝜀 is the energy relative to the Dirac point. The resulting cooling power spectrum 𝑃𝑒𝑝(𝜀) (CPS)-the fundamental quantity accessed in our experiment through local temperature increase 𝛿𝑇(𝜀) ∝ 𝑃𝑒𝑝(𝜀)-is found to peak sharply when the Fermi level 𝜀𝐹 is aligned with the quasi-bound LS resonant energy 𝜀𝐿𝑆. Our analysis (Section 4 (6)) of the observed resonances suggests that they originate from quasi-bound states arising at a carbon vacancy or adatom bonded to a single C atom when its sp2 orbital transforms to an sp3 state (12–15). Such defects are known to produce sharp electronic resonances at energies near the Dirac point (16, 17). While these defects have been extensively investigated by ab initio calculations and STM studies (12–19), their prominent role in dissipation has not been anticipated by previous work. The defect- induced CPS originates from the part of local density of electronic states that mediates electron-phonon coupling (EP-LDOS, 𝐷𝑒𝑝(𝜀)) – a quantity that was inaccessible hitherto – convoluted with the electron and phonon Fermi and Bose energy distribution functions (Section 4 (6)). Our measurement results and analysis indicate that the spectral width of 𝐷𝑒𝑝(𝜀) is much greater than the thermal broadening, in which case the CPS can be approximated as 𝑃𝑒𝑝(𝜀) ∝ 𝐷𝑒𝑝(𝜀) (Section 4 (6)). This quantity can be probed experimentally as illustrated schematically in Figs. 2A-C. By parking the tSOT above the defect and varying 𝑉𝑡𝑔 we can induce local band bending that shifts 𝐷𝑒𝑝(𝜀) with respect to 𝜀𝐹 (Section 6 (6)). The resulting variation in the measured temperature 𝛿𝑇(𝑉𝑡𝑔) provides a spectroscopic measurement of 𝑃𝑒𝑝(𝜀) ∝ 𝐷𝑒𝑝(𝜀) (Fig. 2B). Additional information can be obtained by tuning 𝜀𝐹 by the back-gate 𝑉𝑏𝑔, yielding resonant peaks in 𝐷𝑒𝑝(𝜀) aligned as diagonal lines in the 𝛿𝑇(𝑉𝑡𝑔, 𝑉𝑏𝑔) map as pictured in Fig. 2C. Importantly, in this configuration phonon emission can be turned on and off by applying a potential on the tSOT tip, representing a new instance of controlling non-equilibrium dynamics and probing it at the nanoscale. The experimental 𝛿𝑇(𝑉𝑡𝑔, 𝑉𝑏𝑔) map of defect 'C', presented in Fig. 2D, displays, in agreement with the discussion above, a sharp resonance (Fig. 2E) along a single diagonal line, which passes closely to the origin in the 𝑉𝑡𝑔-𝑉𝑏𝑔 plane. This resonance is due to the presence of a LS with a narrow single peak in 𝐷𝑒𝑝(𝜀) (Fig. 2F) close to the Dirac energy (Section 8 (6)). Such LSs give rise to the sharp thermal rings observed in Fig. 1C as follows. In order for the LS to cause inelastic electron scattering its energy level 𝐸𝐿𝑆 has to be aligned with the energy of the impinging electrons, 𝐸𝑒 ≅ 𝐸𝐹. For a given tip potential 𝑉𝑡𝑔 this resonant condition occurs when the tip is located at a distance 𝑅 from a LS (Fig. 3A, Movie S3 and Section 6 (6)). As a result, each LS displays a sharp peak in 𝛿𝑇(𝑥, 𝑦) and 𝑇𝑎𝑐(𝑥, 𝑦) along a ring of radius 𝑅 as shown in Fig. 3B, describing the ring formations in Figs. 1C and 3C. The variable tip and back gate potentials also provide the spectroscopic means to extract the energy level and the CPS of the LS. By repeatedly scanning the tSOT along the line crossing the defect and incrementing 𝑉𝑏𝑔 a bell-shaped resonance trace is obtained (Fig. 4A). The shape and polarity of the trace confirms the electrostatic picture described above (Figs. 4A-C, and Section 7 and Movies S5-S6 (6)), and its asymptotic 3  𝐿𝑆 describes the energy level 𝐸𝐿𝑆 of the LS. Similar information can be obtained by sweeping 𝑉𝑡𝑔 at value 𝑉𝑏𝑔 various values of 𝑉𝑏𝑔 (Section 7 and Movies S4,S7 (6)). Our spectroscopic analysis of bulk LSs leads to the following conclusions: a) The LS energy resides slightly below the Dirac point, 𝜀𝐿𝑆 = 𝐸𝐿𝑆 − 𝐸𝐷 ≅ −22 meV, as derived from the analysis of the resonance lines (Section 8 (6)). b) The spectral width of the CPS, 𝑃𝑒𝑝(𝜀), is about 13 meV (Fig. 2F and Section 9 (6)). c) No additional LS energy levels are observed at least 180 meV above and below 𝐸𝐿𝑆. If present, additional concentric rings and bell-shaped traces would have formed. These are, however, absent for the entire 𝑉𝑏𝑔 and 𝑉𝑡𝑔 applied range of ±10 V (Section 7 (6)). d) This level spacing puts an upper bound on the spatial extent of the LS of less than 2 nm based on charging energy calculation (Section 7 (6)). e) The sharp energy level and nanometer spatial extent of the LS closely resemble the characteristic features of atomic defects in graphene derived by ab-initio calculations (14) and observed by STM (16, 17) in contrast to more extended non-resonant "puddles" originating from disordered substrate potential (20, 21). f) The resonant character of the defects and the energy level close to Dirac point are consistent with the sp3 band vacancy model. In particular, vacancies and monovalent adatoms are known to form LSs at energies comparable with our measured values (14, 16) (Section 4 (6)) and thus are strong candidates for the observed bulk defects. g) The bulk defects are extremely rare (we found a total of seven such defects in a number of samples) corresponding to an estimated average areal density of 2∙105 cm-2 or volume concentration of 5∙10-5 ppm if originating from the parent graphite. Their spectral properties appear to be the same within our experimental resolution (Section 8 (6)) pointing to a common chemical or structural origin. These conclusions are reinforced by examining the LSs along the edges of graphene which display spectroscopic features very similar to those of the bulk defects, albeit with some notable differences. Figures 4D-E show tSOT line scans along the bottom edge of the sample crossing through numerous LSs, while increasing 𝑉𝑏𝑔 (Movies S6-7 (6)). Each LS is visible as a bell-shaped trace similar to the ones in Figs. 4A-C, indicating the same microscopic origin of different LSs. In contrast to LSs in the bulk, the edge LSs are extremely dense with some adjacent states only about 1 nm apart (Fig. S15E in (6)). This puts an additional bound on their spatial extent (Section 10 (6)). We ascribe their origin to the carbon dangling bonds near the edge with high affinity to adatoms and molecules, giving rise to resonance states formed by the resulting sp3 vacancies (12–15). Notably, quite unlike the LSs in the bulk, the edge LSs display vast variations in 𝐸𝐿𝑆 values, manifested in Figs. 4D-E by the vertical spread of the bell-shaped traces over entire 𝑉𝑏𝑔 range of 20 V. This translates into 260 meV spread in 𝐸𝐿𝑆, limited by our bias range. Remarkably, we resolve states that are less than 2 nm apart, but differ in their 𝐸𝐿𝑆 by as much as 160 meV (Section 10 (6)). Such large energy-level variation may arise from a more diverse chemical origin of the atomic defects at graphene edges as compared to the native bulk defects. An alternative explanation is Coulomb interaction between the charged defects. Indeed, charging a LS by one electron would shift the energy of a neighboring LS at a distance of 2 nm by ~240 meV (Section 7 (6)) which agrees with our observations. The above results suggest that hot electrons lose most of their energy to phonons at the edges of graphene. To verify this conjecture we measured the bare 𝛿𝑇(𝑥, 𝑦) in the absence of tSOT electrostatic influence (𝑉𝑡𝑔 at a flat-band condition, see Section 5 (6)). Measurement on Sample 2 (Fig. 4F), presented in Fig. 4G, reveals enhanced temperature along the sample edges as compared to a relatively lower temperature in the sample 4  bulk. Corroborated by numerical simulations (Section 5 (6)), this finding implies that the LSs along the graphene edges are the dominant source of dissipation at all doping levels reachable in our experiment. Therefore, the excess phonons, corresponding to the overall temperature rise, are not generated locally in the graphene bulk. Instead, the phonons are predominantly emitted through inelastic electron scattering by those LSs at graphene edges that are at resonance at a given 𝑉𝑏𝑔 value (Section 5 (6)). The observed atomic- scale resonant LSs thus emerge as the leading mechanism of dissipation in clean graphene, each acting as point-like lighthouse emitting phonons which then propagate ballistically throughout the entire sample. The observation of sharply localized resonant states and unveiling their role in dissipation should have significant implications for thermal (22), magnetic (12, 16, 23), chemical (24) and transport (25–28) properties of graphene. These states are distinct from the extended edge states anticipated for crystalline graphene edges (23). Further, resonant dissipation is completely different from the conventional non- resonant picture of electron-phonon coupling (2–4), posing interesting questions for future experimental and theoretical work. The new dissipation mechanism may affect the edge transport characteristics (29–31) and explain previous observations of the mean free path being limited by the device size in state-of-the-art encapsulated graphene (7–9). The resonance states, localized at the edge and in the bulk, thus emerge as a key factor governing the dissipation and possibly limiting the carrier mobility in pure graphene. 5  References and Notes: 1. A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, A. K. Geim, The electronic properties of graphene. Rev. Mod. Phys. 81, 109–162 (2009). R. Bistritzer, A. H. Macdonald, Electronic cooling in graphene. Phys. Rev. Lett. 102, 206410 (2009). 2. 3. W. K. Tse, S. Das Sarma, Energy relaxation of hot Dirac fermions in graphene. Phys. Rev. B. 79, 235406 4. 5. (2009). J. C. W. Song, M. Y. Reizer, L. S. Levitov, Disorder-assisted electron-phonon scattering and cooling pathways in graphene. Phys. Rev. Lett. 109, 106602 (2012). D. Halbertal, J. Cuppens, M. Ben Shalom, L. Embon, N. Shadmi, Y. Anahory, H. R. Naren, J. Sarkar, A. Uri, Y. Ronen, Y. Myasoedov, L. S. Levitov, E. Joselevich, A. K. Geim, E. Zeldov, Nanoscale thermal imaging of dissipation in quantum systems. Nature. 539, 407–410 (2016). 6. Materials and methods are available as supplementary materials. 7. C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shepard, J. Hone, Boron nitride substrates for high-quality graphene electronics. Nat. Nanotechnol. 5, 722–726 (2010). A. S. Mayorov, R. V Gorbachev, S. V Morozov, L. Britnell, R. Jalil, L. A. Ponomarenko, P. Blake, K. S. Novoselov, K. Watanabe, T. Taniguchi, A. K. Geim, Micrometer-scale ballistic transport in encapsulated graphene at room temperature. Nano Lett. 11, 2396–2399 (2011). D. A. Bandurin, I. Torre, R. Krishna Kumar, M. Ben Shalom, A. Tomadin, A. Principi, G. H. Auton, E. Khestanova, K. S. Novoselov, I. V Grigorieva, L. A. Ponomarenko, A. K. Geim, M. Polini, Negative local resistance caused by viscous electron backflow in graphene. Science. 351, 1055–1058 (2016). 8. 9. 10. D. Vasyukov, Y. Anahory, L. Embon, D. Halbertal, J. Cuppens, L. Neeman, A. Finkler, Y. Segev, Y. Myasoedov, M. L. Rappaport, M. E. Huber, E. Zeldov, A scanning superconducting quantum interference device with single electron spin sensitivity. Nat. Nanotechnol. 8, 639–44 (2013). 11. A. Finkler, D. Vasyukov, Y. Segev, L. Ne'Eman, E. O. Lachman, M. L. Rappaport, Y. Myasoedov, E. Zeldov, M. E. Huber, Scanning superconducting quantum interference device on a tip for magnetic imaging of nanoscale phenomena. Rev. Sci. Instrum. 83, 73702 (2013). V. M. Pereira, F. Guinea, J. M. B. Lopes dos Santos, N. M. R. Peres, A. H. Castro Neto, Disorder induced localized states in graphene. Phys. Rev. Lett. 96, 36801 (2006). 12. 13. D. W. Boukhvalov, M. I. Katsnelson, A. I. Lichtenstein, Hydrogen on graphene: Electronic structure, total energy, structural distortions and magnetism from first-principles calculations. Phys. Rev. B. 77, 35427 (2008). T. O. Wehling, M. I. Katsnelson, A. I. Lichtenstein, Impurities on graphene: Midgap states and migration barriers. Phys. Rev. B. 80, 85428 (2009). 14. 15. A. V. Shytov, D. A. Abanin, L. S. Levitov, Long-range interaction between adatoms in graphene. Phys. Rev. Lett. 103, 16806 (2009). 17. 16. H. Gonzales-Herrero, J. M. Gomez-Rodriguez, P. Mallet, M. Moaied, J. J. Palacios, C. Salgado, M. M. Ugeda, J. Veuillen, F. Yndurain, I. Brihuega, Atomic-scale control of graphene magnetism by using hydrogen atoms. Science. 352, 437–441 (2016). J. Mao, Y. Jiang, D. Moldovan, G. Li, K. Watanabe, T. Taniguchi, M. R. Masir, F. M. Peeters, E. Y. Andrei, Realization of a tunable artificial atom at a supercritically charged vacancy in graphene. Nat. Phys. 12, 545–550 (2016). Z. H. Ni, L. A. Ponomarenko, R. R. Nair, R. Yang, S. Anissimova, I. V Grigorieva, F. Schedin, P. Blake, Z. X. Shen, E. H. Hill, K. S. Novoselov, A. K. Geim, On resonant scatterers as a factor limiting carrier mobility in graphene. Nano Lett. 10, 3868–3872 (2010). 18. 19. M. Titov, P. M. Ostrovsky, I. V. Gornyi, A. Schuessler, A. D. Mirlin, Charge transport in graphene with resonant scatterers. Phys. Rev. Lett. 104, 76802 (2010). 20. D. Wong, J. J. Velasco, L. Ju, J. Lee, S. Kahn, H.-Z. Tsai, C. Germany, T. Taniguchi, K. Watanabe, A. Zettl, F. Wang, M. F. Crommie, Characterization and manipulation of individual defects in insulating hexagonal boron nitride using scanning tunnelling microscopy. Nat. Nanotechnol. 10, 949–953 (2015). J. Martin, N. Akerman, G. Ulbricht, T. Lohmann, J. H. Smet, K. von Klitzing, A. Yacoby, Observation of 21. 6  22. 23. 24. electron-hole puddles in graphene using a scanning single electron transistor. Nat. Phys. 4, 144–148 (2008). F. Menges, H. Riel, A. Stemmer, C. Dimitrakopoulos, B. Gotsmann, Thermal transport into graphene through nanoscopic contacts. Phys. Rev. Lett. 111, 205901 (2013). C. Tao, L. Jiao, O. V Yazyev, Y. Chen, J. Feng, X. Zhang, R. B. Capaz, J. M. Tour, A. Zettl, S. G. Louie, H. Dai, M. F. Crommie, Spatially resolving edge states of chiral graphene nanoribbons. Nat. Phys. 7, 616– 620 (2011). J. Dauber, B. Terres, C. Volk, S. Trellenkamp, C. Stampfer, Reducing disorder in graphene nanoribbons by chemical edge modification. Appl. Phys. Lett. 104, 83105 (2014). 25. D. Bischoff, A. Varlet, P. Simonet, M. Eich, H. C. Overweg, T. Ihn, K. Ensslin, Localized charge carriers in graphene nanodevices. Appl. Phys. Rev. 2, 31301 (2015). 26. M. Y. Han, J. C. Brant, P. Kim, Electron transport in disordered graphene nanoribbons. Phys. Rev. Lett. 27. 104, 56801 (2010). S. Jung, G. M. Rutter, N. N. Klimov, D. B. Newell, I. Calizo, A. R. Hight-Walker, N. B. Zhitenev, J. A. Stroscio, Evolution of microscopic localization in graphene in a magnetic field from scattering resonances to quantum dots. Nat. Phys. 7, 245–251 (2011). 28. A. G. F. Garcia, M. König, D. Goldhaber-Gordon, K. Todd, Scanning gate microscopy of localized states in wide graphene constrictions. Phys. Rev. B. 87, 85446 (2013). 29. M. T. Allen, O. Shtanko, I. C. Fulga, A. Akhmerov, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, L. S. Levitov, A. Yacoby, Spatially resolved edge currents and guided-wave electronic states in graphene. Nat. Phys. 12, 128–133 (2016). J. Chae, S. Jung, S. Woo, H. Baek, J. Ha, Y. J. Song, Y. W. Son, N. B. Zhitenev, J. A. Stroscio, Y. Kuk, Enhanced carrier transport along edges of graphene devices. Nano Lett. 12, 1839–1844 (2012). 30. 31. M. J. Zhu, A. V Kretinin, M. D. Thompson, D. A. Bandurin, S. Hu, G. L. Yu, J. Birkbeck, A. Mishchenko, I. J. Vera-Marun, K. Watanabe, T. Taniguchi, M. Polini, J. R. Prance, K. S. Novoselov, A. K. Geim, M. Ben Shalom, Edge currents shunt the insulating bulk in gapped graphene. Nat. Commun. 8, 14552 (2017). Acknowledgements: We thank M. E. Huber for SOT readout setup and M. Solodky for data analysis assistance. This work was supported by the Minerva Foundation with funding from the Federal German Ministry of Education and Research, by the NSF/DMR-BSF Binational Science Foundation BSF No. 2015653 and NSF No. 1609519, by Weizmann – UK Making Connections Programme, and by Rosa and Emilio Segré Research Award. A.K.G. and M.B. acknowledge support from EPSRC - EP/N010345/1, and from the European Research Council ARTIMATTER project - ERC-2012-ADG. L.S.L. and E.Z. acknowledge the support of the MISTI MIT-Israel Seed Fund. 7 Tac [μK] Vtg C tSOT hBN/Gr/hBN Vbg SiO2/Si A B 500 nm Idc A C B 500 nm E E e D e p 20 nm Tac [μK] Figure 1. Observing individual dissipation sources in a graphene heterostructure. (A) Schematic side view of the measurement setup with hBN-graphene-hBN heterostructure and SQUID-on-tip image of the device patterned into 4×4 µm2 square nanothermometer (tSOT). (B) Optical chamber (bright). A fixed current (cid:1)(cid:2)(cid:3) (cid:4) 3 µA is driven through the connecting constrictions (cid:7)(cid:8)(cid:9)(cid:10), (cid:12)(cid:13) of the area outlined in (B) at (cid:14)(cid:15)(cid:16) (cid:4) (cid:17)2 V (arrows). (C) Scanning ac nano-thermometry (cid:5)(cid:6)(cid:3) and (cid:14)(cid:18)(cid:16) (cid:4) 9 V at 4.2 K. The tSOT is scanned while oscillating with amplitude (cid:10)(cid:6)(cid:3) (cid:4) 2.7 nm at 12° to (cid:10) axis. Scan area 5.5 × 5 µm2, pixel size 18 nm, scan-speed 20 ms/pixel, (cid:19) (cid:4) 20 nm, (cid:1)(cid:2)(cid:3) (cid:4) 3 µA. The sharp rings (marked 'A','B','C') uncover three isolated sources of dissipation in the bulk of graphene in addition to a dense array of resonances along the graphene edges. (D) Zoom-in on defect 'C' at (cid:14)(cid:18)(cid:16) (cid:4) 5 V. Scan area 140 × 150 nm2, pixel size 1.9 nm, scan-speed 20 ms/pixel, (cid:19) (cid:4) 20 nm, (cid:1)(cid:2)(cid:3) (cid:4) 3 µA. (E) Schematic image of an atomic defect in graphene that creates a localized resonant dissipative state at the center of the ring (D) mediating inelastic scattering of impinging electron (red) into a phonon (orange) and a lower energy electron (blue). The defect, forming an sp3 orbital (green), can arise from carbon vacancy, adatom, or admolecule. 8  Figure 2. Thermal nano-spectroscopy of dissipative localized states. (A) Schematic spectrum of EP-LDOS 𝐷𝑒𝑝(𝜀) (blue) and the graphene Dirac dispersion relation (green). (B) Temperature variation 𝛿𝑇(𝑉𝑡𝑔) measured above the defect vs. tip potential 𝑉𝑡𝑔 provides nanoscale spectroscopy of the cooling power spectrum 𝑃𝑒𝑝(𝜀) by tip-induced band bending. (C) The expected 𝛿𝑇(𝑉𝑏𝑔, 𝑉𝑡𝑔) showing diagonal resonance lines that map the peaks in 𝐷𝑒𝑝(𝜀). (D) The experimental 𝛿𝑇(𝑉𝑏𝑔, 𝑉𝑡𝑔) measured by the tSOT at ℎ = 10 nm above the center of defect 'C' in presence of 𝐼𝑑𝑐 = 3 µA, revealing a single resonance dissipation line. Signal 𝑡𝑔 over 𝑉𝑡𝑔 (Section 2 (6)). (E) A zoomed-in line cut 𝛿𝑇(𝑉𝑡𝑔) along the obtained through integration of 𝑇𝑎𝑐 𝐿𝑆 = 0.2 V. (F) The cooling power spectrum 𝑃𝑒𝑝(𝜀) ∝ 𝐷𝑒𝑝(𝜀), derived from dashed line in (D) at 𝑉𝑏𝑔 = 𝑉𝑏𝑔 measurement of the resonant dissipation ring (Section 9 (6)) showing a single sharp peak near the Dirac point at 𝜀𝐿𝑆 = −22 meV. 9 Dep(ε)E[meV]ka.u.a.u.AVtg[V]δT Vbg= 0 Va.u.BEδT[µK]𝑉𝑏𝑔=𝑉𝑏𝑔𝐿𝑆=0.2 VVtg[V]Experimental dataSchematic EP-LDOSPep(ε) [a.u.]ε[meV]FδE=13 meVεLSδT[µK]Empty stateOccupied stateVtg[V]Vbg[V]DVtg[V]Vbg[V]CδT[a.u.]Vtg[V]Vbg[V]CδT[a.u.] Figure 3. Origin of the resonant ring structures. (A) Schematic description of the systems' energy levels corresponding to tip position marked by black point in (C): Two localized states with a peak in EP-LDOS at 𝐸𝐿𝑆 pinned to Dirac energy 𝐸𝐷. The tip potential 𝑉𝑡𝑔 induces band-bending resulting in the calculated position- dependent 𝐸𝐷(𝑥) (blue). The left LS is off resonance while the tSOT positioned at 𝑥 =– 𝑅 brings the right LS into resonance for inelastic scattering of electrons injected at energy 𝐸𝑒 (red arrows). The result is a point- source of phonon emission (red arcs). The Fermi energy 𝐸𝐹 (dashed line) is determined by the back gate voltage 𝑉𝑏𝑔. (B) Dashed line: Schematic temperature profiles due to ballistic 2D phonon emission from LS at resonant conditions, 𝛿𝑇𝑜𝑛(𝑥) ∝ (1 + (𝑥/ℓ)2)−1/2 with ℓ = 100 nm (Section 5.2 (6)). Red line: Calculated temperature variation 𝛿𝑇(𝑥) measured by the scanning tSOT that brings the LS into resonance at a distance ±𝑅 from each defect. Blue: Calculated 𝑇𝑎𝑐(𝑥) = 𝑥𝑎𝑐𝑑𝑇(𝑥)/𝑑𝑥 measured by the tSOT vibrating parallel to the surface. (C) 𝑇𝑎𝑐(𝑥, 𝑦) measured in the central region of Fig. 1C showing dissipation rings around defects 'B' and 'C'. The dashed line describes the scan direction depicted in (A). Scan area 3.5  0.4 µm2, pixel size 8 nm, scan-speed 20 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = −0.3 V, 𝑉𝑡𝑔 = 5.5 V, 𝐼𝑑𝑐 = 6 µA and 𝑥𝑎𝑐 = 2.7 nm directed at 18 to the x-axis. 10 200 nmCBCTac[μK]BδTon,leftδTTacδTon,rightxT-RR0AED(x)eELSELSxE-RR0tSOTEFEe Figure 4. Thermal spectroscopy of individual bulk and edge localized states. (A) Map of 𝑇𝑎𝑐(𝑥) line scans through defect C upon varying 𝑉𝑏𝑔 at 𝑉𝑡𝑔 = − 5 V showing the bell-shaped resonance trace. (B) Resonance 𝐿𝑆. (C) Resonance trace as in (A) at 𝑉𝑡𝑔 = 4.2 V. traces for various values of 𝑉𝑡𝑔 that switch their polarity at 𝑉𝑏𝑔 Scan parameters for (A-C): ℎ = 20 nm, pixel width 5 nm, pixel height 30 mV, scan-speed 20 ms/pixel, 𝑥𝑎𝑐 = 2.7 nm, 𝐼𝑑𝑐 = 3 µA, linear subtraction was applied to each line to emphasize the resonance traces (see Fig. S11 and Movie S5 for raw images). (D-E), Maps of 𝑇𝑎𝑐(𝑥) line scans along the bottom edge of the square graphene sample in Fig. 1C at 𝑉𝑡𝑔 = −10 V (D) and 10 V (E). Each bell-shaped trace originates from a single dissipative atomic defect. Scan parameters: ℎ = 20 nm, pixel width 4 nm, pixel height 100 mV, scan-speed 60 ms/pixel, 𝑥𝑎𝑐 = 2.7 nm, 𝐼𝑑𝑐 = 3 µA, high-pass filtering was used to emphasize the pertinent bell-shaped resonance traces. (F) Optical image of Sample 2 with patterned hBN/graphene/hBN heterostructure (bright) on SiO2 substrate (dark) and dc current 𝐼𝑑𝑐 = 1 μA rms chopped at 35.5 Hz applied as indicated by the arrows. (G) Thermal image 𝛿𝑇(𝑥, 𝑦) revealing dissipation along the graphene edges. Scan area 4.7  3.7 µm2, 𝐹𝐵 at this pixel size 50 nm, scan-speed 200 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = 8 V, 𝑉𝑡𝑔 =0.85 V corresponding to 𝑉𝑡𝑔 𝑉𝑏𝑔. 11 Vtg[V]Vbg[V]x[nm]𝑉𝑏𝑔𝐿𝑆BTac[μK]Vtg=-5 VVbg[V]Ax[nm]Vtg=4.2 VCVbg[V]x[nm]Tac[μK]500 nmVbg=8 VGδT[μK]Idc500 nmFIdcVtg=-10 VTac[μK]Vbg[V]DVtg=10 Vx[μm]Vbg[V]E Supplementary Materials for: Imaging resonant dissipation from individual atomic defects in graphene Materials and Methods 1. Sample fabrication and transport characterization Our devices are made from single-layer graphene placed atop a relatively thick (30 nm) crystal of hexagonal boron nitride (hBN), and covered by a thinner (10 nm) hBN. The heterostructures are assembled using the dry-peel technique on top of an oxidized Si wafer (90 nm of SiO2) which served as a back gate, and then annealed at 300 C in Ar-H2 atmosphere for 3 hours. After this, a PMMA mask is designed by electron beam lithography to define the heterostructures' boundaries and, in a similar step, to define the contacts (50 nm of superconducting Nb) positions. The exposed heterostructure is etched by a reactive ion plasma combination of CHF3 and O2 which is optimized to uncover a narrow strip (~10 nm wide) of the otherwise encapsulated graphene. This procedure yields low contact resistance (31) helping to reduce the local heating at the metal-graphene interface. The graphene/Nb contact interface was designed to be located far outside the imaging chamber (10 µm away). In the process of defining the imaging chambers geometry the plasma etching conditions were optimized to minimize the uncovered graphene edge to about 1 nm. The fabrication procedure results in high-quality samples with momentum-relaxation mean free path that is limited by the chamber dimensions over a wide range of carrier concentrations and temperatures (9). The studied sample, shown in Fig. 1 and referred to as Sample 1, consisted of a square-shaped chamber of 4  4 µm2 with three constrictions at the corners connected to external electrical contacts. The three contact points to the chamber were insufficient to perform a four-probe measurement of its transport properties. The top constriction, however, had additional contacts above it, which allowed us to carry out four-probe measurement of its magneto-resistance as a function of 𝑉𝑏𝑔 as shown in Fig. S1. The quantum Hall oscillation data allow us to extract the back-gate capacitance to the graphene 𝐶𝑏𝑔 = 1.25 ∙ 1011 1/cm2V and determine 𝐶𝑁𝑃 = −0.26 V. The CNP of the 500 nm wide the charge neutrality point (CNP) of the constriction, 𝑉𝑏𝑔 𝐶𝑁𝑃 = constriction, however, may differ from the CNP in the square bulk chamber which we estimate to be 𝑉𝑏𝑔 0.48 V as described in Section 8 below. A second sample (Sample 2) of similar structure but with thicker hBN top layer of 30 nm was used for the measurements described in some sections in the SM as noted. Sample 2 had 𝐶𝑏𝑔 = 0.9∙ 1011 1/cm2V and 𝑉𝑏𝑔 𝐶𝑁𝑃 = 0.6 V. 2. tSOT characterization and thermal imaging schemes The tSOT was fabricated by self-aligned three-step Pb deposition onto a pulled quartz pipette as described previously (5, 10). Figure S2A shows a SEM image of the tSOT device used in this work for measurements on Sample 1. The I-V characteristics of the tSOT at 4.2 K are shown in Fig. S2C at various applied magnetic fields. The current 𝐼𝑡𝑆𝑂𝑇 is measured using a SQUID series array amplifier (11) (SSAA) as shown in the inset of Fig. S2C. For 𝐼𝑡𝑆𝑂𝑇 < 𝐼𝑐(𝐻) the device is in the superconducting state, while at currents above the critical current 𝐼𝑐(𝐻) a finite voltage 𝑉𝑡𝑆𝑂𝑇 is developed redirecting part of the externally applied current to the shunt resistor 𝑅𝑠. The interference pattern of the critical current 𝐼𝑐(𝐻) in Fig. S2D shows the central lobe of period 2.4 T corresponding to effective tSOT diameter 𝑑 = 33 nm. The higher lobes are suppressed and truncated by the upper critical field 𝐻𝑐2 = 1.73 T of the Pb film. The magnetic and thermal sensitivities of the tSOT were characterized as described in Refs. (5, 10). The flux noise of the tSOT in the white noise region above about 1 kHz at 4.2 K was 145 n0/Hz1/2 at applied field 𝜇0𝐻 = 0.56 T, corresponding to spin noise of 0.85 µB/Hz1/2; the thermal noise at zero field was 510 nK/Hz1/2. The thermal coupling between the tSOT and the sample was provided by 60 mBar of He exchange gas (5). Sample 2 was studied using a similar tSOT of 48 nm effective diameter. 12  The tSOT was attached to a quartz tuning fork (TF) as shown in Fig. S2B. The TF was excited by applying an ac excitation voltage 𝑉𝑇𝐹 to it at the resonance frequency 𝑓𝑇𝐹 = 36.9867 kHz and the current through the TF was monitored. As the tip approaches the sample surface to within a few nm, a shift in 𝑓𝑇𝐹 and in the phase of the current is detected and used as a feedback for height control. After approaching the surface, the tSOT was retracted to a height ℎ of 10 to 40 nm (as indicated in figure captions) above the surface of the hBN, the feedback was switched off, and the scanning was performed at a fixed height ℎ. 𝑇𝐹(𝑥, 𝑦) = 𝑥𝑎𝑐𝑑𝛿𝑇(𝑥, 𝑦)/𝑑𝑥. Figures 1, 3 and 4 show such 𝑇𝑎𝑐 In order to avoid the 1/𝑓 noise of the tSOT, in addition to the dc 𝛿𝑇 measurement, in our studies we also employed 𝑇𝑎𝑐 measurements using two different methods. The first method uses the ac oscillation of the tSOT, 𝑥𝑎𝑐, parallel to the sample surface by applying excitation 𝑉𝑇𝐹 to the TF at its resonance frequency 𝑇𝐹 reflects the local temperature gradient along the vibration direction of the 𝑓𝑇𝐹. The resulting measured 𝑇𝑎𝑐 𝑇𝐹 data which we refer to as 𝑇𝑎𝑐 in the TF, 𝑇𝑎𝑐 𝑇𝐹(𝑥, 𝑦) is imaged over a sufficiently large area, the dc main text for brevity. If 𝑥𝑎𝑐 is sufficiently small and 𝑇𝑎𝑐 temperature variation can be reconstructed by integration 𝛿𝑇(𝑥, 𝑦) = ∫ 𝑑𝑥′𝑇𝑎𝑐 /𝑥𝑎𝑐. Such integration, however, is usually unnecessary since the essential information can be readily extracted directly 𝑎𝑐 to the tip in addition to the dc 𝑉𝑡𝑔, which from 𝑇𝑎𝑐 𝑡𝑔(𝑥, 𝑦) = measures the change in the local temperature in response to a change in the tip potential, 𝑇𝑎𝑐 𝑡𝑔 vs. 𝑉𝑡𝑔 at a fixed tip position the dc temperature variation 𝛿𝑇(𝑉𝑡𝑔) 𝑉𝑡𝑔 can be obtained by integration, 𝛿𝑇(𝑉𝑡𝑔) = ∫ which provides direct spectroscopic information on the electron-phonon cooling power spectrum (see Sections 4 and 9 below). The data presented in Figs. 2D and 2E were obtained by this method. 𝑇𝐹. In the second method we apply an ac potential 𝑉𝑡𝑔 𝑎𝑐𝑑𝛿𝑇(𝑥, 𝑦)/𝑑𝑉𝑡𝑔. By measuring 𝑇𝑎𝑐 𝑡𝑔(𝑉𝑡𝑔 ′ ) 𝑇𝐹(𝑥′, 𝑦) 𝑑𝑉𝑡𝑔 ′ 𝑇𝑎𝑐 𝑥 𝑉𝑡𝑔 Usage of these three measurement methods is demonstrated in Figs. S3B-D, which details imaging of the dissipation ring of a defect on the left edge of Sample 2 (Figs. S3A and S4A). Imaging was done by 𝑎𝑐 = 0.1 V rms at 𝑓 = 3.06 kHz and 𝑥𝑎𝑐 = 1.5 nm rms at 𝑓𝑇𝐹 in the presence of 𝐼𝑑𝑐 = 1 µA rms applying 𝑉𝑡𝑔 chopped at 35.5 Hz (see Section 5 below). The 𝛿𝑇(𝑥, 𝑦) in Fig. S3B shows a ring of enhanced temperature, described schematically by the solid red line in Fig. 3B in the main text. Note that the intensity of the ring is higher within the graphene heterostructure region (right) than outside of it (left) due to better heat conductivity of the hBN/graphene/hBN as compared to SiO2 substrate. Figure S3C shows the simultaneously 𝑇𝐹(𝑥, 𝑦), described schematically by the blue line in Fig. 3B. Figure S3D shows thermal imaging measured 𝑇𝑎𝑐 using the ac tip gating method. Since varying tip potential 𝑉𝑡𝑔 changes the radius 𝑅 of the ring, the resulting 𝑎𝑐𝑑𝑇(𝑟, 𝜃)/𝑑𝑟, where 𝑟 is the signal can be expressed as 𝑇𝑎𝑐 radial distance from the center of the ring. The two ac methods are thus equivalent to measuring two spatial derivatives of the dc image, 𝑇𝑎𝑐 𝑎𝑐𝑑𝑇(𝑥, 𝑦)/𝑑𝑉𝑡𝑔 = (𝑑𝑅/𝑑𝑉𝑡𝑔)𝑉𝑡𝑔 𝑇𝐹(𝑥, 𝑦) ∝ 𝑑𝑇(𝑥, 𝑦)/𝑑𝑥 and 𝑇𝑎𝑐 𝑡𝑔(𝑥, 𝑦) ∝ 𝑑𝑇(𝑟, 𝜃)/𝑑𝑟. 𝑡𝑔(𝑥, 𝑦) = 𝑉𝑡𝑔 In the studies described in SM we have mainly employed the TF ac method (Fig. S3B) due to its superior 𝑇𝐹(𝑥, 𝑦) of the S/N performance and operational convenience. Figure 1C presents such ac thermal image 𝑇𝑎𝑐 sample. Note that the bulk of the sample has a uniform background 𝛿𝑇(𝑥, 𝑦) temperature (which is not affected by the local tip gating) that is slightly higher than that of the surrounding SiO2 substrate (see Section 5 below). This small step-wise change in 𝛿𝑇(𝑥, 𝑦) at the sample edges translates accordingly into a strip of a 𝑇𝐹(𝑥, 𝑦) signal along the left edge of the sample and a similar negative strip along the bright background 𝑇𝑎𝑐 right edge in Fig. 1C. 3. Calibration of the vibration amplitude of the tuning fork The vibration amplitude of the tSOT attached to the TF excited at its resonance frequency can be 𝑇𝐹(𝑥, 𝑦) images. We demonstrate here the procedure derived by a direct comparison of the 𝛿𝑇(𝑥, 𝑦) and 𝑇𝑎𝑐 by imaging a constriction in Sample 2 as marked in Fig. S4A. A dc current 𝐼𝑑𝑐 = 4 µA rms, chopped by a square wave at 35.5 Hz was applied to the constriction as indicated by the arrows. The resulting images of 𝐹𝐵, are presented in Figs.S4B,F. 𝛿𝑇(𝑥, 𝑦) and 𝑇𝑎𝑐 𝑇𝐹(𝑥, 𝑦), measured simultaneously at a flat-band condition 𝑉𝑡𝑔 13  The chopping of the current allows measuring 𝛿𝑇(𝑥, 𝑦) at a frequency of 35.5 Hz using a lock-in amplifier, thus significantly reducing the 1/𝑓 noise contribution to the tSOT signal. (The presented 𝛿𝑇(𝑥, 𝑦) has been corrected for the numerical factor of 2√2/𝜋 between the first harmonic lock-in reading of a square wave 𝑇𝐹(𝑥, 𝑦) = and the corresponding time averaged value.) The measured 𝑇𝑎𝑐 𝑥𝑎𝑐 𝜕𝛿𝑇 𝜕𝑥⁄ + 𝑦𝑎𝑐 𝜕𝛿𝑇 𝜕𝑦 =⁄ 𝒓𝑎𝑐 ∙ ∇𝛿𝑇 where 𝑥𝑎𝑐 and 𝑦𝑎𝑐 are the rms values of the tSOT vibration along the imaging axes and 𝒓𝑎𝑐 is the corresponding vibration vector. In order to derive 𝑥𝑎𝑐 and 𝑦𝑎𝑐 we take numerical derivatives 𝜕 𝜕𝑥⁄ of Fig. S4B (Figs. S4C,D) and perform numerical two-parameter fit of a linear 𝑇𝐹(𝑥, 𝑦) image of Fig. S4F. The resulting best fit numerical combination of Figs. S4C,D to the measured 𝑇𝑎𝑐 image, which is presented in Fig. S4E, provides values 𝑥𝑎𝑐 = 4.05 nm and 𝑦𝑎𝑐 = 0.31 nm, indicating that the TF oscillated at an angle of 4.3 deg with respect to the 𝑥 axis. This procedure was then repeated for various values of TF excitation voltage 𝑉𝑇𝐹. Figure S4G shows the resulting linear dependence of the tSOT vibration amplitude 𝑟𝑎𝑐 = √𝑥𝑎𝑐 𝑇𝐹(𝑥, 𝑦) is expected to follow 𝑇𝑎𝑐 2 on 𝑉𝑇𝐹. and 𝜕 𝜕𝑦⁄ 2 + 𝑦𝑎𝑐 Supplementary Text 4. Theoretical analysis of the electron-phonon cooling power spectrum Graphene is known to host a wide variety of atomic-scale defects which can act as resonant scatterers, trapping electrons in quasibound states. Here we show that such defects enhance the inelastic electron- phonon scattering and provide a pathway for electron-lattice energy dissipation. We define the notion of the electron-phonon cooling power spectrum, which is the quantity measured by our instrument, and show that in realistic cases it factors into a product of a contribution that depends on the defect density of states and a contribution that depends on the electron and phonon energy distributions. While here, for simplicity, we focus on the two-temperature case (𝑇𝑒 and 𝑇𝑝 for the electron and phonon bath respectively), the notion of the cooling power spectrum can be easily generalized to a more general non-equilibrium transport regime. We stress that such resonant cooling pathways have not been studied previously, neither theoretically nor experimentally. We start with a brief summary of the studies of resonant scatterers in graphene and then proceed to describe inelastic scattering mediated by these defects. Ab initio (14) and STM (16, 17) studies have shown that quasi-bound states with energies near the Dirac point arise in a robust manner when adatoms or polar groups like H, F, CH3 or OH- bind covalently to carbon atoms (Fig. 1E), transforming the trigonal sp2 orbital to the tetrahedral sp3 orbital. Each transformed C atom represents a vacancy in the 𝜋-band that produces a quasibound state i.e. a sharp resonance positioned near the Dirac point. In contrast, the defects having other symmetries (e.g. adatoms positioned between two C atoms or at a hexagon center) typically form resonances far away from the Dirac point. This behavior can be captured using a tight-binding model on the honeycomb lattice with an on-site disorder 𝐻𝑒𝑙 = ∑ 𝛿(𝑟 − 𝑟𝑖), (S1) where 𝑟𝑖 are the defect positions, and the energy parameters 𝑢𝑖 are on the order or greater than graphene bandwidth, W=6t0 (15). Focusing on a single defect placed at the origin 𝑟 = 0 on sublattice B, and passing to the momentum representation, we obtain 𝑢(𝑟) = ∑ 𝑢𝑖 †𝜓𝑟′ + ℎ. 𝑐. ) + ∑ 𝑢(𝑟) †𝜓𝑟, 𝑟−𝑟′=1 𝜓𝑟 𝑡0 (𝜓𝑟 𝑟 𝑖 𝐻𝑒𝑙 = ∑ [ 𝒌 † 𝜓𝐴,𝑘 † ] [ 𝜓𝐵,𝑘 0 𝑡0𝑓∗(𝒌) 𝑡0𝑓(𝒌) 0 ] [ 𝜓𝐴,𝑘 𝜓𝐵,𝑘 ] + ∑ 𝒌,𝒌′ 𝜓𝐵,𝒌′ 𝑢0 † 𝜓𝐵,𝒌 , (S2) where 𝑓(𝒌) = 𝑒𝑖𝒏1𝒌 + 𝑒𝑖𝒏2𝒌 + 𝑒𝑖𝒏3𝒌 with n1,2,3 the vectors connecting a C atom to its nearest neighbors, and the sum is taken over k and k' in the Brillouin zone. At low defect concentration and for 𝑢0 values large compared to the bandwidth 𝑊 = 6𝑡0, each defect hosts a single resonance state with the energy 𝜀𝐿𝑆 close to the Dirac point, broadened due to hybridization with the states in the Dirac continuum. Scattering of band electrons on the defect is described by a T-matrix, i.e. the defect potential renormalized by multiple scattering processes: 14  𝑡(𝜀) = 𝜋𝑣2 𝜀 ln(𝑖𝑊 𝜀⁄ ) + 𝛿 , 𝛿 = 𝜋 𝑣2 𝑢0⁄ ≪ 𝑊. (S3) that features a resonance centered at an energy 𝜀𝐿𝑆, near the Dirac point (𝜀𝐿𝑆 ≈ − 𝛿 ln⁄ (𝑊 𝛿⁄ )). Here the parameter δ describes detuning of the resonance from the Dirac point, 𝑊 is the bandwidth, 𝑊 ≈ 6 eV, 𝑣 ≈ 106 𝑚 𝑠⁄ is the Fermi velocity, and for simplicity we set Planck's constant equal unity. At 𝑢0 large compared to 𝑊 the resonance energy 𝜀𝐿𝑆 is nonzero and small compared to 𝑊. The sign of 𝜀𝐿𝑆 is opposite to that of 𝑢0, i.e. for 𝑢0 positive 𝜀𝐿𝑆 is of a negative sign. This is in good agreement with the results of ab initio studies (14) predicting the resonance energy values: 𝜀𝐿𝑆 = −0.03, −0.11, −0.70, −0.67 eV for the sp3 orbital state induced by H, CH3, OH, and F, respectively. The contribution of a defect to the single-particle density of states is given by 𝜋𝑣2𝜀/2 (𝜀ln (𝑊/𝜀)+𝛿)2+(𝜋𝜀/2)2 , 1 sgn𝜀. The energy dependence 𝜋 𝜋 2 (S4) 1 𝜋 Im 𝑡(𝜀) = 𝑖𝑊 𝜀 𝑊 𝜀 Im 𝑡(𝜀) defines a sharp where we used an identity ln ( peak positioned near the Dirac point. The half-width of the resonance in 𝑡(𝜀) predicted by the above model is 𝛾 ≈ 𝜋𝛿 (2ln (𝑊 𝛿⁄ )) , i.e. is small when 𝛿 is small. ) = ln ( ) + 𝑖 ⁄ The spatial behavior of the states associated with such resonances can be understood in a simple way using the low-energy Dirac Hamiltonian 𝐻 = [ 0 −𝑖𝑣𝜕+ −𝑖𝑣𝜕− 𝑢0𝛿(𝑥) ], 𝜕±=𝜕𝑥 ± 𝑖𝜕𝑦. (S5) 𝑎 𝑥−𝑖𝑦 1 ( 0 𝜓(𝑥, 𝑦) = For a large on-site energy 𝑢0, the Hamiltonian hosts a single zero-energy eigenstate of the form ), wherein the approximate particle-hole symmetry becomes exact in the large-𝑢0 limit. The wavefunction 𝜓(𝑥, 𝑦) features a power-law tail and a log-diverging normalization, and is therefore marginally delocalized. When the defect resides on the graphene B sublattice, the zero-mode wavefunction takes nonvanishing values on the A sublattice, and vice versa. Interestingly, the zero modes survive when many defects are placed on the same sublattice, A or B (15). Next we proceed to calculate the cooling power spectrum for a defect and show that in realistic cases it is proportional to the defect density of states with a prefactor that depends on the electron and phonon energy distributions. The contribution of a resonant scatterer to electron-phonon scattering can be understood pictorially as resonant trapping of band electrons on the quasi-bound state with the released energy difference transferred to phonons. Our treatment of defect-assisted electron-phonon scattering accounts for the key aspects of interaction between electrons and phonons in graphene. First, the unusually high energy of optical phonons makes optical phonons irrelevant for electron-lattice cooling at temperatures below few hundred kelvin. As a result, the long-wavelength acoustic phonons play a dominant role despite their relatively weak coupling to electrons. Second, momentum conservation strongly limits the phase volume for electron-phonon scattering in graphene bulk. Atomic defects can absorb phonon recoil momentum and thus relieve the bottleneck due to the small phase volume, giving rise to an enhanced cooling rate near the defect. Microscopically this can be described with the help of the Hamiltonian 𝐻 = 𝐻𝑒𝑙 + 𝐻𝑝ℎ + 𝐻𝑖𝑛𝑡, (S6) where 𝐻𝑒𝑙, given in Eq. (S1), describes electrons and their scattering by the defects as discussed above, 𝐻𝑝ℎ = ∑ 𝜔𝑘𝑏𝑘 + h.c. describes electron- phonon interaction. Here 𝑔√𝜔𝑘 is the graphene deformation potential matrix element (see Ref. (4)). Starting with the Hamiltonian (S6) we calculate the energy dissipation rate as describes acoustic phonons, and 𝐻𝑖𝑛𝑡 = ∑ 𝑔√𝜔𝑘𝑎𝒑′ † 𝑎𝒑𝑏𝒌 𝒑′=𝒑+𝒌 †𝑏𝑘 𝑘 𝑃𝑒𝑝(𝜀) = ∑ 𝜔𝑊𝑝→𝑝′𝑘(1 − 𝑛′)𝑛(𝑁𝑘 + 1)𝛿(𝜀′ + 𝜔 − 𝜀) 𝑝′=𝑝+𝑘 − ∑ 𝜔𝑊𝑝′𝑘→𝑝(1 − 𝑛)𝑛′𝑁𝑘𝛿(𝜀′ + 𝜔 − 𝜀) (S7) 𝑝′+𝑘=𝑝 15  Here 𝑛, 𝑛′ and 𝑁𝑘 are Fermi and Bose distributions for electrons and phonons with momenta p, p' and k, and with the energies 𝜀 = 𝜀𝑝, 𝜀′ = 𝜀𝑝′, 𝜔 = 𝜔𝑘, respectively. The two terms in (S7) describe processes of phonon emission and phonon absorption. The microscopic transition rate is evaluated from Fermi's Golden Rule as 𝑀2 with the matrix element 𝑀 describing defect-assisted electron-phonon scattering 𝑊𝒑→𝒑′𝒌 = comprising several different contributions. Namely, phonon emission can take place either before or after the carrier is trapped on the quasibound state (for details see Ref. (4)), giving 2𝜋 ℎ 𝑀 = 𝑀0(𝑘) 𝐺(𝜀, 𝑝′ + 𝑘) 1 + 𝜎3 2 𝑡(𝜀) + 𝑡(𝜀′) 1 + 𝜎3 2 𝐺(𝜀′, 𝑝 − 𝑘)𝑀0(𝑘) + ∑ 𝑡(𝜀′) 𝑞 1+𝜎3 2 𝐺 (𝜀′, 𝑞 − 𝑘 2 ) 𝑀0(𝑘)𝐺 (𝜀, 𝑞 + 𝑘 2 ) 1+𝜎3 𝑡(𝜀) (S8) 2 ⁄ where 𝑀0(𝑘) = 𝑔√𝜔𝑘 is the electron-phonon matrix element, 𝐺(𝜀, 𝑝) = 1 (𝜀 − 𝑣𝜎𝑝) is the graphene electron Greens function, with the 2x2 matrices accounting for the A/B sublattice structure. The quantity 𝑡(𝜀) is the scatterer T-matrix evaluated above, the factors account for the sublattice A or B on which the defect is located. 1+𝜎3 2 Starting from a completely general expression (S8), our subsequent analysis proceeds differently for the cases when the values of temperature 𝑇 and bias eV are (i) much smaller, or (ii) much greater than the resonance width 𝛾 = 𝜋𝛿 (2ln (𝑊 𝛿⁄ )) . In the case (i) the 2x2 matrix structure in (S8) simplifies after taking into account that the change of electron energy-which is of the order of 𝑘𝐵𝑇 and 𝑒𝑉, whichever is greater-is small on the scale of 𝛾 and thus the process is quasi-elastic. Also, at not too low temperatures the phonon momentum values k are typically large compared to electron p and p' values, allowing us to replace 𝐺(𝜀, 𝑝) with ± in the first two terms and giving an expression ⁄ 1 𝑣𝜎𝑘 𝑖𝜎×𝑘 1 2𝜋𝑣2 𝑀0(𝑘)ln ( 𝑊 𝑣𝑘 𝑣𝑘2 𝑡(𝜀)+ 𝑀(𝑘) = 𝑀0(𝑘) (S9) For temperatures of interest the phonon momentum 𝑘𝑝 ≈ 𝑘B𝑇𝑒/𝑠, where 𝑠 ≈ 2 ∙ 104 𝑚 𝑠⁄ , the sound velocity, is much greater than Fermi momentum for electron Fermi energy on resonance with the defect, 𝜀𝐿𝑆 = 𝜀𝐹. This allows us to disregard the first term in (S9) as compared to the second term. After making this approximation, plugging the matrix element 𝑀2 in (S7) and carrying out standard algebra, we arrive at an expression for the cooling rate which is a product of the defect density of states and a function of the electron and phonon temperatures 𝑡(𝜀′)𝑡(𝜀). ) 2 1+𝜎3 5 𝑃𝑒𝑝(𝜀) = 𝐴(𝜀) ((𝑘𝐵𝑇𝑒)5 − (𝑘𝐵𝑇𝑝) ) , 𝐴(𝜀) = [(𝜀ln (𝑊/𝜀)+𝛿)2+(𝜋/2)2𝜀2]2 , (S10) where we used the relation between the electron-phonon coupling constant and graphene deformation potential 𝑔2 = 𝐷/2𝜌𝑠2 with 𝐷 ≈ 50 eV and correspondingly ℏ2𝜌𝑠2𝑣2 = 1.86 × 1019 Joule-1. The quantity 𝑃𝑒𝑝(𝜀) vanishes in equilibrium as a result of detailed balance, but is nonzero when the system is driven out of equilibrium. For numerical estimates we rewrite Eq. (S10) scaling energy by 1 meV and temperature by 1 K: 48𝜁(5) ℏ3𝜌𝑠4 𝐷2 𝜋2 𝐷2 (𝜋/2)4𝜀2ln2( 𝑊 𝑣𝑘𝑝 ) 𝑃𝑒𝑝(𝜀) = 𝜀2ln2( 𝑊 𝑣𝑘 )(𝑇𝑒 5−𝑇𝑝 5) [1 K]−5 [1 meV]2 [(𝜀ln (𝑊/𝜀)+𝛿)2+(𝜋/2)2𝜀2]2 × 260 fW , (S11) Figure S5 shows 𝑃𝑒𝑝(𝜀) derived from Eq. (S11) for several indicated values of 𝜀𝐿𝑆 along with the experimental 𝛿𝑇(𝜀). The cooling power spectrum 𝑃𝑒𝑝(𝜀) exhibits resonance behavior with a sharp peak at 𝜀 = 𝜀𝐿𝑆 consistent with the experimental findings. Using for illustration 𝑇𝑝 = 4.2 K, ∆𝑇 = 1 K, and ln(𝑊 𝑣𝑘𝑝 ) ≈ ln(270) ≈5.6, the dissipated power at the defect is on the order of 6 pW at resonant conditions (Fig. S5) comparable to the experimental evaluations in Section 5. Using Eq. (S10) and a definition for the electron-phonon heat conductivity of a defect as 𝑃𝑒𝑝(𝜀) = 𝜅𝑒𝑝(𝜀, 𝑇𝑒, 𝑇𝑝)𝛥𝑇 we obtain: ⁄ 𝜅𝑒𝑝(𝜀, 𝑇𝑒, 𝑇𝑝) = 𝐴(𝜀) 5−𝑇𝑝 5 𝑇𝑒 𝛥𝑇 ≈ 5𝑇𝑒 4𝐴(𝜀) (S12) 16  which is proportional to the cooling-active part of the single-particle density of states. This result elucidates the meaning of "electron-phonon cooling power spectrum", predicting that the cooling power exhibits a resonance behavior as a function of 𝜀, which is due to an enhancement of phonon emission by trapping of an electron on the quasibound state. Factorization of the dissipated power 𝑃𝑒𝑝(𝜀) into a part due to the scatterer that mediates cooling and a part due to hot electrons impinging on the scatterer is similar to that underlying the notion of scattering cross-section introduced in the analysis of scattering probes to separate the properties of the target and the source. We also emphasize that the cooling-active part of the single-particle density of states is distinct from the total DOS measured by STM probes, since STM probes the density of all electronic states exposed at the surface whereas our technique selects only the states that mediate electron-lattice cooling. In the case (ii), where 𝑘𝐵𝑇, 𝑒𝑉 ≫ 𝛾, the scatterer energy dependence can be treated as a delta function, which gives a simple expression 𝑃𝑒𝑝(𝜀) = ∫ 𝑑𝜔 ∞ 0 (𝑁𝜔,𝑒 − 𝑁𝜔,𝑝)(𝑛𝜀𝐿𝑆−𝜔 − 𝑛𝜀𝐿𝑆+𝜔) 𝑔2𝜔2𝜈2(𝜇)𝑣 𝛾𝑎 (S13) where 𝑁𝜔,𝑒 and 𝑁𝜔,𝑝 are Bose functions evaluated at temperatures 𝑇𝑒 and 𝑇𝑝, respectively, and 𝑛𝜔 is electron Fermi distribution. The result (S13) describes a peak of width 𝑇𝑒 or 𝑇𝑝, whichever is greater. In our experiment the width of the resonance is on the order of 10 meV whereas electron temperature inferred from Joule heat dissipated at the voltage bias values used in our measurement is a few times smaller. We therefore expect the regime (i) to provide a better description of the measurement results than the regime (ii). 5. Resonant states as a dominant source of dissipation in clean graphene 5.1. Thermal imaging at the flat-band condition In order to understand the significance of the cooling pathway due to resonant LSs and to quantify its overall contribution to dissipation we performed a set of measurements on a second sample (Sample 2) at a flat-band condition that did not involve the "on" and "off" switching of the atomic sources during measurement. The sample was patterned into a rectangular chamber with four constrictions (Fig. 4F), and a tSOT of 48 nm effective diameter (~100 nm outer diameter) was used. A dc current 𝐼𝑑𝑐 = 1 μA rms chopped at 35.5 Hz was applied between two of the constrictions as indicated by the arrows in Fig. 4F. In order to reveal the thermal landscape unperturbed by the tip electrostatic potential we acquired 𝛿𝑇(𝑥, 𝑦) images at a 𝐹𝐵 as shown in Figs. 4G. The striking observation, which is prominent in Fig. 4G, flat-band condition 𝑉𝑡𝑔 = 𝑉𝑡𝑔 was 𝛿𝑇(𝑥, 𝑦) enhanced along the edges of the rectangular chamber. Such behavior is observed at all the positive and negative values of 𝑉𝑏𝑔 (Figs. 4G, S6A, S8A, and S9A) including the charge neutrality point (Fig. S9A), providing strong evidence that the dissipation occurs predominantly at the graphene edges. The reasoning leading to this conclusion is described below step by step. We note that since the flat-band 𝐹𝐵 depend on the position of the Fermi energy in graphene and hence on 𝑉𝑏𝑔 values, the flat- conditions 𝑉𝑡𝑔 band value 𝑉𝑡𝑔 = 𝑉𝑡𝑔 𝐹𝐵(𝑉𝑏𝑔) was determined for each 𝑉𝑏𝑔 separately. 5.2. Numerical simulations In order to understand what sources of dissipation could lead to 𝛿𝑇(𝑥, 𝑦) enhanced along the edges, we performed COMSOL numerical simulations of diffusive heat flow using the Sample 2 shape and the estimated values of 2000 W/m/K and 0.2 W/m/K for the heat conductivities of graphene and SiO2 substrate at 4.2 K respectively (our qualitative conclusions weakly depend on these values). The thickness of graphene in the simulation was taken as 50 nm for the reasons of practical convenience, electric and thermal conductances were scaled by the physical thickness. Figure S6 presents the calculated surface temperature distributions for four cases that describe four different cooling regimes: 17  i) Joule heating. A current of 1 µA is applied to the sample through two constrictions on the right as in the experimental configuration. Self-consistent current distribution and the corresponding local dissipation and heat diffusion were calculated numerically using a uniform sheet resistance value of 180 /□ in graphene. The resulting calculated surface temperature distribution is presented in Fig. S6C. This case yields a temperature map in agreement with expectations for the diffusive electron transport regime. The elevated temperature in the constrictions results from their significantly higher resistance, which also gives rise to the overall temperature gradient from right to left. ii) Uniform bulk-dominated dissipation. Since the ballistic electron transport cannot be simulated in COMSOL we study this regime by assuming for simplicity a uniform heat dissipation density of 80 pW/µm2 in the entire area of the graphene heterostructure, which amounts to total heat power of 1.8 nW, a value equal to that in case (i). Figure S6D shows the resulting temperature distribution which agrees with the behavior expected for ballistic electron transport with a weak and homogeneous inelastic electron-phonon scattering in the bulk. This case represents the behavior in the bulk of the rectangular chamber ignoring enhanced heating in the constrictions due to the higher current density. The essential observation in cases (i) and (ii) is that the temperature profile 𝛿𝑇(𝑥, 𝑦) through the bulk of the sample has a maximum in the central region as shown by the red and cyan lines in Fig. S6B. A similar behavior is also expected in the case of hydrodynamic electron flow in the presence of bulk-dominated inelastic electron-phonon scattering. The general reason for temperature to peak away from the sample edges is that in the case of bulk dissipation the edges of the sample will always be colder than the sample interior due to a more efficient cooling to the substrate along the sample perimeter. iii) Edge-dominated dissipation with diffusive heat propagation. We study edge-dominated dissipation by carrying out simulations similar to case (ii) in which the same total power of 1.8 nW is dissipated at discrete point defects positioned along the edges of the sample (spaced 250 nm apart, as marked by blue points in Fig. S6E). Dissipated power was 12.7 pW per defect, and no dissipation in the bulk. The resulting temperature distribution in Fig. S6E exhibits a pronounced minimum of 𝛿𝑇(𝑥, 𝑦) in the center of the sample as seen by the blue profile in Fig. S6B. This is in stark contrast to the two bulk-dominated dissipation cases (i) and (ii) characterized by temperature peaks at the sample center. We checked that the temperature minimum is rapidly washed out upon adding a small bulk dissipation. The semblance to the pronounced minimum at the center of the sample observed in experiment (green curve in Fig. S6B) suggests that dissipation in clean encapsulated graphene is dominated by the edge defects and any bulk dissipation, if present, is substantially weaker. iv) Edge-dominated dissipation with ballistic phonon propagation. Interestingly, our measurement results show a minimum in 𝛿𝑇(𝑥, 𝑦) that is even more pronounced than the simulated case of a pure edge dissipation (see temperature profile displayed in Fig. S6B). We speculate that this finding points to the ballistic propagation of emitted phonons. Indeed, in the case of heat propagation dominated by the 2D heat flow in graphene, solution of diffusive 2D heat equations results in universal −log 𝑟 temperature decay away from a point source. In contrast, in the 2D ballistic phonon regime the resulting phonon density decays much faster, as 1/𝑟 from the point source, giving rise to steeper temperature decay. Physically, in clean samples at low temperatures phonons are expected to propagate ballistically on our relevant length scales. In this case the measured 𝛿𝑇(𝑥, 𝑦) should be thought of as a map of the excess power density of the vibrational modes given by the local density of the excess phonons and by their energy. In order to simulate the ballistic phonon regime we use a simple model temperature distribution of 𝛿𝑇(𝑟) = 𝛿𝑇0(1 + (𝑟/ℓ)2)−1/2 with 𝛿𝑇0 = 100 µK imposed around each defect (shown by the dashed 𝛿𝑇𝑜𝑛(𝑥) line in Fig. 3B). Here ℓ is a characteristic length scale for the cut-off of the 1/𝑟 divergence determined by the diameter of the tSOT and the scanning height above the graphene, which we took as ℓ = 100 nm. Figure S6F shows the resulting 𝛿𝑇(𝑥, 𝑦) given by superposition of such point sources positioned at the same locations as in Fig. S6E. As illustrated in Fig. S6B, this simple model gives a good agreement with the experimental results, providing strong support for the dissipation mechanism that is dominated by inelastic electron scattering by resonant LSs at the clean graphene edge. 18  To simplify the analysis, in Fig. S6F we have assigned the same power dissipation to all the defects. In reality, however, the LSs in the current-carrying constrictions will dissipate more power due to the larger local current density resulting in higher collision rate of the electrons with the LSs. This will result in the higher intensity of the rings in the constrictions and in their proximity as indeed clearly seen in Figs. 4G, S6A, S8A, and S9A. 5.3. The density of cooling-active LSs at graphene edge Figures 4G, S6A, S8A, and S9A demonstrate that edge-dominated dissipation is present at all values of 𝑉𝑏𝑔. Since each defect creates a dissipative resonant state at a specific energy, this observation implies that different defects contribute to dissipation at different 𝑉𝑏𝑔. This is illustrated by a wide spread of resonance energies in Fig. 4 in the main text. Signatures of a multitude of different LSs can be discerned also from the measurements under the flat band condition. Indeed a closer inspection of Figs. 4G, S6A, S8A, and S9A shows inhomogeneity in 𝛿𝑇(𝑥, 𝑦) along sample edges that changes upon varying 𝑉𝑏𝑔. In Fig. 4 our numerical algorithm detected 135 LSs along 𝐿 = 3.5 µm along graphene edge in the energy window of ~260 meV (SM section 10). Taking the spectral width of each LS to be about 13 meV (Fig. 2 and SM Section 9), this implies that at any given 𝑉𝑏𝑔 there are about 7 out of 135 defects which are cooling-active. This corresponds to an average separation of 500 nm for the active defects. Since our algorithm detects only a fraction of the LSs, the actual density of the cooling-active LSs is probably somewhat greater. In simulations shown in Figs. S6E,F we used point heat sources spaced by 250 nm. 5.4. Resolving localized states by 𝑽𝒃𝒈 𝒂𝒄 modulation Since the temperature fields created by nearby LSs overlap, their individual contributions cannot be readily resolved in the 𝛿𝑇(𝑥, 𝑦) images of Figs. 4G and S6A at a flat-band condition. Upon variation in 𝑉𝑏𝑔 some LSs move out of resonance, reducing their contribution to dissipation while at the same time other LSs 𝑎𝑐 = 0.1 V to are shifted into resonance forming new point sources of dissipation. By applying an ac voltage 𝑉𝑏𝑔 𝑑𝑐 the details of this process can be resolved by imaging the the back gate in addition to the dc 𝑉𝑏𝑔 𝑏𝑔(𝑥, 𝑦) corresponding ac change 𝑇𝑎𝑐 𝑑𝑐 = –8 V and –6.7 V. The bright and dark spots that appear at random locations along images acquired at 𝑉𝑏𝑔 𝑎𝑐 the edges of graphene reflect the LSs that are correspondingly shifted in and out of resonance by the 𝑉𝑏𝑔 𝑏𝑔(𝑥, 𝑦), namely in these regions the change in 𝑉𝑏𝑔. Note that the dark spots correspond to negative 𝑇𝑎𝑐 dissipation decreases as 𝑉𝑏𝑔 increases, while in the bright spots 𝛿𝑇(𝑥, 𝑦) increases with 𝑉𝑏𝑔 giving rise to 𝑏𝑔(𝑥, 𝑦). This picture would look quite differently if the microscopic dissipation sources were positive 𝑇𝑎𝑐 independent of 𝑉𝑏𝑔. In this case, instead of strong bright and dark spots we would see only a weak and 𝑏𝑔(𝑥, 𝑦) originating from the gate dependence of graphene resistivity. smooth constant-sign background 𝑇𝑎𝑐 𝑏𝑔 in the local temperature. Figure S7 shows two examples of such 𝑇𝑎𝑐 5.5. Resolving individual LSs by tip gating The above 𝑉𝑏𝑔 𝑎𝑐 method of imaging the resonant LSs is somewhat limited since it reveals only the states that are close to resonance at a given 𝑉𝑏𝑔 , which makes resolving individual LSs that are located close to each other a challenging task. However, by varying the tip potential 𝑉𝑡𝑔 and moving away from the flat-band condition, a wide range of LSs can be individually resolved with very high spatial resolution. This enhanced functionality arises from the fact that instead of imaging the net 𝛿𝑇(𝑥, 𝑦) which is a sum of contributions due to all active LSs contributing to dissipation at a particular 𝑉𝑏𝑔, the local potential induced by the tip can be used for selective and position dependent imaging of individual LSs. This is done by switching individual LSs 𝑇𝐹(𝑥, 𝑦) due to in and out of resonance and measuring the corresponding change in the local temperature 𝑇𝑎𝑐 modulation of a single resonant state by the vibrating tip. The striking difference between 𝛿𝑇(𝑥, 𝑦) at a flat- 𝑇𝐹(𝑥, 𝑦) in the presence of tip potential is illustrated in band condition and the local temperature change 𝑇𝑎𝑐 19  Fig. S8. While Fig. S8A clearly shows that the dissipation occurs predominantly along the graphene edges, the presence of individual LSs cannot be discerned. Figure S8B, in contrast, reveals the high density of resonant LSs at graphene edges with a wide-spread energy level distribution giving rise to numerous rings of different 𝑇𝐹(𝑥, 𝑦) imaging is the reason it is mainly the data diameters. The wealth of information provided by the 𝑇𝑎𝑐 obtained by this method that are presented in the main text. The amount of detail obtained by this method is illustrated in Figure S8 which provides an interesting insight into the origin of dissipation enhancement in the current-carrying constrictions. In narrow constrictions both the current density and the edge-to-bulk ratio are high. As a result the rates for electron collision with the LSs and for the corresponding phonon emission are much higher than in the chamber, resulting in enhanced heating in the two constrictions on the right that serve as current source and drain. The high rate of phonon emission in the constrictions contributes to the overall right to left gradient in 𝛿𝑇(𝑥, 𝑦) within the chamber seen in Fig. S8A. This gradient is significantly enhanced near CNP as presented in Fig. S9 and observed as the bright horizontal streaks in Figs. S11D and S11F around 𝑉𝑏𝑔 = 0. Interestingly, the ballistic nature of electron transport in our clean graphene causes dissipation also within the two floating constrictions on the left where no current flows. We interpret the ring structures seen in the floating constrictions in Fig. S8B as being due to the hot carriers that travel ballistically across the chamber, enter the constrictions and dissipate their energy while dwelling inside the constrictions. 5.6. Current dependence The observed overall temperature increase in the sample arises predominantly from the phonons that are emitted by the resonant LSs at graphene edges and propagate throughout the sample rather than being generated in the bulk of graphene. In order to study the current dependence of the overall sample temperature we carried out thermal imaging at flat-band condition as presented in Fig. S9. Temperature profile across the sample along the dashed line in Fig. S9A was then acquired at different values of the applied current 𝐼𝑑𝑐. Examples of the resulting 𝛿𝑇 profiles at different currents are presented in Figure S9B showing the temperature increase with the current 𝐼𝑑𝑐. Figure S9C presents the average temperature 〈𝛿𝑇〉 2 behavior over a wide range of currents down to our along the profiles vs. the current, showing 〈𝛿𝑇〉 ∝ 𝐼𝑑𝑐 lowest current of 20 nA. Such low currents are well comparable to the currents typically employed in transport studies thus indicating the relevance of the observed dissipation mechanisms to the transport properties of graphene. 6. Simulations of the electrostatic interaction between the tSOT and the sample The powerful spectroscopic capabilities of the tSOT nanothermometry arise from the local electrostatic interaction between the tip and sample. In this section we derive this interaction and provide some simulations by numerical solution of Laplace's equation for the potential 𝜑(𝑟, 𝑧) in an axial-symmetric geometry. Our simulated space consisted of graphene residing in 𝑧 = 0 plane while the tSOT was modeled for simplicity as a disc with outer diameter 𝑑0 and thickness 𝑡 located at a height 𝑧𝑡𝑖𝑝 above the graphene. A fixed voltage boundary condition of 𝑉𝑡𝑔 was applied to the disc and a fixed voltage boundary condition of Vbg was applied at 𝑧 = −𝑑𝑏𝑔 to account for the back gate. The entire space above and below the graphene was taken to be homogeneous with dielectric constant 𝜖 to simplify the numerics. Zero Neumann boundary conditions were used at large 𝑟 and 𝑧. The surface charge density 𝜎(𝑟) in the graphene at 𝑧 = 0 was solved self-consistently as follows. The tip potential induces a local band bending 𝐸𝐷(𝑟) of the graphene Dirac energy and changes the local chemical potential 𝜇(𝑟) = 𝐸𝐹 − 𝐸𝐷(𝑟), while the electro-chemical potential 𝐸𝐹 remains constant throughout the graphene and fixed to 𝐸𝐹 = 0. Unlike in an ideal metal, a finite in-plane field can be present in graphene. Equilibrium conditions require that a test charge experiences a zero net in- plane force. This dictates that the in-plane field is balanced by the gradient of the chemical potential, which results in a boundary condition 𝐸𝐷(𝑟) = −𝑒𝜑(𝑟, 𝑧 = 0) + 𝐸𝐷,∞, where 𝐸𝐷,∞ is the global band bending due 20  to the back-gate voltage 𝑉𝑏𝑔. Considering the Dirac density of states 𝐷(𝜀) = 𝜕𝑛/𝜕𝜀 = 2𝜀(√𝜋ℏ𝑣𝐹) (where 𝜀 = 𝐸𝐹 − 𝐸𝐷) and the geometrical capacitance between the two parallel plates (graphene and back gate) it is straightforward to derive at analytic expression 𝐸𝐷,∞ = −𝑠𝑖𝑔𝑛(𝑉𝑏𝑔)√𝜋ℏ𝑣𝐹√𝑛∞, where: −2 𝑛∞ = (√(𝜖𝜖0√𝜋ℏ𝑣𝐹 2𝑒2𝑑𝑏𝑔 2 ) + 𝜖𝜖0 𝑒∙𝑑𝑏𝑔 𝑉𝑏𝑔 − 𝜖𝜖0√𝜋ℏ𝑣𝐹 2𝑒2𝑑𝑏𝑔 ) 2 On the other hand, 𝜎(𝑟) is determined by 𝐸𝐷(𝑟) through 𝐷(𝜀), resulting in 𝜎(𝑟) that is given by the relation 𝐸𝐷(𝑟) − 𝐸𝐹 = sign(𝜎(𝑟))√𝜋/𝑒ℏ𝑣𝐹√𝜎(𝑟), resulting in a closed set of equations that can be self- consistently solved. In the simulation results of Fig. S10 the following parameter values were used: outer tip diameter 𝑑0 = 50 nm, tip thickness 𝑡 = 5 nm, 𝜖 = 3, 𝑣𝐹 = 106 m/s, 𝑑𝑏𝑔 was set to 125 nm to match the experimentally derived 𝐶𝑏𝑔 (see Section 1 above), and 𝑧𝑡𝑖𝑝 was set to 57 nm to attain comparable 𝐶𝑡𝑔 and 𝐶𝑏𝑔 capacitances as in the experiment. The graphene band bending 𝐸𝐷(𝑥) is shown in Fig. S10A for tip potential 𝑉𝑡𝑔 = −1.5 V for various back gate voltages 𝑉𝑏𝑔. The 𝐸𝐷(𝑥) curve in Fig. 3 and Movie S3 was derived for 𝑉𝑡𝑔 = −3 V and 𝑉𝑏𝑔 = 0.25 V. The strongest tip-induced band bending occurs when 𝐸𝐷(𝑥) crosses zero where the graphene is incompressible. At high values of 𝑉𝑏𝑔 the bending is reduced due to a more effective screening. Figure S10B shows similar results for a constant 𝑉𝑏𝑔 = 1 V and different values of 𝑉𝑡𝑔. From these results we can calculate the evolution of the radius 𝑅 of the dissipation ring assuming that the LS is pinned to the Dirac point, 𝐸𝐿𝑆 = 𝐸𝐷(𝑥), where 𝑥 is the tip distance to the defect, and the carriers are injected close to the Fermi energy 𝐸𝑒 ≅ 𝐸𝐹 = 0. Figure S10C shows the resonant dissipation traces as a function of 𝑉𝑡𝑔 for various values of 𝑉𝑏𝑔 akin the experimental results in Figs. S11A-C. The points along the traces indicate the lateral displacement 𝑥 of the tSOT from the defect at which resonant conditions for inelastic scattering are met for various values of 𝑉𝑡𝑔 and 𝑉𝑏𝑔. For example, for 𝑉𝑏𝑔 = 1 V, as described in Fig. S10B, the resonant conditions occur only for negative 𝑉𝑡𝑔 < −1 V for which 𝐸𝐷(𝑥) crosses 𝐸𝐹 = 0. The distance 2𝑅 between the crossing points grows with decreasing 𝑉𝑡𝑔 resulting in the resonance trace marked by an arrow in Fig. S10C. Figure S10D presents the resonant bell-shaped traces as a function of 𝑉𝑏𝑔 for various values of 𝑉𝑡𝑔 describing the experimental results in Figs. 4A-C. 7. Search for additional energy levels of the localized state In this section we provide additional information on the spectroscopic analysis of bulk defects and explore the widest energy range for searching for additional possible energy levels of the LS. By repeatedly scanning the tSOT along the line crossing defect 'C' and incrementing 𝑉𝑡𝑔 we derive the resonant traces describing the evolution of the radius of the dissipation ring 𝑅(𝑉𝑡𝑔) as shown in Figs. S11A-C. These traces correspond to the numerical results presented in Fig. S10C. In the case of 𝑉𝑏𝑔 = 2.5 V (Fig. S11C), the concave trace reflects the situation exemplified in Fig. 3 where in absence of tip potential, 𝐸𝐿𝑆 resides below 𝐸𝐹 and hence the LS is occupied and does not dissipate until a negative 𝑉𝑡𝑔 brings it into resonance with 𝐸𝑒. For 𝑉𝑏𝑔 = − 1 V (Fig. S11A), the opposite situation occurs in which the empty state does not dissipate because it is well above 𝐸𝑒 and a positive 𝑉𝑡𝑔 is required for bringing it into resonance resulting in a convex trace. The transition from convex to concave traces upon varying 𝑉𝑏𝑔 is presented in Fig. S11B and Movie S4. Alternatively, the tSOT can be scanned through the defect while incrementing 𝑉𝑏𝑔 at constant 𝑉𝑡𝑔 (Figs. 𝐿𝑆 as expected 4A-C and Movie S5). In this case the traces are bell-shaped, in which 𝑅 diverges when 𝑉𝑏𝑔 = 𝑉𝑏𝑔 (Fig. S10D). Figures S11D and S11F show the bell-shaped resonance traces at 𝑉𝑡𝑔 = 10 V and 𝑉𝑡𝑔 = −10 V over the entire range of 𝑉𝑏𝑔 = ±10 V. Traces at various intermediate values of 𝑉𝑡𝑔 are presented in Fig. S11E. These data are summarized in Fig. S11G showing the ring radius 𝑅 in the 𝑉𝑏𝑔-𝑉𝑡𝑔 plane. In the white areas no dissipation is present since 𝐸𝐿𝑆 is far from resonance. In the colored regions inelastic scattering 21  occurs depending on the tip position. The transition line (dark blue) maps the resonant conditions when the tip is positioned directly above the defect (𝑅 = 0) as in Fig. 2D, while the red color reflects the conditions for 𝐹𝐵 at which 𝑉𝑡𝑔 corresponds to flat- diverging 𝑅. The black dot describes the bare resonance point 𝑉𝑏𝑔 band conditions of the tip and 𝑉𝑏𝑔 aligns 𝐸𝐿𝑆 with 𝐸𝑒 ≅ 𝐸𝐹 in absence of tip potential. This is the point at 𝐿𝑆 describes the position of the LS energy level 𝐸𝐿𝑆. which the resonant traces switch their polarity and 𝑉𝑏𝑔 The slope of the 𝑅 = 0 resonance line in Fig. S11G (dark blue) describes the ratio between the back gate and the tip capacitances 𝐶𝑏𝑔/𝐶𝑡𝑔 that is affected by the tip height ℎ as described in Section 6 above. With 𝐶𝑏𝑔 = 1.25 ∙ 1011 1/cm2V attained from the magneto-transport data of Section 1, we derive 𝐶𝑡𝑔 = 1.3 ∙ 1011 1/cm2V at ℎ = 12 nm. 𝐿𝑆, 𝑉𝑡𝑔 𝑚𝑎𝑥 and ±𝑉𝑏𝑔 𝑚𝑎𝑥 + 𝐶𝑏𝑔𝑉𝑏𝑔 The fact that only a single resonance trace is visible in Figs. S11D and S11F shows that the LS has just 𝑚𝑎𝑥. This translates into a range of carrier concentrations of one energy level in this range of ±𝑉𝑡𝑔 𝑚𝑎𝑥) = ±2.6 ∙ 1012 cm-2 and corresponding energy span of ±186 meV. These results ±(𝐶𝑡𝑔𝑉𝑡𝑔 show that the LS has just a single energy level 𝐸𝐿𝑆 near the Dirac point with no additional levels in the range of at least Δ𝐸 = ±186 meV from 𝐸𝐷. Considering the LS as a quantum dot, the single electron changing energy and hence the corresponding energy level separation is given by Δ𝐸 ≅ 𝑒2 (4𝜋𝜀ℎ𝐵𝑁𝜀0𝑑𝐿𝑆) where 𝑑𝐿𝑆 is the diameter of the LS and 𝜀ℎ𝐵𝑁 ≅ 3 is the dielectric constant of hBN. Taking the lower bound estimate of Δ𝐸 ≥ 186 meV we attain the maximal effective size of the LS 𝑑𝐿𝑆 ≤ 2.4 nm. ⁄ 8. Derivation of the local CNP and of the energy level 𝑬𝑳𝑺 of bulk localized state The LS resonance line in Fig. 2D provides a valuable tool for determining the local charge neutrality 𝐿𝑆. To describe the derivation point (CNP) in terms of 𝑉𝑏𝑔 method we carry out numerical simulations of the resonance line, which is the critical line at which the dissipation ring nucleates from 𝑅 = 0. 𝐶𝑁𝑃 and the energy level 𝐸𝐿𝑆 of the LS in terms of 𝑉𝑏𝑔 We first assume that the LS is pinned to the Dirac point, 𝜀𝐿𝑆 = 𝐸𝐿𝑆 − 𝐸𝐷 = 0. In this case the resonance curve is defined by the geometric place of all the 𝑉𝑏𝑔, 𝑉𝑡𝑔 pairs for which the 𝐸𝐷(𝑥) profile at the tip location 𝑥 = 0 is tangent to the Fermi energy 𝐸𝐹 = 0 as shown in Fig. S12A, such that the impinging electrons with energy 𝐸𝑒 ≅ 𝐸𝐹 are at resonance with the LS. The resulting 𝑅 = 0 resonance curves are shown in Fig. S12B for various heights 𝑧𝑡𝑖𝑝 of the tip above the graphene. The slope of the curves away from the CNP reflects the ratio between the effective capacitances 𝐶𝑏𝑔/𝐶𝑡𝑔 of the back and tip gating and is thus affected by 𝑧𝑡𝑖𝑝. 𝐹𝐵 which depends on its The intersection point of the curves is given by the flat-band condition of the tip 𝑉𝑡𝑔 𝐹𝐵 the tip has no effect on the sample, 𝐸𝐷(𝑥) is work function and is taken to be zero in the simulations. At 𝑉𝑡𝑔 𝐿𝑆 independent of the tip height. The crossing point thus constant, and the resonance occurs at 𝑉𝑏𝑔 = 𝑉𝑏𝑔 𝐿𝑆. determines the energy of the LS in terms of 𝑉𝑏𝑔 The resonance curves are not straight and show a kink which is clearly resolved by plotting the derivatives of the curves (Fig. S12C). This kink is the result of the quantum capacitance and reflects the change in effectiveness of graphene in screening the tip potential at different 𝑉𝑏𝑔. The kink occurs at the 𝐶𝑁𝑃, which is taken to be zero in the simulations, and is defined by the location of charge neutrality point 𝑉𝑏𝑔 𝐶𝑁𝑃 = 0 and hence the the minimum of the 𝑑𝑉𝑡𝑔/𝑑𝑉𝑏𝑔 curves (Fig. S12C). In the 𝜀𝐿𝑆 = 0 case, 𝑉𝑏𝑔 crossing point of the resonance curves coincides with the kink location, however, in the general case these two features are separated as demonstrated in Figs. S12D-F for 𝜀𝐿𝑆 = 30 meV. Here the crossing point of the 𝐿𝑆, while the kink is fixed curves is determined by the top-gate flat-band condition 𝑉𝑡𝑔 by the back gate CNP, 𝑉𝑏𝑔 𝐹𝐵 = 0 at which 𝑉𝑏𝑔 = 𝑉𝑏𝑔 𝐶𝑁𝑃 = 0 (Figs. S12E-F). 𝐿𝑆 = 𝑉𝑏𝑔 In the experimental resonance lines of defect 'C' in Fig. S13D (and their derivative in Fig. S13E) the two 𝐶𝑁𝑃 ≅ 480 mV. From these two values we 𝐿𝑆 ≅ 200 mV and 𝑉𝑏𝑔 features are well resolved with resulting 𝑉𝑏𝑔 derive the energy level of the LS to be 22  𝜀𝐿𝑆 = 𝐸𝐿𝑆 − 𝐸𝐷 = 𝑠𝑖𝑔𝑛(𝑉𝑏𝑔 𝐶𝑁𝑃 ≅ −22 meV. 𝐿𝑆 − 𝑉𝑏𝑔 𝐶𝑁𝑃)√𝜋/𝑒ℏ𝑣𝐹√𝐶𝑏𝑔𝑉𝑏𝑔 𝐿𝑆 − 𝑉𝑏𝑔 Figures S13A-C show the corresponding numerical simulations for the case of 𝜀𝐿𝑆 = −20 meV reproducing the experimental features. In order for the LS to be at resonance with the electrons at 𝐸𝐹 the tip potential has to induce 𝐸𝐷(𝑥) profile such that 𝐸𝐷(0) − 𝐸𝐹 = 20 meV (Fig. S13A). Theoretical calculations and STM studies (13, 14, 16, 17) show that carbon vacancies and monovalent adatoms form LSs close to the Dirac point consistent with our findings. In particular, hydrogen atoms on graphene were shown (13, 14) to give rise to sharply localized state at 30 meV below the Dirac point. Our derived 𝜀𝐿𝑆 ≅ −22 meV thus raises the possibility that the observed resonant inelastic scattering in the bulk of graphene may arise from individual hydrogen adatoms. 𝐶𝑁𝑃 and 𝑉𝑏𝑔 Note that in graphene heterostructures the CNP is known to be hysteretic with respect to 𝑉𝑏𝑔. 𝐿𝑆 are slightly dependent on the span of previously applied 𝑉𝑏𝑔 and on its sweep 𝐿𝑆 in Figs. 4A-C and S11A-C. Since 𝜀𝐿𝑆 is determined 𝐶𝑁𝑃 the above described procedure diminishes the effect of hysteresis by Accordingly, 𝑉𝑏𝑔 direction, giving rise to small variations in the apparent 𝑉𝑏𝑔 only by the difference 𝑉𝑏𝑔 measuring the two quantities in a single continuous 𝑉𝑏𝑔 sweep. 𝐿𝑆 − 𝑉𝑏𝑔 The bulk defects are extremely rare in high quality encapsulated graphene. Besides the three defects in Sample 1 we found only four additional bulk defects in other samples, while in some of the samples, like Sample 2, we could not resolve any bulk defects. From this limited statistics we can estimate the average areal density of 2∙105 cm-2 of bulk defects in graphene. Such a low density and their atomic nature may indicate that the bulk defects are not formed during the sample fabrication process but rather originate from native defects in the parent graphite with a corresponding concentration of 5∙10-5 ppm. The spectral properties of bulk defect 'C' were thoroughly investigated as presented here. The spectral properties of the other six defects were attained by measurements similar to those presented in Movies S4-5. From these assessments we conclude that for all bulk defects the LS energy level resides within at most 50 meV from the Dirac point, pointing to a common chemical or structural origin in form of carbon vacancies or hydrogen adatoms. 9. Experimental derivation of the electron-phonon cooling power spectrum Our thermal spectroscopy allows evaluation of the broadening 𝛿𝐸 of the energy level of the LS and derivation of its electron-phonon cooling power spectrum (CPS), 𝑃𝑒𝑝(𝜀), as follows. For a LS that is pinned to the Dirac point, 𝐸𝐿𝑆 = 𝐸𝐷, a finite spectral width 𝛿𝐸 of the LS translates into a finite width of the back gate voltage 𝛿𝑉𝑏𝑔, 𝛿𝐸 = 2√𝜋/𝑒ℏ𝑣𝐹√𝐶𝑏𝑔𝛿𝑉𝑏𝑔/2, over which the dissipation occurs. This broadening translates into a finite width 𝛿𝑥 = 𝛿𝑉𝑏𝑔/(𝑑𝑉𝑏𝑔 𝑑𝑥⁄ ) of the dissipation ring as described in Fig. S14. In addition to the intrinsic broadening, however, there is an extrinsic broadening due to a finite energy distribution of the injected energetic carriers and the tSOT modulation 𝑥𝑎𝑐 by the TF. In order to assess the extrinsic broadening and to minimize its contribution we performed the measurements presented in Fig. S14. Figure S14A shows the bell-shaped dissipation trace of defect 'B' at 𝑉𝑡𝑔 = −10 V. The back-gate voltage was then fixed to 𝑉𝑏𝑔 = 5.3 V, where the slope of the bell-shaped trace was 𝑑𝑉𝑏𝑔 𝑑𝑥⁄ = 47 mV/nm, as shown in Fig. S14A. Next, we performed line scans through the LS and measured the width of the ring 𝛿𝑥 (Fig. S14D) as a function of the rms amplitude of the tip oscillation 𝑥𝑎𝑐 as shown in Fig. S14B. The resulting 𝛿𝑥 decreases linearly with decreasing 𝑥𝑎𝑐, as expected when tip oscillation is larger than the intrinsic width of the ring, and saturates for 𝑥𝑎𝑐 ≲ 1.5 nm. The average energy of the injected electrons 𝜀𝑒 = 𝐸𝑒 − 𝐸𝐷 and the width of their energy distribution, which should be comparable to 𝜀𝑒, are determined by 𝑅𝑐 and the dc current 𝐼𝑑𝑐, 𝜀𝑒 = 𝑒𝑅𝑐𝐼𝑑𝑐, where 𝑅𝑐 = 890 Ω is the resistance of the constriction at 𝑉𝑏𝑔 = 5.3 V (see Fig. S1). Figure S14C shows that the ring width 𝛿𝑥 decreases with decreasing 𝐼𝑑𝑐 and saturates for 𝐼𝑑𝑐 ≲ 5 µA. The corresponding dissipation ring measured in the limit of low tip 23  oscillation of 𝑥𝑎𝑐 = 0.5 nm and low current 𝐼𝑑𝑐 = 2 µA is presented in Fig. S14D (note the 𝑇𝑎𝑐 scale of just 1 µK). The cross section 𝑇𝑎𝑐(𝑥) of the ring (Fig. S14E) shows ring width of 𝛿𝑥 = 7 nm. After minimizing the extrinsic contributions to the broadening, we can directly derive the CPS from the spectroscopic thermal signal, 𝑃𝑒𝑝(𝜀) ∝ 𝛿𝑇(𝜀). To attain this, we integrate the 𝑇𝑎𝑐(𝑥) profile of Fig. S14E to derive 𝛿𝑇(𝑥) = ∫ 𝑑𝑥′𝑇𝑎𝑐(𝑥′) /𝑥𝑎𝑐, and use coordinate transformation to translate the lateral tip position 𝑥 𝑥 is determined from Fig. S14A, the origin of 𝑥 is defined in Fig. S14E, and 𝑉𝑏𝑔 into the induced energy shift of the LS given by 𝜀 = √𝜋/𝑒ℏ𝑣𝐹√𝐶𝑏𝑔𝑉𝑏𝑔 𝐶𝑁𝑃 was taken 𝑑𝑉𝑏𝑔 𝑑𝑥⁄ from Section 8 above. The resulting 𝛿𝑇(𝜀) is presented in Fig. S14F and describes the CPS of the LS, 𝑃𝑒𝑝(𝜀) ∝ 𝛿𝑇(𝜀), which shows a sharp peak with spectral width of 𝛿𝐸 = 13 meV. Since 𝜀𝑒 = 𝑒𝑅𝑐𝐼𝑑𝑐 = 1.8 meV ≪ 𝛿𝐸, the observed 𝛿𝐸 reflects an upper bound on the intrinsic broadening of the energy level of the LS. The derived 𝑃𝑒𝑝(𝜀) is presented in Fig. 2F. 𝐿𝑆 − 𝑉𝑏𝑔 , where 𝐿𝑆 − 𝑉𝑏𝑔 𝐶𝑁𝑃 + 𝑥 𝑑𝑉𝑏𝑔 𝑑𝑥⁄ 10. Analysis of localized states along the graphene edges The numerous LS along the graphene edges were analyzed as described in Fig. S15. The tSOT was scanned along the bottom edge of the graphene sample and the bell-shaped resonance traces were acquired by sweeping the back gate 𝑉𝑏𝑔 at various 𝑉𝑡𝑔 (see Movie S6). Figure S15A shows the data at 𝑉𝑡𝑔 = − 10 V reproduced here from Fig. 4 for clarity. Figure S15B presents the same data with overlaid fits to the traces 𝐶𝑁𝑃 + 𝑉𝑝/(1 + (𝑥 − 𝑥𝑖)2/𝑤2) with four fitting parameters: (black) of empirical form 𝑉𝑏𝑔(𝑥) = 𝑉𝑏𝑔 𝐶𝑁𝑃 is the 𝑉𝑏𝑔 value corresponding to the energy 𝐸𝐿𝑆 of the LS, 𝑉𝑝 describes the height of the bell 𝑉𝑏𝑔 shape determined by 𝑉𝑡𝑔, 𝑤 is the width of the bell shape governed by the scanning height ℎ of the tSOT, and 𝑥𝑖 is the location of the LS along the edge. 𝐿𝑆 − 𝑉𝑏𝑔 𝐿𝑆 − 𝑉𝑏𝑔 Our numerical algorithm detected 𝑚 = 135 traces along 𝐿 = 3.5 µm long graphene edge which reflects only a fraction of the total number of LSs. The average density of the identified LSs is 𝜌 = 𝑚/𝐿 = 38.6 µm-1 corresponding to average distance of 〈∆𝑥〉 = 26 nm. Analysis of the distances ∆𝑥𝑛𝑛 between the nearest neighbors (Fig. S15C) reveals that the probability density of ∆𝑥𝑛𝑛 is described by Poisson distribution 𝑃(∆𝑥𝑛𝑛) = 𝜌𝑒−∆𝑥𝑛𝑛𝜌 (black curve) indicating that LSs are distributed randomly, excluding presence of clustering or long-range spatial correlations. Interestingly, Fig. S15E shows ∆𝑥𝑛𝑛 separations of down to below 1 nm putting an upper bound on the spatial extent of the individual LSs and emphasizing the atomic- scale nature of the defects. Figure S15D presents the 𝑉𝑏𝑔 𝐿𝑆 values of the different LSs that shows a very wide distribution with no apparent correlations. Such a large random variability over very short length-scales is inconsistent with the 𝐿𝑆 result from formation of "puddles" due to long-range substrate potential possibility that the variations in 𝑉𝑏𝑔 𝐿𝑆 values appear to be uniformly disorder (21) and support the Coulomb repulsion scenario. Moreover, the 𝑉𝑏𝑔 distributed, without visible clustering around specific values that may characterize particular adatom defects. Chemical variability of adatoms is therefore not a likely explanation of the large variability of the energy of 𝐿𝑆 (Fig. S15E) reaches ±15 V, limited by the LSs. The difference in 𝑉𝑏𝑔 our accessible 𝑉𝑏𝑔 span, and seems to be independent of the distance ∆𝑥𝑛𝑛 between the neighbors from sub nm to 100 nm distances. Note, however, that we analyze only a fraction of the defects. Remarkably, we 𝐿𝑆 detect neighboring defects that are separated on nm scale while showing variation of up to 10 V in their 𝑉𝑏𝑔 energy. Examples of such pairs of nearest neighbors are marked in color in Fig. S15B. The green traces show an example of five LSs within a segment of 8 nm along the graphene edge with significantly different energies. 𝐿𝑆 between the nearest neighbors ∆𝑛𝑛𝑉𝑏𝑔 24  Supplementary Figures Figure S1. Magneto-transport characterization of the graphene sample. Four-probe measurement of 𝑅𝑥𝑥 of the top constriction of the sample in Fig. 1B vs. back gate voltage 𝑉𝑏𝑔 and the perpendicularly applied magnetic field 𝐻 at 𝑇 = 4.2 K, showing quantum Hall oscillations. A linear fit to the QH minima results in back gate capacitance of 𝐶𝑏𝑔 = 2 ∙ 10−4 F/m2. 25 Vbg[V]μ0H[T]Change colormap Figure S2. tSOT characterization. (A) SEM image of the tSOT used in this work showing two Pb leads connecting to the Pb ring at the apex of the tip and the gap between them. (B) SEM image of the tSOT pipette attached to one tine of the quartz tuning fork. (C) I-V characteristics of the tSOT at 4.2 K and various applied magnetic fields. The inset shows a simplified measurement circuit which includes the bias source 𝑉𝐵𝐼𝐴𝑆, bias resistors 𝑅𝑏1 = 𝑅𝑏2 = 11 k, shunt resistor 𝑅𝑠 = 5.4 , parasitic resistance 𝑅𝑝 = 0.6 , and the inductively coupled SSAA. (D) Quantum interference pattern of the critical current 𝐼𝑐(𝐻) showing a period of 2.4 T corresponding to effective diameter 𝑑 = 33 nm of the tSOT. The higher lobes of the interference pattern are suppressed by 𝐻𝑐2 of Pb. 26 Rb1Rb2RsVBIASSSAAItSOTRPtSOTVtSOTVbias_wp=-3Vnoise=9.7908 pA/sqrt(Hz)thermal response is 19.7572 uA/Kthermal noise is 0.49556 uK/sqrt(Hz)μ0H [T]Ic[μA]D50 nmAPbPbgap50 μmBtuning forktSOTVtSOT[μV]ItSOT[μA]C Figure S3. Demonstration of simultaneous measurement by three thermal imaging methods. 𝑇𝐹(𝑥, 𝑦) image of the left part of Sample 2 revealing the dissipation rings along the edges of (A) 𝑇𝑎𝑐 hBN/graphene/hBN heterostructure (dashed). The arrow indicates the current 𝐼𝑑𝑐 (chopped at 35.5 Hz) that flows from the right side of the sample into the constriction. (B) Zoomed-in thermal image 𝛿𝑇(𝑥, 𝑦) of a 𝑇𝐹(𝑥, 𝑦) dissipation ring marked by dashed square in (A). (C) Simultaneously measured ac thermal image 𝑇𝑎𝑐 𝑡𝑔(𝑥, 𝑦) due due to 𝑥𝑎𝑐 tip oscillation by the TF, 𝑇𝑎𝑐 𝑎𝑐𝑑𝑇(𝑥, 𝑦)/𝑑𝑉𝑡𝑔. Scan parameters: (A) Scan area 1.4  3.9 µm2, to 𝑉𝑡𝑔 pixel size 13 nm, scan-speed 200 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = -2 V, 𝑉𝑡𝑔 = 2 V, 𝐼𝑑𝑐 = 2 µA rms chopped at 35.5 Hz, 𝑥𝑎𝑐 = 1.4 nm. (B-D) Scan area 400  410 nm2, pixel size 3.3 nm, scan-speed 250 ms/pixel, ℎ = 20 nm, 𝑎𝑐 = 0.1 V at 𝑉𝑏𝑔 = 0 V, 𝑉𝑡𝑔 = 5 V, 𝐼𝑑𝑐 = 1 µA rms chopped at 35.5 Hz, 𝑥𝑎𝑐 = 1.5 nm at 4.5 to the 𝑥 axis, 𝑉𝑡𝑔 3.06 kHz. 𝑇𝐹(𝑥, 𝑦) = 𝑥𝑎𝑐𝑑𝛿𝑇(𝑥, 𝑦)/𝑑𝑥. (D) The ac thermal image 𝑇𝑎𝑐 𝑎𝑐 tip gate modulation, 𝑇𝑎𝑐 𝑡𝑔(𝑥, 𝑦) = 𝑉𝑡𝑔 27 δT[mK]B40 nm𝑇𝑎𝑐𝑇𝐹[µK]C𝑇𝑎𝑐𝑡𝑔[μK]DA500 nm𝑇𝑎𝑐𝑇𝐹[µK]Idc Figure S4. Measurement of the tSOT vibration amplitude. (A) Optical image of Sample 2 showing the etched hBN/graphene/hBN heterostructure (bright) on SiO2 substrate (dark). A dc current 𝐼𝑑𝑐 = 4 μA rms chopped at 35.5 Hz was applied as indicated by the arrows. (B) Thermal 𝛿𝑇(𝑥, 𝑦) image of the constriction region marked by the dashed square in (A). (C,D) Numerical derivatives 𝜕𝛿𝑇 𝜕𝑥⁄ of image (B). (E) Best fit linear combination of (C) and (D), 𝑇𝐹(𝑥, 𝑦) image (F) providing the tSOT vibration of 𝑥𝑎𝑐 𝜕𝛿𝑇(𝑥, 𝑦) 𝜕𝑥⁄ + 𝑦𝑎𝑐 𝜕𝛿𝑇(𝑥, 𝑦) 𝜕𝑦⁄ , to the measured 𝑇𝑎𝑐 𝑇𝐹(𝑥, 𝑦) measured simultaneously with (B) at TF excitation 𝑉𝑇𝐹 = 𝑥𝑎𝑐 = 4.05 nm and 𝑦𝑎𝑐 = 0.31 nm. (F) 𝑇𝑎𝑐 2 vs. 𝑉𝑇𝐹 (blue dots) and a linear 1.66 mV. (G) The measured rms tSOT vibration amplitude 𝑟𝑎𝑐 = √𝑥𝑎𝑐 fit (red) with slope of 2.41 nm/mV. Scan parameters of (B,F): Scan area 22 µm2, pixel size 20 nm, scan-speed 80 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = -2 V, 𝑉𝑡𝑔 =-0.53 V corresponding to 𝑉𝑡𝑔 𝐹𝐵 at this 𝑉𝑏𝑔. and 𝜕𝛿𝑇 𝜕𝑦⁄ 2 + 𝑦𝑎𝑐 28 400 nmBδT[mK] δ D400 nmF400 nm𝑇𝑎𝑐𝑇𝐹[µK]E400 nmrac∙ δ [µK] δ C400 nm𝜕𝜕𝑥𝜕𝜕𝑦rac[nm]VTF[mV]GIdc500 nmAIdcver2 Figure S5. Electron-phonon cooling power spectrum of a single defect. Cooling power spectrum of a quasi-bound state,𝑃𝑒𝑝(𝜀) , calculated from Eq. (S11) for three indicated values of 𝜀𝐿𝑆 (left axis) and the experimental 𝛿𝑇(𝜀) ∝ 𝑃𝑒𝑝(𝜀) of defect 'C' (right axis) as a function of the Fermi energy 𝜀. Parameters of Eq. (S11): 𝑇𝑝 =4.2 K, 𝑇𝑒 =5.2 K, ln(𝑊 𝑣𝑘𝑝 )=5.6. ⁄ 29 𝛿𝑇𝜀[µK] ∝ 𝑃𝑒𝑝𝜀𝑃𝑒𝑝𝜀[pW]𝜀[meV]𝜀𝐿𝑆[meV] Figure S6. Edges as dominant source of dissipation in hBN/graphene/hBN. (A) Thermal image 𝛿𝑇(𝑥, 𝑦) of Sample 2 at tSOT flat-band condition as described in Figs. 4F,G but at 𝑉𝑏𝑔 = -6 V revealing dissipation along the graphene edges. Scan area 4.7  3.7 µm2, pixel size 50 nm, scan-speed 200 𝐹𝐵 at this 𝑉𝑏𝑔. (B) ms/pixel, ℎ = 20 nm, 𝐼𝑑𝑐 = 1 μA rms chopped at 35.5 Hz, 𝑉𝑡𝑔 =-1.08 V corresponding to 𝑉𝑡𝑔 Temperature profiles taken along the vertical dashed lines comparing Joule-heating simulation (C – red), uniform bulk heating simulation (D – cyan), edge point sources with diffusion model (E – blue), edge point sources with ballistic 2D phonons (F – black), and the experimental profile (A – green). (C-E) 3D heat flow simulations of surface temperature of graphene flake on SiO2 substrate shaped as in experiment for three cases of heat sources. (C) Joule heating: A current of 1 μA is injected into the graphene through the right constrictions as in experiment. A sheet resistance of 180 /□ is assigned to the graphene and the current density is solved. The current density is used as a heat source (resulting in a total heat load of 1.8 nW) to solve for the 3D heat flow problem. (D) Surface temperature from the solution of a heat flow problem with uniform power density dissipated in the graphene with the same total power as in (C). (E) Surface temperature from the solution of a diffusive heat flow problem with heat injected at discrete points (blue) along the edges with the same total power as in (C). (F) Surface temperature attained by superposition of thermal profiles of the form 𝑇(𝑟) ∝ (1 + (𝑟/ℓ)2)−1/2 around each marked point (blue) representing the case of 2D ballistic phonons emitted from point sources along the sample edges. 30 ExperimentNumerical simulations500 nmDδT[μK]500 nmCδT[μK]δT[μK]500 nmEδT[μK]500 nmF500 nmVbg=-6 VAδT[μK]Joule diss. (sim.)Uniform bulk (sim.)Edge diss. diffusion (sim.)Experimenty[μm]δT[μK]BEdge diss. ballistic (sim.) 𝒂𝒄 . Figure S7. Resolving resonant dissipation sources by ac back-gate voltage modulation 𝑽𝒃𝒈 𝑎𝑐 = 0.1 V rms at 322 Hz is applied to the back gate in addition to 𝑉𝑏𝑔 𝑏𝑔(𝑥, 𝑦) in presence of 𝐼𝑑𝑐 = 6 µA injected as indicated in Fig. 4F. An ac modulation (A) Thermal image 𝑇𝑎𝑐 𝑑𝑐 = –8 V, 𝑉𝑡𝑔 =-1.35 V corresponding 𝑉𝑏𝑔 𝐹𝐵 at this 𝑉𝑏𝑔. The to 𝑉𝑡𝑔 bright and dark spots reveal the LSs that are shifted in and out of resonance by the back gate modulation 𝑉𝑏𝑔 𝑎𝑐. Scan area 4.6  3.7 µm2, pixel size 40 nm, scan-speed 60 ms/pixel, ℎ = 20 nm. 𝐹𝐵 at this 𝑉𝑏𝑔. (B) Same as (A) at 𝑉𝑏𝑔 𝑑𝑐 = –6.7 V, 𝑉𝑡𝑔 =-1.17 V corresponding to 𝑉𝑡𝑔 31 B𝑇𝑎𝑐𝑏𝑔[μK]A𝑇𝑎𝑐𝑏𝑔[μK]500 nm𝑉𝑏𝑔𝑑𝑐=-8 V𝑉𝑏𝑔𝑑𝑐=-6.7 V Figure S8. Revealing dissipation from individual LSs by tip gating. (A) Thermal image 𝛿𝑇(𝑥, 𝑦) at flat-band condition in presence of 𝐼𝑑𝑐 = 1 μA rms chopped at 35.5 Hz injected through the two right constrictions as indicated in Fig. 4F. Scan area 4.7  3.9 µm2, pixel size 25 nm, scan- 𝑇𝐹(𝑥, 𝑦) speed 200 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = –2 V, 𝑉𝑡𝑔 = -0.53 V corresponding to 𝑉𝑡𝑔 thermal image away from flat-band condition revealing the individual resonant LSs at graphene edges in form of rings. The faint rings at the bottom of the image arise from the graphene edges in the adjacent chamber (see Fig. S4A). Scan area 4.7  3.9 µm2, pixel size 20 nm, scan-speed 60 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = –2 V, 𝑉𝑡𝑔 = 10 V, 𝐼𝑑𝑐 = 4 μA rms chopped at 35.5 Hz. 𝐹𝐵 at this 𝑉𝑏𝑔. (B) 𝑇𝑎𝑐 32 500 nmAδT[µK]500 nmB𝑇𝑎𝑐𝑇𝐹[µK]x3Vbg=-2 V, 𝑉𝑡𝑔=-0.53 V (𝑉𝑡𝑔𝐹𝐵)Vbg=-2 V, 𝑉𝑡𝑔=10 V Figure S9. Current dependence of the thermal map at flat-band condition. (A) Thermal image 𝛿𝑇(𝑥, 𝑦) in presence of 𝐼𝑑𝑐 = 120 nA rms chopped at 35.5 Hz injected through the two right constrictions as indicated in Fig. 4F (the dotted line outlines the graphene edges). Scan area 4.7  3.7 µm2, pixel size 60 nm, scan-speed 600 ms/pixel, ℎ = 20 nm, 𝑉𝑏𝑔 = 0.33 V close to the CNP, 𝑉𝑡𝑔 =-0.16 V 𝐹𝐵 at this 𝑉𝑏𝑔. (B) 𝛿𝑇 profiles along the white dashed line in (A) at several indicated corresponding to 𝑉𝑡𝑔 values of 𝐼𝑑𝑐. (C) Current dependence of the average temperature 〈𝛿𝑇〉 along the profiles (blue) and a quadratic fit 〈𝛿𝑇〉 ∝ 𝐼𝑑𝑐 2 (red). 33 BCδT[μK]500 nmVbg=0.33 VAX2 Figure S10. Numerical simulations of tip-induced potentials and spectroscopy. (A) Band bending 𝐸𝐷(𝑥) induced in graphene by the tSOT located at 𝑥 = 0 at potential 𝑉𝑡𝑔 = –1.5 V for various back gate voltages 𝑉𝑏𝑔. (B) Band bending 𝐸𝐷(𝑥) in graphene at 𝑉𝑏𝑔 = 1 V and different values of 𝑉𝑡𝑔. (C) Resonant dissipation traces describing the lateral displacement 𝑥 of the tSOT from the defect at which resonant conditions for inelastic scattering are met as a function of 𝑉𝑡𝑔 for various values of 𝑉𝑏𝑔 depicting the experimental results in Fig. S11B. (D) Resonant dissipation bell-shaped traces as a function of 𝑉𝑏𝑔 for various values of 𝑉𝑡𝑔 describing the experimental results in Fig. 4B and Fig. S11E. 34 x[nm]Vbg[V]Vtg[V]DVtg[V]Vbg=1 Vx[nm]E[meV]BED(x)EFVbg[V]Vtg=-1.5 Vx[nm]E[meV]AED(x)EFx[nm]Vtg[V]Vbg[V]CVbg=1 V Figure S11. Thermal spectroscopy of a single localized state. (A) Map of 𝑇𝑎𝑐(𝑥) line scans through the center of defect 'C' upon varying 𝑉𝑡𝑔 at constant 𝑉𝑏𝑔 = − 1 V and ℎ = 20 nm. Pixel width 3 nm, pixel height 50 mV, scan-speed 60 ms/pixel. (B) The resulting resonance traces 𝐹𝐵. (C) Map of 𝑅(𝑉𝑡𝑔) for various values of 𝑉𝑏𝑔. The transition from convex to concave traces occurs at 𝑉𝑏𝑔 𝑇𝑎𝑐(𝑥) line scans as in (A) at 𝑉𝑏𝑔 = 2.5 V. (D) Map of 𝑇𝑎𝑐(𝑥) line scans showing the bell-shaped resonance trace at 𝑉𝑡𝑔 = −10 V over the full range of 𝑉𝑏𝑔 = ±10 V at ℎ = 12 nm, 𝑥𝑎𝑐 = 2.7 nm, and 𝐼𝑑𝑐 = 6 µA. The bright strip at 𝑉𝑏𝑔 ≈ 0 V arises from enhanced heating at the nearby constriction, the resistance of which becomes very large close to CNP. (E) The resulting fitted bell-shaped resonance traces 𝑅(𝑉𝑏𝑔) for various 𝐹𝐵. (F) Same as (D) at 𝑉𝑡𝑔 =10 V. The fact that only a single values of 𝑉𝑡𝑔 that change their polarity at 𝑉𝑏𝑔 resonance trace is observed in (D) and (F) shows that the LS has only one energy level in the accessible range of Vtg and Vbg. (G) The dissipation ring radius 𝑅(𝑉𝑏𝑔, 𝑉𝑡𝑔) of the LS over the entire range of tip and back gate voltages with the 𝑉𝑏𝑔 𝐹𝐵 point marked by black dot. 𝐿𝑆, 𝑉𝑡𝑔 𝐿𝑆, 𝑉𝑡𝑔 𝐿𝑆, 𝑉𝑡𝑔 35 Vbg=-1 VTac[μK]Vtg[V]x[nm]AVtg[V]x[nm]Vbg[V]𝑉𝑡𝑔𝐹𝐵BIs the parabola filppedin x?Is it really the right defect (there are features coming from the left in this sweep)Tac[μK]Vbg=2.5 VVtg[V]x[nm]CVtg=10 VFVbg[V]x[nm]Tac[μK]R[nm]Vtg[V]Vbg[V]GOccupied stateEmpty stateVtg[V]Vbg[V]x[nm]𝑉𝑏𝑔𝐿𝑆EDTac[μK]Vtg=-10 VVbg[V]x[nm] Figure S12. Numerical simulation of resonance lines of LS. (A) Band bending 𝐸𝐷(𝑥) induced in graphene by the tSOT located at 𝑥 = 0 at 𝑧𝑡𝑖𝑝 = 60 nm above the graphene for critical (𝑉𝑏𝑔, 𝑉𝑡𝑔) pairs such that 𝐸𝐷(𝑥 = 0) = 0 providing resonance conditions for inelastic scattering from LS with 𝜀𝐿𝑆 = 0 (all other simulation parameters are identical to Fig. S10). Inset: Schematic Dirac cone with LS aligned with 𝐸𝐹. (B) The resulting 𝑅 = 0 resonance curves for different 𝑧𝑡𝑖𝑝 heights. The 𝐿𝑆 (cross). (C) Derivative of the resonance curves intersect at flat-band conditions determined by 𝑉𝑡𝑔 𝐶𝑁𝑃. For 𝜀𝐿𝑆 = 0 the crossing point coincides curves showing the kink location determined by the local CNP, 𝑉𝑏𝑔 with the kinks. (D) Same as (A) for the case of 𝜀𝐿𝑆 = 30 meV. The resonance conditions are attained when 𝐸𝐷(𝑥 = 0) = −30 meV. (E,F) Same as (B,C) for the case of 𝜀𝐿𝑆 =30 meV. The kink and the crossing point are well separated and their difference, 𝑉𝑏𝑔 𝐶𝑁𝑃, provides a direct measure of 𝜀𝐿𝑆. 𝐹𝐵 and 𝑉𝑏𝑔 𝐿𝑆 − 𝑉𝑏𝑔 36 Vtg[V]𝑉𝑡𝑔𝐹𝐵𝑉𝑏𝑔𝐶𝑁𝑃𝑉𝑏𝑔𝐿𝑆EVtg[V]𝑉𝑡𝑔𝐹𝐵𝑉𝑏𝑔𝐶𝑁𝑃=𝑉𝑏𝑔𝐿𝑆BC𝑑𝑉𝑡𝑔𝑑𝑉𝑏𝑔⁄Vbg[V]FVbg[V]𝑑𝑉𝑡𝑔𝑑𝑉𝑏𝑔⁄Dx[nm]E[meV]Vtg[V]Vbg[V]EFεLS=30 meV-505ED(x)-404Ax[nm]E[meV]Vtg[V]Vbg[V]EFεLS=0 meV-505ED(x)0-44 Figure S13. Calculated and measured resonance lines of LS. (A) Calculated band bending 𝐸𝐷(𝑥) induced in graphene by the tSOT as in Fig. S12 for the case of 𝜀𝐿𝑆 = −20 meV. Inset: Schematic Dirac cone with LS aligned with 𝐸𝐹 at resonance conditions. (B,C) The resulting calculated resonance curves for different 𝑧𝑡𝑖𝑝 values (B) and the derivatives of the curves (C), showing the 𝐶𝑁𝑃. (D) Experimental resonance lines of defect 'C' extracted from intersection and the kink points 𝑉𝑏𝑔 the data of Fig. 2D and a similar measurement performed at ℎ = 20 nm above the hBN surface with otherwise the same measurement parameters, and their numerical derivatives (E). A kink is apparent at the 𝐶𝑁𝑃 = 0.48 V, defined by the minima in (E). Experimental curves intersect at local change neutrality point, 𝑉𝑏𝑔 𝑉𝑏𝑔 𝐹𝐵 = –0.26 V (indicated by the cross in (D)), resulting in 𝜀𝐿𝑆 = −22 meV. 𝐿𝑆 and 𝑉𝑏𝑔 𝐿𝑆 = 0.2 V and 𝑉𝑡𝑔 37 Vtg[V]D𝑉𝑡𝑔𝐹𝐵𝑉𝑏𝑔𝐿𝑆𝑉𝑏𝑔𝐶𝑁𝑃ExperimentVbg[V]E𝑑𝑉𝑡𝑔𝑑𝑉𝑏𝑔⁄SimulationVtg[V]B𝑉𝑡𝑔𝐹𝐵𝑉𝑏𝑔𝐿𝑆𝑉𝑏𝑔𝐶𝑁𝑃SimulationVbg[V]𝑑𝑉𝑡𝑔𝑑𝑉𝑏𝑔⁄Cx[nm]AE[meV]Vtg[V]Vbg[V]EFED(x)-404-505εLS=-20 meV Figure S14. Estimating the spectral width of the localized state. (A) Bell-shaped 𝑇𝑎𝑐 dissipation trace of defect 'B' at 𝑉𝑡𝑔 = −10 V, ℎ = 20 nm, 𝑥𝑎𝑐 = 2.7 nm, and 𝐼𝑑𝑐 = 3 µA. The slope of the trace 𝑑𝑉𝑏𝑔 𝑑𝑥⁄ (at 𝑉𝑏𝑔 = 5.3 V, green line) translates the width of the ring, 𝛿𝑥, into corresponding width 𝛿𝑉𝑏𝑔 and 𝛿𝐸 of the LS. (B) Width 𝛿𝑥 of the dissipation ring vs. the tip vibration rms amplitude 𝑥𝑎𝑐 at 𝐼𝑑𝑐 = 3 µA (red curve is a guide to the eye). (C) Width 𝛿𝑥 of the dissipation ring vs. 𝐼𝑑𝑐 at 𝑥𝑎𝑐 = 0.5 nm. (D) 𝑇𝑎𝑐 dissipation ring at 𝑉𝑡𝑔 = −10 V, 𝑉𝑏𝑔 = 5.3 V, ℎ = 20 nm, 𝑥𝑎𝑐 = 0.5 nm, and 𝐼𝑑𝑐 = 2 µA (Scan area 180  180 nm2, pixel size 1.2 nm, scan-speed 600 ms/pixel). (E) Several line-cuts in the rectangle in (C) (blue, x-axis parallel to long edge of the rectangle), and the average of the profiles (red) determining the width of the ring 𝛿𝑥 = 7 nm. (F) The corresponding 𝛿𝑇(𝜀) ∝ 𝑃𝑒𝑝(𝜀), with 𝛿𝐸 = 13 meV. 38 CIdc[µA]δx[nm]δx[nm]xac[nm]BTac[μK]x[nm]EδxDTac[μK]20 nmδxx[nm]Tac[μK]Vbg[V]δVbgδxA𝑑𝑉𝑏𝑔𝑑𝑥Pep(ε)∝δT(ε)[μK]ε[meV]FδE Figure S15. Spectroscopic thermal imaging of the localized edge states. (A) Map of 𝑇𝑎𝑐(𝑥) line scans along the bottom edge of the graphene sample (reproduced from Fig. 4D) vs. 𝑉𝑏𝑔 at 𝑉𝑡𝑔 = −10 V showing bell-shaped resonant dissipation traces of the individual LSs. (B) Numerical fits 𝐿𝑆 describe (black) to the bell-shaped traces of 135 clearly resolved LSs. The asymptotic values of the traces 𝑉𝑏𝑔 the energy 𝐸𝐿𝑆 of the LSs. The colored traces (red, blue and yellow) show examples of pairs of adjacent 𝐿𝑆. The green traces show an example of five LSs within 8 nm neighbors with significantly different energy 𝑉𝑏𝑔 interval. (C) Probability density of the nearest neighbor distances ∆𝑥𝑛𝑛 between the LSs (20 bins) and the 𝐿𝑆 values of the resolved states vs. their theoretical exponential decay for random distribution (black). (D) 𝑉𝑏𝑔 location 𝑥𝑖 along the edge showing large uncorrelated variation in the energy of the LSs. (E) The difference in 𝑉𝑏𝑔 𝐿𝑆 vs. the distance ∆𝑥𝑛𝑛 between the neighboring LSs. 𝐿𝑆 between the nearest neighbors ∆𝑛𝑛𝑉𝑏𝑔 39 Tac[μK]x[μm]Vbg[V]Axi[μm]𝑉𝑏𝑔𝐿𝑆[V]D nm C𝑃 [μm-1]------------------group #1: max dx=2.25 nm, max dVbg=4.1964 Vx0=0.61383 0.61608eps=1.463 -2.7334peaks=7.465 2.3854------------------group #2: max dx=0 nm, max dVbg=4.0735 Vx0=0.818 0.818eps=-5.742 -9.8154peaks=-1.2269 -5.9611------------------group #3: max dx=1.853 nm, max dVbg=8.261 Vx0=1.9269 1.9288eps=1.7184 -6.5427peaks=7.5259 -1.9513------------------group #4: max dx=7.794 nm, max dVbg=9.3296 Vx0=2.6362 2.6364 2.6395 2.6421 2.6439eps=-4.598 -9.2302 -6.5325 0.09942 -2.0848peaks=0.13822 -5.2973 -2.2774 5.5571 3.2037E xnn[nm] nn𝑉𝑏𝑔𝐿𝑆 [V]𝜌𝑒−∆𝑥𝑛𝑛𝜌Tac[μK]x[μm]Vbg[V]B Movie Captions Movie S1. Thermal imaging of dissipation in hBN encapsulated graphene Scanning nanothermometry movie of 𝑇𝑎𝑐(𝑥, 𝑦) of 44 µm2 hBN/graphene/hBN sample. The tSOT of 33 nm effective diameter is scanned at ℎ = 20 nm above the sample surface in presence of a dc current 𝐼𝑑𝑐 = 3 µA applied between the top and the bottom-right constrictions at back gate voltage 𝑉𝑏𝑔 = −2 V. The tSOT is attached to a tuning fork and oscillates at 36.9 kHz with rms amplitude 𝑥𝑎𝑐 = 2.7 nm parallel to the surface at an angle of 12° relative to the 𝑥 axis, providing sensitive 𝑇𝑎𝑐(𝑥, 𝑦) imaging. The movie shows a sequence of images as the tip potential 𝑉𝑡𝑔 is incremented from 0 to 9 V. Numerous dissipation rings appear along the graphene edges and expand as 𝑉𝑡𝑔 increases. In addition, three isolated dissipation rings form and expand inside the sample revealing rare atomic defects in the bulk of graphene. The bulk and edge defects form sharp localized states acting as point source emitters of phonons due to resonant inelastic scattering of electrons. Scan area 5.5  5 µm2, pixel size 18.3 nm, scan-speed 20 ms/pixel. Movie S2. Thermal imaging of resonant dissipation of three bulk localized states Scanning nanothermometry movie of 𝑇𝑎𝑐(𝑥, 𝑦) of the central part of hBN/graphene/hBN sample. Three resonant dissipation rings appear and expand upon increasing 𝑉𝑡𝑔 from 2 to 10 V revealing inelastic electron scattering from three individual atomic defects. The small differences in the onset potential of the rings can be ascribed to long-range substrate potential disorder modifying the local CNP. 𝑉𝑏𝑔 = −1 V, 𝐼𝑑𝑐 = 3 µA, ℎ = 20 nm, 𝑥𝑎𝑐 = 2.7 nm, scan area 3.5  2.8 µm2, pixel size 17.5 nm, scan-speed 20 ms/pixel. The slightly non- monotonic expansion of the rings is a result of fluctuations in the tSOT scanning height. Movie S3. Principle of spectroscopic thermal imaging of localized states Animation describing thermal spectroscopy of the dissipation process due to resonant inelastic scattering of electrons from a localized state. A single LS with an energy level at 𝐸𝐿𝑆 pinned to the Dirac energy 𝐸𝐷 is located at 𝑥 = 0 marked by a sharp peak in 𝑃𝑒𝑝(𝜀). The Fermi energy 𝐸𝐹 (dashed) is determined by the back gate voltage 𝑉𝑏𝑔. Electrons are injected from left at energy 𝐸𝑒 (red arrows). The potential 𝑉𝑡𝑔 of the moving tSOT induces local band bending of the graphene Dirac energy 𝐸𝐷(𝑥) (blue) which shifts the energy of the LS. Inelastic scattering of the electrons occurs when the tip-induced potential brings the LS into resonance, 𝐸𝐿𝑆 = 𝐸𝑒, resulting in phonon emission from the defect (red arcs) which is detected by the tSOT as a sharp 𝛿𝑇(𝑥) thermal signal (red curve) when positioned at 𝑥 = ±𝑅. In 2D thermal imaging this results in a ring of enhanced 𝛿𝑇(𝑥, 𝑦) with its center at the LS. The 𝛿𝑇𝑜𝑛(𝑥) curve (dashed) describes the local temperature that would be measured by the tSOT if the LS stayed at resonance constantly. The band bending curve 𝐸𝐷(𝑥) (blue) was calculated using the electrostatic simulations as described in Section 6. Movie S4. Spectroscopic dissipation traces of a single bulk localized state vs. 𝑽𝒕𝒈 Spectroscopic nanothermometry movie of 𝑇𝑎𝑐(𝑥) of defect 'C'. The tSOT is scanned along a line through the center of the LS upon varying 𝑉𝑡𝑔. The different frames show the evolution of the resonant dissipation trace 𝐿𝑆. 𝐼𝑑𝑐 = 3 µA, ℎ = 20 nm, pixel width 3 nm, pixel height 50 with 𝑉𝑏𝑔. The curvature of the trace flips at 𝑉𝑏𝑔 mV, scan-speed 60 ms/pixel. Movie S5. Spectroscopic bell-shaped dissipation traces of a single bulk localized state vs. 𝑽𝒃𝒈 Spectroscopic nanothermometry movie of 𝑇𝑎𝑐(𝑥) of defect 'C'. The tSOT is scanned along a line through the center of the LS upon varying 𝑉𝑏𝑔. The different frames show the evolution of the resonant trace upon 𝐹𝐵. The bright 𝑉𝑡𝑔-independent signal in a changing the tip potential 𝑉𝑡𝑔. The curvature of the trace flips at 𝑉𝑡𝑔 form of a strip around 𝑉𝑏𝑔 ≈ 0 V is the result of heating of the nearby constriction that peaks when 𝑉𝑏𝑔 reaches CNP of the constriction. 𝐼𝑑𝑐 = 3 µA, ℎ = 20 nm, pixel width 5 nm, pixel height 30 mV, scan-speed 20 ms/pixel. 40  Movie S6. Spectroscopic bell-shaped dissipation traces of edge localized states vs. 𝑽𝒃𝒈 Spectroscopic nanothermometry movie of 𝑇𝑎𝑐(𝑥) of edge LSs. The tSOT is scanned along the bottom edge of the graphene upon varying 𝑉𝑏𝑔, revealing numerous resonant bell-shaped dissipation traces. The different frames show the evolution of the resonant traces with 𝑉𝑡𝑔. The resonant dissipative edge states are 𝐿𝑆. 𝐼𝑑𝑐 = 3 µA, ℎ = 20 nm, pixel width 4 nm, pixel observed at all values of 𝑉𝑏𝑔 with very large variations in 𝑉𝑏𝑔 height 100 mV, scan-speed 60 ms/pixel. High-pass filtering was applied to emphasize the pertinent bell- shaped resonance traces. Movie S7. Spectroscopic dissipation traces of edge localized states vs. 𝑽𝒕𝒈 Spectroscopic nanothermometry movie of 𝑇𝑎𝑐(𝑥) of edge LSs. The tSOT is scanned along the bottom edge of the graphene upon varying 𝑉𝑡𝑔, revealing numerous resonant dissipation traces. The different frames show the evolution of the resonant traces with 𝑉𝑏𝑔. The resonant edge states are observed at all values of 𝑉𝑏𝑔 with 𝐿𝑆 at which the curvature of trace originating from a given LS flips. 𝐼𝑑𝑐 = 3 µA, ℎ = very large variations in 𝑉𝑏𝑔 20 nm, pixel width 10 nm, pixel height 100 mV, scan-speed 20 ms/pixel. 41 

1201.0267

1

1201

Multi-dimensional laser spectroscopy of exciton-polaritons with spatial light modulators

We describe an experimental system that allows one to easily access the dispersion curve of exciton-polaritons in a microcavity. Our approach is based on two spatial light modulators (SLM), one for changing the excitation angles (momenta), and the other for tuning the excitation wavelength. We show that with this setup, an arbitrary number of states can be excited accurately and that re-configuration of the excitation scheme can be done at high speed.

cond-mat.mes-hall

cond-mat

Multi-dimensional laser spectroscopy of exciton polaritons with spatial light modulators Institut fur Experimentalphysik, Universitat Innsbruck, Technikerstrasse 25, 6020 Innsbruck, Austria P. Mai, B. Pressl, M. Sassermann, Z. Voros, and G. Weihs Technische Physik, Physikalisches Institut, Wilhelm Conrad Rontgen Research Center for Complex Material Systems, Universitat Wurzburg, Am Hubland, D-97074 Wurzburg, Germany C. Schneider, A. Loffler, S. Hofling, and A. Forchel We describe an experimental system that allows one to easily access the dispersion curve of exciton- polaritons in a microcavity. Our approach is based on two spatial light modulators (SLM), one for changing the excitation angles (momenta), and the other for tuning the excitation wavelength. We show that with this setup, an arbitrary number of states can be excited accurately and that re-configuration of the excitation scheme can be done at high speed. Cavity polaritons are the coherent superpositions of a confined electric field, and some excitation of the solid, be it a phonon, plasmon, or exciton. Amongst these quasi- particles, owing to both possible applications and their role in fundamental research, exciton-polaritons in a pla- nar microcavity are perhaps the most extensively studied. These quasi-particles are relatively short-lived (their lifetime is of the order of several ps), and they can de- cay by the emission of a photon. From the experimental point of view, one of the advantages of using a planar structure is that the momentum of the decaying polari- ton is mapped into the angle of the out-going photon. Conversely, by exciting the system at a particular angle and energy, a well-defined polariton state can be popu- lated at will. While this seems simple in theory, practice tends to be less straightforward. In order to excite, or to measure a specific polariton state, several approaches have been developed through the years. These involve either one or more goniome- ters [1 -- 3], a confocal setup [4, 5], or beam shifting [6]. Common to all these schemes is a mechanical means of moving the beams, which implies that re-configuration of the excitation geometry is slow and complicated. Fur- thermore, the complexity of the setup increases consid- erably, if more than one excitation beam is involved, and experiments become impractical for three beams. In this Letter, we would like to introduce a scheme that does not suffer from the above-mentioned difficul- ties. Instead of mechanically moving optical elements, we apply spatial light modulators (SLM), thereby eliminat- ing the main limiting factor. SLMs are special reflective or transmissive liquid crystal devices that can modulate either the phase, or amplitude of the impinging light, or both at the same time. SLMs have found many uses, both in fundamental, and applied research. They are extensively utilized as opti- cal tweezers in biological applications [7, 8], to beat the diffraction limit in optical imaging [9], or to create light with orbital angular momentum in quantum information experiments [10, 11], just to name a few examples. This latter approach has successfully been applied in polari- ton experiments to create a polariton superfluid with a FIG. 1. (color online) Experimental configuration: A phase- only SLM (SLM1) is used to displace the beam, thus to change the excitation momentum. SLM2 is an amplitude-only de- vice, which selects the excitation wavelengths in a standard 4-f pulse compressor. The half-wave plate (HWP) and the polarizing beam splitter (PBS1) are optional, and are used only when two-color excitation is required. The optical fibers are used in momentum-selective detection, either in reflection (F1), or transmission configuration (F2, F3). For clarity, not all split beams are shown. well-defined angular momentum [12, 13]. Our experimental setup is shown in Fig. 1. A colli- mated laser beam impinges on a phase-only spatial light modulator (SLM1) [14]. The beam expander only serves to make the illumination more uniform. On the SLM, we define a phase-mask equivalent to a lens (a parabolic phase profile), thereby focusing the beam to a 100-µm large spot on the entrance side of a microscope objective with working distance of 2.3 mm [15]. It is important to note that since the beam is focused on the objective, from the objective's perspective, the incoming beam can be approximated as a localized plane wave, which is, in turn, focused by the objective to an about 10-µm large spot on the sample. By moving the SLM's phase pro- file, the focal point can be shifted in the plane of the microscope objective, thus changing the excitation an- gle. With the 100-µm focal spot on the objective, we can achieve an angular selection of about 2 degrees. This translates into about 2 · 105 m−1 momentum selection in the total accessible polariton momentum range of about 2 First, we measured the reflectivity of such a microcav- ity sample as a function of the excitation angle. For this, we kept the laser wavelength fixed at λ = 770.5 nm. The laser beam is split by a beam splitter, (BS1) in Fig. 1. SLM1 focuses the passing component on the objective, while the other half serves as power reference. The beam reflected by the sample is then directed towards a large- area power meter, which replaces the fiber (F1) in Fig. 1. If the incoming laser beam is now scanned in the x − y plane (the plane of the back of the microscope objective), the reflected intensities will be reduced, if the correspond- ing momentum coincides with that dictated by the polari- ton dispersion, and the energy of the laser. This polariton resonance manifests itself as a circle of lower intensity in Fig. 2. The right hand side of the figure shows the inten- sity along the vertical line that passes through the origin of momentum space, and it clearly demonstrates the two resonances at k = ±2 · 106 m−1. By shifting the mask positions on SLM2, the mea- surement can be repeated for other energies, giving a three-dimensional map of the polariton dispersion. A two-dimensional projection of the dispersion of the lower polariton branch is shown in Fig. 3. The points were obtained by fitting a circular profile to the reflectivity minimum in figures similar to Fig. 2, and then taking the circles' coordinates at the leftmost, and rightmost points, respectively. In Fig. 3, the points are not com- pletely symmetric owing to the small variation in the cir- cles' center position. We then fitted the analytical lower polariton dispersion to the measured points with a Rabi splitting of ¯hΩRabi = 11 meV, cavity energy Ec = 1.612 meV, and exciton energy Ex = 1.6088 meV [19]. These FIG. 3. Lower polariton dispersion extracted from reflectiv- ity scans similar to those in Fig. 2. The solid line is a fit to the analytical lower polariton dispersion with parameters ¯hΩRabi = 11 meV, Ec = 1.612 meV, and Ex = 1.6088 meV [19]. The horizontal error was determined from the error of the fit to the circles on Fig. 2, while the error in the energy of the laser is less than 0.05 meV, the size of the dots in the figure. FIG. 2. (color online) Reflectivity scan of an AlGaAs/GaAs microcavity. The color range is relative to the maximum of the reflected intensity. The intensity cut-off at around 4.5 · 106 m−1 is due to the numerical aperture of the microscope objective. The right hand side shows the intensity along the vertical line passing through k = 0. 5 · 106 m−1 [16]. (In our case, the accessible momentum range is limited by the clear aperture of the objective, shown in Figs. 2, and 4.) When tuning of the laser's energy is required, we make the laser pass through a standard pulse-shaping configu- ration containing two 1200-line gratings, and a pair of spherical mirrors with a focal length of 500 mm (up- per half of Fig. 1). A laser beam with duration 180 fs is dispersed by the first grating (GR1), and then com- bined again with the second grating (GR2). In the com- mon focal plane of the spherical mirrors (M1, M2), an amplitude-only SLM (SLM2) with 128 pixels is located [17], and it acts as a re-configurable wavelength filter. The configuration above allows us to select an arbitrary wavelength within the bandwidth of laser pulse with a precision of about 0.1 nm. This setup can easily be ex- tended, in order to make two-color excitation schemes possible: instead of one, two slits are defined on SLM2, and the polarization of one of the colors is rotated by 90 degrees by an achromatic waveplate (HWP). In this way, after re-combining the beams on GR2, we can separate them on a polarizing beam splitter (PBS), so that the two colors can be steered independently. The waveplates WP1, and WP2, and the delay line in one of the beams are optional, and can be used for generating arbitrary lin- ear or circular polarization, and delays. We should point out that apart from translation stages required for adjust- ing the foci at the beginning of an experiment, the delay line is the only variable mechanical part in our setup. In order to prove the viability of our setup, we carried out various measurements on a AlGaAs/GaAs microcav- ity sample identical to that discussed in [18]. The sample consists of a λ/2 cavity with three sets of four quantum wells located at the antinodes of the confined field. -4-2024Wave vector [106 m1]-4-2024Wave vector [106 m
1]0.51.0-4-3-2-101234In-plane wave vector [106 m−1]1.6051.6071.609Energy [eV]3 (PLE) measurement. Again, the wavelength of the laser is kept constant at 770.2 nm, while the excitation mo- mentum is scanned. However, instead of detecting the reflected power, we measure the luminescence intensity at the center of the dispersion curve. For momentum se- lection on the detection side, we used a multimode fiber of diameter 62.5 µm, (F1 in Fig. 1), and fed the collected light into a spectrometer. Resonance behavior is apparent in Fig. 4, where we show the luminescence intensities as a function of the excitation momentum. The intensities are taken from spectra integrated over a range of 3 nm. The range was chosen such that that it suppresses the excitation laser's contribution, which, however, is still visible at normal incidence, and high momenta. Further suppression of the scattered laser light can be achieved by measuring in orthogonal polarization with respect to the excitation. In comparison with the reflectivity dip in Fig. 2, the PLE intensities in Fig. 4 give a much better defined res- onance, and we can easily fit a circle to the maxima. Repeating this procedure for various wavelengths, we get a three-dimensional representation of the polariton dis- persion, as depicted in Fig. 5. We would like to point out that our setup has tremen- dous advantages in terms of speed. Since the required phase masks can efficiently be calculated in the video card of a computer, and the amplitude masks do not have to be calculated, the displayed phases or amplitudes can be re-defined in 1/60 or 1/100 of a second, respectively. These numbers are determined by the maximum refresh rate of the SLMs. The two-dimensional scans in Fig. 2 (reflectivity) and in Fig. 4 (PLE), can be acquired in about 40 seconds. In conclusion, we introduced a versatile beam-steering system that can be used to easily access the complete po- lariton phase-space, even if an excitation scheme requires multiple momenta and energies. This approach should make it possible to easily, and simultaneously excite non- degenerate polariton states. Due to the technical difficul- ties mentioned in the introduction, non-degenerate exci- tation of polaritons is a largely unexplored field. Beyond its fundamental importance, it could be useful for the generation of entangled photons [20, 21]. We would also like to mention that the scheme that we outlined above can also be used for studying polaritons in confined ge- ometries, either in wires, or in micropillars. We acknowledge partial funding of this work by the Austrian Science Fund (FWF), Project P22979-N16. We are also indebted to D. Snoke for providing us with sam- ples at the initial stages of our work. FIG. 4. (color online) Luminescence intensity at k = 0 as a function of the excitation momentum. The color range is relative to the maximum of the luminescence intensity. On the right, a cut is shown along a vertical line through k = 0. The red circle with a radius of about k = 4.5· 106 m−1 is laser light scattered by the aperture of the microscope objective, while the small peak at k = 0 is the tail of the excitation laser's spectrum. FIG. 5. the measured polariton dispersion. (color online) Three-dimensional representation of values are in reasonable agreement with those reported earlier for identical samples. Next, we conducted a photoluminescence excitation [1] Ryan Balili, Bose-Einstein condensation of microcavity polaritons, Phd Thesis, Univeristy of Pittsburgh (2009) [2] P. G. Savvidis, J. J. Baumberg, R. M. Stevenson, M. S. Skolnick, D. M. Whittaker, J. S. Roberts, Phys. Rev. B 62 R13278 (2000) [3] S. Kundermann, T. Guillet, M. Saba, C. Ciuti, O. El Daıf, J. L. Staehli and B. Deveaud, Physica Status Solidi (a) 201, 381 [4] Wolfgang Langbein, Phys. Stat. Sol. b 242 2260 (2005) [5] S. Savasta, O. Di Stefano, V. Savona and W. Langbein, -4-2024Wave vector [106 m1]-4-2024Wave vector [106 m
1]0.00.51.0[106m-1]¡-2024[106m¢-1]¡-202Energy[eV]1.6061.6084Wave vectorWave vectorPhys. Rev. Lett. 94 246401 (2005) [6] R. Huang, F. Tassone and Y. Yamamoto, Phys. Rev. B 61 R7854 (2000) [7] Christian Maurer, Alexander Severin Furhapter, Stefan Bernet and Monika Ritsch-Marte, New Journal of Physics 9 78 (2007) Jesacher, [8] David G. Grier, Nature 424 810 (2003) [9] E. G. van Putten, D. Akbulut, J. Bertolotti, W. L. Vos, A. Lagendijk and A. P. Mosk, Phys. Rev. Lett. 106 193905 (2011) [10] Alipasha Vaziri, Gregor Weihs, Anton Zeilinger, Journal of Optics B: Quantum and Semiclassical Optics 4 S47 (2002) [11] B. Jack, J. Leach, J. Romero, S. Franke-Arnold, M. Ritsch-Marte, S. M. Barnett and M. J. Padgett, Phys. Rev. Lett. 103 083602 (2009) [12] D. N. Krizhanovskii, D. M. Whittaker, R. A. Bradley, K. Guda, D. Sarkar, D. Sanvitto, L. Vina, E. Cerda, P. Santos, K. Biermann, R. Hey and M. S. Skolnick, Phys. Rev. Lett. 104 126402 (2010) 4 [13] K. Lagougadis, private communication [14] PLUTO phase-only modulator with a 1920 x 1080 pixels, and 15.36 x 8.64 mm size, www.holoeye.de [15] Zeiss Ld Plan Neofluar 63x 0.75, www.zeiss.com [16] The focal spot size and angular resolution are determined by several factors: a short focal length on the SLM would increase the angular resolution at the expense of the fi- nal spot size on the sample. At the same time, the focal length defined on the SLM has to be long enough, so that the various diffraction orders can be separated. [17] SLM-128 with 128 pixels, and 12.8 x 10 mm size, www.cri-inc.com [18] J. Bloch, T. Freixanet, J. Y. Marzin, V. Thierry-Mieg and R. Planel, Appl. Phys. Lett. 73 1694 (1998) [19] H. Deng, H. Haug and Y. Yamamoto, Rev. Mod. Phys. 82 1489 (2010) [20] T. C. H. Liew and V. Savona, Phys. Rev. A 84 032301 (2011) [21] S. Portolan, O. Di Stefano, S. Savasta and V. Savona, EPL (Europhysics Letters) 88 20003 (2008) 

1002.0685

1

1002

The effect of disorder within the interaction theory of integer quantized Hall effect

We study effects of disorder on the integer quantized Hall effect within the screening theory, systematically. The disorder potential is analyzed considering the range of the potential fluctuations. Short range part of the single impurity potential is used to define the conductivity tensor elements within the self-consistent Born approximation, whereas the long range part is treated self-consistently at the Hartree level. Using the simple, however, fundamental Thomas-Fermi screening, we find that the long range disorder potential is well screened. While, the short range part is approximately unaffected by screening and is suitable to define the mobility at vanishing magnetic fields. In light of these range dependencies we discuss the extend of the quantized Hall plateaus considering the "mobility" of the wafer and the width of the sample, by re-formulating the Ohm's law at low temperatures and high magnetic fields. We find that, the plateau widths mainly depend on the long range fluctuations of the disorder, whereas the importance of density of states broadening is less pronounced and even is predominantly suppressed. These results are in strong contrast with the conventional single particle pictures. We show that the widths of the quantized Hall plateaus increase with increasing disorder, whereas the level broadening is negligible.

cond-mat.mes-hall

cond-mat

The effect of disorder within the interaction theory of integer quantized Hall effect S. E. Gulebaglan1, G. Oylumluoglu2, U. Erkarslan2, A. Siddiki2,3 and I. Sokmen1 1Dokuz Eylul University, Physics Department, Tınaztepe Campus, 35100 Izmir, Turkey 2Mugla University, Physics Department, Faculty of Arts and Sciences, 48170-Kotekli, Mugla, Turkey 3Istanbul University, Faculty of Sciences, Physics Department, Vezneciler-Istanbul 34134, Turkey E-mail: [email protected] Abstract. We study effects of disorder on the integer quantized Hall effect within the screening theory, systematically. The disorder potential is analyzed considering the range of the potential fluctuations. Short range part of the single impurity potential is used to define the conductivity tensor elements within the self-consistent Born approximation, whereas the long range part is treated self-consistently at the Hartree level. Using the simple, however, fundamental Thomas-Fermi screening, we find that the long range disorder potential is well screened. While, the short range part is approximately unaffected by screening and is suitable to define the mobility at vanishing magnetic fields. In light of these range dependencies we discuss the extend of the quantized Hall plateaus considering the "mobility" of the wafer and the width of the sample, by re-formulating the Ohm's law at low temperatures and high magnetic fields. We find that, the plateau widths mainly depend on the long range fluctuations of the disorder, whereas the importance of density of states broadening is less pronounced and even is predominantly suppressed. These results are in strong contrast with the conventional single particle pictures. We show that the widths of the quantized Hall plateaus increase with increasing disorder, whereas the level broadening is negligible. This work focuses on the disorder effects on the integer quantized Hall effect within the screening theory. Since the early days of QHE, disorder played a very important role, however, interactions were completely neglected. Here we present our results which also includes interactions in a self-consistent manner and show that even without localization one can obtain the quantized Hall plateaus. We investigated different aspects of the impurity potential and suggested a criterion on mobility at high magnetic fields. We think that our work will shed light on the understanding of the QHE and is interest to condensed matter community. Keywords: Article preparation, IOP journals Submitted to: J. Phys. C: Solid State Phys. Disorder in screening theory 1. Introduction 2 The integer quantized Hall effect (IQHE), observed at two dimensional charge systems (2DCS) subject to strong perpendicular magnetic fields B, is usually discussed within the single particle picture, which relies on the fact that the system is highly disordered [1, 2]. These quantized (spinnless) single particle energy levels are called the Landau levels (LLs) and the discrete energy values are given by EN = ωc(n + 1/2), where n is the Landau index and ωc = eB/m∗c is the cyclotron frequency of an electron with an effective mass m∗ (≈ 0.067me, me being the bare electron mass at rest) and c is the speed of light in vacuum. In single particle models the disorder plays several roles, such as Landau level broadening [3], leading to a finite longitudinal conductivity [4, 5], spatial localization [6] etc. Disorder can be created by inhomogeneous distribution of dopant ions which essentially generates the confinement potential [7] for the electrons. In the absence of disorder, the density of states are Dirac delta-functions D(E) = N =0δ(E − EN ), where l = (cid:112)/eB is the magnetic length, and the longitudinal (cid:80)∞ 1 2πl2 conductivity (σl) vanishes. For a homogeneous two dimensional electron system (2DES), by the inclusion of disorder and due to collisions, LLs become broadened. Therefore the longitudinal conductance becomes non-zero in a finite energy (in fact magnetic field) interval. Long range potential fluctuations generated by the disorder result in the so called classical localization [8], i.e. the guiding center of the cyclotron orbit moves along closed equi -- potentials [9]. In contrast to the above mentioned bulk theories, the edge theories usually disregard the effect of disorder to explain the (quantized) Hall resistance RH and accompanying (zero) longitudinal resistance RL. However, the non-interacting edge theories still require disorder to provide a reasonable description of the transition between the plateaus. The Landauer-Buttiker approach (known as the edge channel picture) [10] and its direct Coulomb interaction generalized version, i.e. the non-self- consistent Chklovskii picture [11], also needs localization assumptions in order to obtain quantized Hall (QH) plateaus of finite width (see for a review e.g. Datta's book [12] and Ref. [9] for the estimates of plateau widths at the high disorder limit). In contrast to above discussions very recent experimental [13, 14, 15, 16] and theoretical [17, 18] results point the incomplete treatment of the disorder potential and scattering mechanisms. Fairly recent theoretical approaches [19], the QH plateaus are obtained by the inclusion of direct Coulomb interaction self-consistently [20] and the effect of the disorder was handled via conductivity tensor elements [21], however, the source of the disorder and its properties was left unresolved [22]. Whereas, the influence of potential fluctuations on the QH plateaus were discussed briefly [23, 24]. This work provides a systematic investigation of the disorder potential and its influence on the quantized Hall effect including direct Coulomb interaction. The investigation is extended to realistic experimental conditions in determining the widths of the quantized Hall plateaus. We, essentially study the effect of disorder in two distinct regimes, namely the short range and the long range. The short range part is included to the density of states (DOS), thereby influences the widths of the current carrying Disorder in screening theory 3 edge-states and the entries of the conductivity tensor. Whereas, the long range part is incorporated to the self-consistent calculations, which determines the extend of the plateaus in turn. In Sec 2 we introduce two types of single impurity potentials, namely the Coulomb and the Gaussian, and compare their range dependencies considering damping of the dielectric material. In the next step we discuss the screened disorder potential within a pure electrostatic approach, by considering an homogeneous two dimensional electron system (2DES) without an external magnetic field and show that the long range part is well screened, whereas the short range part is almost unaffected. Section 2.2 is devoted to investigate impurities numerically, where we solve the Poisson equation self-consistently in three dimensions. The numerical and analytical calculations are compared, considering the estimations of the disorder potential range and its variation amplitude. We finalize our discussion with Sec. 3, where we calculate the plateau widths under experimental conditions for different sample widths and mobilities. 2. Impurity potential The disorder potential experienced by the 2DES, resulting from the impurities has quite complicated range dependencies. Since, the potential generated by an impurity is (i) damped by the dielectric material in between the impurity and the plane where the 2DES resides (ii) is screened by the homogeneous 2DES depending on the density of states, which changes drastically with and without magnetic field. It is common to theoreticians to calculate the conductivities from single impurity potentials, such as Lorentzian [19], Gaussian [25] or any other analytical functions [26, 27]. However, the landscape of potential fluctuations is also important to define the actual mobility of the sample at hand, in particular in the presence of an external magnetic field. 2.1. Pure Electrostatics We first discuss the different range dependencies of the Coulomb and Gaussian donors, assuming open boundary conditions. Next, the effect of the spacer thickness on the disorder potential is discussed, namely the damping of the external (Coulomb) potential, and is compared with the Thomas-Fermi screening. The different damping/screening dependencies of the resulting potentials are discussed in terms of range. The Coulomb potential presents long range part, which leads to long range fluctuations due to overlapping if several donors are considered. Whereas, the Gaussian potential decays exponentially on the length scale comparable with the separation thickness. Since the Gaussian potential is relatively short ranged, no overlapping of the single donor potentials occur. Hence, the external potential experienced by the electrons can be approximated to a homogeneous potential fairly good. Thus one can conclude that approximating the total disorder potential by Gaussians is not sufficient to recover the long range part. Similar arguments are also found in the literature [6, 9, 24]. In order to overcome the difference observed at the long range potential fluctuations Disorder in screening theory 4 between the Coulomb and the Gaussian impurities, the following procedure is applied: First we calculate the total disorder potential considering many impurities then we perform a two-dimensional Fourier transformation of the Coulomb potential and make a back transformation keeping the first few momentum q components in each direction, hence only the long range part of the potential is left [24]. Then we add the long range part of the Coulomb potential to the potential created by donors, i.e. to the confinement potential. We take this as a motivation to simulate the short range part of the impurity potential by Gaussian impurities, and calculate the Landau level broadening and the conductivities, described within the self-consistent Born approximation (SCBA) [25]. Here we point to the effect of the spacer thickness on the impurity potential experienced in the plane of 2DES. It is well known from experimental and theoretical investigations that, if the distance between the electrons and donors is large, the mobility is relatively high and it is usually related with suppression of the short range fluctuations of the disorder potential. These results agree with the experimental observations of high mobility samples and are easy to understand from the z dependence of the Fourier expansion of the Coulomb potential, (cid:90) V (cid:126)q(z) = d(cid:126)re−i(cid:126)q·(cid:126)r N(cid:88) j (cid:112)((cid:126)r − (cid:126)rj)2 + z2 e2/¯κ = 2πe2 ¯κq e−qzN S((cid:126)q), (1) where S((cid:126)q) contains all the information about the in-plane donor distribution and N is the total number of the ionized donors. We observe that if the spacer thickness is increased, the amplitude of the potential decreases rapidly. We also see that the short range potential fluctuations, which correspond to higher order Fourier components, are suppressed more efficiently. Next, we discuss electronic screening of the external potential created by the donors discussed above. For a dielectric material the relation between the external and the screened potentials are given by, scr = V q V q ext/
(q), (2) where
(q) is the dielectric function and is given by
(q) = 1 + 2πe2D0 , with the constant ¯κq 2D density of states D0 = m/(π2) in the absence of an external B field, and is known as the Thomas-Fermi (TF) function. The simple linear relation above, together with the TF dielectric function essentially describes the electronic screening of the Coulomb potential given in Eq. 1, if there are sufficient number of electrons [9] (nel > 0.1 · 1015 m−2). Consider a case where the q component approaches to zero, then the external (damped) potential is well screened, hence the long range part of the disorder potential. Whereas, the short range part remain unaffected, i.e. high q Fourier components. Now we turn our attention to the second type of impurities considered, the Gaussian ones. As well known, the Fourier transform of a Gaussian is also of the form of a Gaussian, therefore, similar arguments also hold for this kind of impurity. We should emphasize once more the clear distinction between the effect of the spacer on the external potential and the screening by the 2DES, i.e. via
(q). The Disorder in screening theory 5 Figure 1. Schematic representation of the crystal, which we investigate numerically. The crystal is grown on a thick GaAs substrate, where the 2DES is formed at the interface of the AlGaAs/GaAs hetero-junction. The top AlGaAs layer is doped with Silicon 30 nm above the interface. The crystal is spanned by a 3D matrix (128 × 128 × 60). former depends on the Fourier transform of the Coulomb potential and the important effect is the different decays of the different Fourier components (see Eq. 1), so that the short range part of the disorder potential is well dampened, whereas the latter depends on the relevant DOS of the 2DES and the screening is more effective for the long range part. We continue our investigation by solving the 3D Poisson equation iteratively for randomly distributed single impurities, where three descriptive parameters (i.e. the number of impurities, the amplitude of the impurity potential and the separation thickness) are analyzed separately. Next, we discuss the long range parts of the potential fluctuations investigating the Coulomb interaction of the 2DES, numerically. The range is estimated from these investigations by performing Fourier analysis and is related to the samples used in experiments [15, 16] (Sec. 4). 2.2. 3D simulations In the previous section we took a rather simple way to study the effect of interactions by assuming an homogeneous 2DES and screening is handled by the TF dielectric function. Here, we present our results obtained from a rather complicated numerical method. We solve the Poisson equation in 3D starting from the material properties of the wafer at hand, the typical material we consider is sketched in Fig. 1. Namely, using the growth parameters, we construct a 3D lattice where the potential and the charge distributions are obtained iteratively assuming open boundary conditions, i.e. V (x → ±∞, y → ±∞, z → ±∞) = 0. For such boundary conditions, we chose a lattice size which is considerably larger than the region that we are interested in. We preserve the above conditions within a good numerical accuracy (absolute error of 10−6). A forth order grid approach [28] is used to reduce the computational time, which is successfully used to describe similar structures [29]. Figure 1 presents the schematic drawing of the hetero-structure which we are interested in. The donor layer is δ− doped by a density of 3.3×1016 m−2 (ionized) Silicon Disorder in screening theory 6 Figure 2. (a) Electron density fluctuation considering 3300 impurities 30 nm above the electron gas. (b) The long-range part, arrows are to guide the distance between two maxima. The calculation is repeated for 50 random distributions, which lead to a similar range. Vimp zD atoms, ∼ 30 nm above the 2DES, which provide electrons both for the potential well at the interface and the surface. It is worthwhile to note that most of the electrons (∼ %90) escape to the surface to pin the Fermi energy to the mid-gap of the GaAs. In any case, for such wafer parameters there are sufficient number of electrons (nel (cid:38) 3.0× 1015 m−2) at the quantum well to form a 2DES. To investigate the effect of impurities we place positively charged ions at the layer where donors reside. From Eq. 1 we estimate the amplitude of the potential of a single impurity to be e2 = 0.033 eV and assume that κ some percent of the ionized donors are generating the disorder potential, that defines the long range fluctuations. In our simulations we perform calculations for a unit cell with areal size of 1.5 µm×1.5 µm which contains 3.3×1016 donors per square meters, thus with 10 percent disorder we should have NI ∼ 3300 impurities. Figure 2a depicts the actual density distribution, when considering 3300 impurities, whereas Fig. 2b presents only the long range part of the density fluctuation. The arrows show the average distance between two maxima, which is calculated approximately to be 550 nm. To estimate an average range of the disorder potential, we repeated calculations for such randomly distributed impurities, where number of repetitions scales with NI. Such a statistical investigation, sufficiently ensembles the system to provide a reasonable estimation of the long range fluctuations. We also tested for larger number of random distributions, however, the estimation deviated less than tens of nanometers. We show our main result of this section in Fig. 3, where we plot the estimated long range part of the disorder potential considering various number of impurities NI and impurity potential amplitude Vimp. Our first observation is that the long range part of the total potential becomes less when NI becomes large, not surprisingly. However, the range increases nonlinearly while decreasing NI, obeying almost an inverse square law and tend to saturate at highly disordered system. When fixing the distributions and NI, and changing the amplitude of the impurity potential we observe that for large amplitudes the range can differ as √ Disorder in screening theory 7 Figure 3. Statistically estimated range of the density fluctuations as a function of number of impurities, considering various impurity strengths (a) and spacer thicknesses (b). The calculations are done at zero temperature considering Coulomb impurities. The long range potential fluctuations become larger than the size of the unit cell if one considers less than %5 disorder. large as 200 nm at all impurity densities. We found that for impurity concentration less than %3, the range of the potential is larger than the unit cell we consider, i.e R > 1.5µm. In contrast to the long range part, the short range part is almost unaffected by the impurity concentration, however, is affected by the amplitude. Therefore, while defining the conductivities we will focus our investigation on Vimp. Another important result is that the estimates of long range fluctuations does not depend strongly on the spacer thickness, if one keeps the amplitude of single impurity potential amplitude fixed, Fig. 3b. All of the above numerical observations coincide fairly good with our analytical investigations in the previous section. However, the range dependency on the impurity concentration cannot be estimated with the analytical formulas given. We should also note that, similar or even complicated numerical calculations are present in the literature [6, 7]. A indirect measure of the screening effects on the potential can also be inferred by capacitance measurements, supported by the above calculation scheme in the presence of external field [14]. Next section is devoted to investigate the widths of the quantized Hall plateaus utilizing our findings. We consider mainly two "mobility" regimes, where the long range fluctuations is at the order of microns (high mobility) and is at the order of few hundred nanometers, low mobility. However, the amplitude of the total potential fluctuations will be estimated not only depending on the number of impurities but also depending on the spacer thickness, range and amplitude of single impurity potential. 3. Quantized Hall plateaus The main aim of this section is to provide a systematic investigation of the quantized Hall plateau (QHP) widths within the screening theory of the IQHE [20], therefore here we summarize the essential findings of the mentioned theory. In calculating the Disorder in screening theory 8 QHPs one needs to know local conductivities, namely the longitudinal σl(x, y) and the transverse σH(x, y). To determine these quantities it is required to relate the electron density distribution nel(x, y) to the local conductivities explicitly. Here we utilize the SCBA [25]. However, the calculation of the electron density and the potential distribution including direct Coulomb interaction is not straightforward, one has to solve the Schrodinger and the Poisson equations simultaneously. This is done within the Thomas-Fermi approximation which provides the following prescription to calculate the electron density nel(x, y) = dED(E) 1 e(EF −V (x,y))/kBT + 1 , (3) (cid:90) (cid:90) 2e2 ¯κ where D(E) is the appropriate density of states calculated within the SCBA, where kB is the Boltzmann constant and T temperature. The total potential is obtained from V (x, y) = (4) and the Kernel K(x, y, x(cid:48), y(cid:48)) is the solution of the Poisson equation satisfying the boundary conditions to be discussed next. dxdyK(x, y, x(cid:48), y(cid:48))nel(x, y), In the following we assume a translation in variance in y-direction and implement the boundary conditions V (−d) = V (d) = 0 (2d being the sample width), proposed by Chklovskii et.al. [11], such a geometry allows us to calculate the Kernel in a closed form. Hence, Eqs. (3) and (4) forms the self-consistency. For a given initial potential distribution, the electron concentration can be calculated at finite temperature and magnetic field, where the density of states D(E) contains the information about the quantizing magnetic field and the effect of short range impurities. Here we implicitly assume that the electrons reside in the interval −b < x < b (where, dl = d−b/d is called the depletion length), and is fixed by the Fermi energy, i.e. the number of electrons, hence donors. As a direct consequence of Landau quantization and the locally varying electrostatic potential, the electronic system is separated into two distinct regions, when solving the above self-consistent equations iteratively: i) The Fermi energy equals to (spin degenerate) Landau energy and due to DOS the system illustrates a metallic behavior, the compressible region, ii) The insulator like incompressible region, where EF falls in between two consequent eigen-energies and no states are available [11, 30]. It is usual to define the filling factor ν, to express the electron density in terms of the applied B field as, ν = 2πl2nel. Since all the states below the Fermi energy are occupied the filling factor of the incompressible regions correspond to integer values (e.g. ν = 2, 4, 6...), whereas the compressible regions have non-integer values, due to partially occupied higher most Landau level. The spatial distribution and widths of these regions are determined by the confinement potential [11], magnetic field [31], temperature [32] and level broadening [19, 20]. For the purpose of the present work we fix the confinement potential profile by confining ourselves to the Chklovskii geometry and keeping the donor concentration (and distribution) constant. Moreover we perform our calculations at a default temperature given by kBT /E0 F is the F = 0.02, where E0 Disorder in screening theory 9 Figure 4. The Hall resistances versus magnetic field, calculated at default temperature and considering a 10 µm sample for different ranges of the single impurity potential. Inset depicts a larger B field interval, where ν = 4 plateau can also be observed. Fermi energy calculated for the electron concentration at the center of the sample and is typically similar to 10 meV. The next step is to calculate the global resistances, i.e. the longitudinal RL and Hall RH resistances, starting from the local conductivity tensor elements. Such a calculation is done within a relaxed local model that relates the current densities j(x, y) to the electric fields E(x, y), namely the local Ohm's law: j(x, y) = σ(x, y)E(x, y). (5) The strict locality of the conductivity model is lifted by an spatial averaging process [20] over the quantum mechanical length scales and an averaged conductivity tensor σ(x, y) is used to obtain the global resistances. It should be emphasized that, such an averaging process also simulates the quantum mechanical effects on the electrostatic quantities. To be explicit: if the widths of the current carrying incompressible strips become narrower than the extend of the wave functions, these strips become "leaky" which can not decouple the two sides of the Hall bar and back-scattering takes place. Therefore, to simulate the "leakiness" of the incompressible strips we perform coarse-graining over quantum mechanical length scales. Now let us relate the local conductivities with the local filling factors. Since the compressible regions behave like a metal within these regions there is finite scattering Disorder in screening theory 10 dl = 70 nm Rg = 10 nm 20 nm 40 nm 80 nm 0.050 2d= 2 µm 0.035 0.020 0.010 0.010 0.120 0.125 0.115 0.095 0.085 0.120 0.135 0.140 0.135 0.130 0.100 0.090 0.070 0.050 0.040 3 µm 5 µm 8 µm 10 µm leading to finite conductivity. In contrast, within the incompressible regions the back-scattering is absent, hence, the longitudinal conductivity (and simultaneously resistivity) vanishes. Therefore, all the imposed current is confined to these regions. The Hall conductivity, meanwhile is just proportional to the local electron density. The explicit forms of the conductivity tensor elements are presented elsewhere [20]. Having the electron density and local magneto-transport coefficients at hand, we perform calculations to obtain the widths of the quantized Hall plateaus utilizing the above described, microscopic model assisted by the local Ohm's law at a fixed external current I. Further details of the calculation scheme is reviewed in Ref. [23]. 3.1. Single impurity potentials: Level broadening and conductivities Since the very early days of the charge transport theory, collisions played an important role. Such a scattering based definition of conduction also applies for the system at hand, i.e. a two-dimensional electron gas subject to perpendicular magnetic field. Among many other approaches [33, 19, 27] the SCBA emerged as a reasonable model to describe the DOS assuming Gaussian impurities, considering short range scattering. A single impurity has two distinct parameters that represents the properties of the resulting potential, the range Rg (at the order of separation thickness) and the amplitude of the potential (in relevant units), (cid:101)Vimp. However, these two parameters are not enough to define the widths of the Landau levels (Γ), another important parameter is the number of the impurities, NI. In the previous section we have already investigated these three parameters in scope of potential landscape, now we utilize our findings to define the level widths and the conductivities. It is more convenient to write the single impurity potential of the form, (cid:101)Vimp πR2 g Vg(r) = exp (− r2 R2 g ). (6) NI(cid:101)V 2 (cid:113) 4NI(cid:101)V 2 3 Together with the impurity concentration, the relaxation time is defined as τ0 = impm∗ and in the limit of delta impurities (i.e. Rg → 0) the Landau level width Γ takes the form Γ = . It is useful to define the impurity strength parameter to investigate the effect of disorder by 2πl2 imp I = (Γ/ωc)2 = γ2 , (7) 2NI(cid:101)V 2 impm π3ωc Disorder in screening theory 11 dl = 150 nm Rg = 10 nm 20 nm 40 nm 80 nm 0.075 2d= 2 µm 0.055 0.035 0.020 0.015 0.140 0.160 0.180 0.180 0.175 3 µm 5 µm 8 µm 10 µm 0.140 0.150 0.150 0.130 0.120 0.125 0.120 0.095 0.070 0.060 Table 1. The ν = 2 plateau widths obtained at default temperature for two depletion lengths dl (left 75 nm, right 150 nm), while γI = 0.05 is fixed (defined in Eq. 7 and the related text below). The widths are given in units of ωc/E0 F = Ωc/E0 F . given in units of magnetic energy ωc = fix the magnetic energy at 10 T. eB m = Ωc and as a normalization parameter we At this point we would like to make a remark on the concepts short/long range impurities and short/long range potential fluctuations, which is commonly mixed. By short range impurity potential we mean that Rg (cid:46) l, however, by short range potential fluctuation a length scale of the order of 200 − 300 nm is meant. The long range impurity potential corresponds to Rg > l and long range potential fluctuation is of the order of micrometers. Thus, when considering short range impurities the potential fluctuations may be long range, if NI is not large (< %5 of the donor concentration). We have also observed that, the long-range potential fluctuations are more efficiently screened by the 2DES and their range can be at the order of 500 nm at most, when assuming large impurity concentration, i.e. NI > %10. In light of the above findings and formulation we now investigate the widths of the quantized Hall plateaus. Figure 4 presents the calculated Hall resistances at a fixed temperature for typical single impurity ranges. We observe that, when increasing Rg the plateau widths remain approximately the same, with a small variation, which is in contrast to the experimental findings, if the system is low mobility (small Rg ⇒ highly broadened DOS) the plateau i.e. are larger. In fact changing Rg from 10 nm to 20 nm should increase the zero B field mobility almost an order of magnitude, when fixing the other parameters (see e.g table I of Ref. [20]). The contradicting behavior is due to the fact that the levels become broader when increasing the single impurity range, therefore the incompressible strips become narrower, which results in a narrower plateau. However, the long range potential fluctuations are completely neglected, therefore the effect(s) of disorder on the quantized Hall plateaus cannot be described in a complete manner. To investigate the effect of the single impurity range we systematically calculated the plateau widths; table 1 depicts the calculated widths of the Hall plateaus considering different sample widths, depletion lengths, filling factors and Rg. One sees that the plateau widths are affected by the increase of impurity range, however, in a completely wrong direction, i.e. plateaus become narrower when decreasing the mobility. As we show in the next section, it is not sufficient to describe mobility only considering the range of a single impurity. Moreover, we also show that the other two parameters defining B = 0 mobility are either Disorder in screening theory 12 Figure 5. The calculated Hall resistances at default temperature assuming a 5 µm sample considering three characteristic value of broadening parameter. The lowest mobility (γI =0.3) shows the narrowest plateau. not important or behaves in the opposite direction when calculating the resistances. Next we investigate the effect of the remaining two parameters, (cid:101)Vimp and NI. However, these two parameters both effect the level width simultaneously, thereby the widths of the incompressible strips. Hence, one cannot to distinguish their influence on the QHPs separately. Typical Hall resistances are shown in Fig. 5 calculated at default temperature considering different impurity parameters. Similar to the range parameter, we observe that the plateau widths become narrower when the mobility is low, which also points that our single particle based level broadening calculations are not in the correct direction. Such a behavior is easy to understand, when we decrease the mobility either by increasing the impurity concentration or by the amplitude of the impurity potential, the Landau levels become broader due to collisions. This means that, both the energetic and spatial gap between two consequent levels is reduced, hence the resulting incompressible strips are also narrower and fragile even at low temperatures. A detailed investigation on the incompressible widths depending on impurity parameters are reported in Ref. [19]. It is known that if there exists an incompressible strip wider than the Fermi wavelength the system is in the quantized Hall regime [20], therefore, if the gap is reduced the incompressible strips are smeared, thus the quantized Hall plateau vanish. As a general remark on the single particle theories, we should note that such a Disorder in screening theory 13 reduced gap is also a gross problem for the non-interacting models [34, 35, 36], however, one can overcome this discrepancy by making localization assumptions [1]. Namely, one assumes that even within the broadened Landau levels there are states, which are localized, therefore electrons cannot contribute to transport. Hence, although the gap is small (levels are broad) these localized states serves as a reasonable candidate to explain the low mobility behavior. In the early days of IQHE it was a great challenge to describe and observe these localized states [3]. Recent experiments [37, 38, 15, 39] show clearly that, the localization assumptions are not relevant in all the cases, i.e. narrow and high mobility samples. Moreover, the universal behavior of the localization length dictated by these theories fail [40]. An explicit treatment of the activation energy [41] and critical exponents are left to an other publication. 3.2. Size effects on plateau widths Another important parameter in defining the plateau widths is the depletion length dl. The slope of the confinement potential close to the edges essentially determines the widths of the incompressible strips [11], which in turn determines the plateau widths. In Fig. 6 we show the ν=2 plateau calculated for two different depletion lengths, we see that for the larger depletion the plateau is more extended. Since, the larger the depletion is, the smoother the electron density is. Therefore, resulting incompressible strips are wider, hence the plateau. Such an argument will fail if one considers a highly disordered large sample, which we discuss in Sec. 3.3. Next, we compare the plateau widths of different sample sizes while keeping constant the disorder parameters and depletion length. Figure 6 depicts the sample size dependency of ν = 2 plateau width. It is seen that the larger samples present wider plateaus, if the magnetic field is normalized with the center Fermi energy, E0 F . One can understand this by similar arguments given above, i.e. if the sample is narrow the variation of the confinement potential is stronger, therefore the incompressible strips become narrower, hence, the plateaus. The discrepancy between the experimental results and the screening theory of the IQHE is solved if one considers not only the single impurity potentials but also the overall disorder potential landscape generated by the impurities. In the next part of this section, we investigate the effect of the long range potential fluctuations on the quantized Hall plateaus and find that, when the mobility is reduced the plateaus become wider and stabile, as it is observed in many experiments, (see e.g. Refs. [42, 15, 16]). 3.3. Many many impurities: Potential fluctuations So far we have investigated the effect of single impurity potentials on the overall potential landscape in Sec. 2.2 and on the widths of the plateaus in Sec. 3.1. We have seen that, at high impurity concentration the overall potential fluctuates over a length scale of couple of hundred nanometers, whereas for low NI concentration such length scale can be as large as micrometers. Now we include the effect of this long range potential fluctuations into our screening calculations via modulation potential defined Disorder in screening theory 14 Figure 6. a) The calculated Hall resistances at a large B interval at default temperature, setting 2d = 5 µm, Rg = 20 nm and γI = 0.05, while changing the It is clearly seen that depletion length is much more important depletion length. than the single impurity parameters in determining the plateau widths. (b) The direct comparison of the plateau widths considering different sample sizes. The impurity parameters and depletion lengths are kept constant. Calculations are done at kBT /E0 F = 0.02, whereas the donor density is 4 × 1015 m−2 for all sample sizes. Disorder in screening theory 15 Figure 7. Self-consistently obtained Hall resistances for a modulated system considering a sample of 3 µm. The depletion lengths and other single impurity parameters are kept fixed, whereas the parity of the modulation period is set 5. 2d as Vmod(x) = V0 cos ( 2πxmp ) where, the modulation period mp, is chosen such that the boundary conditions are preserved. At the moment, we consider two modulation periods regardless of the sample width and vary the amplitude of the modulation potential. In the next section, however, we select these two parameters from our estimations obtained in Sec. 2 and Sec. 2.2. Figure 7 depicts the self-consistently calculated Hall resistances, considering different modulation amplitudes V0 for a fixed sample width (2d = 3 µm) and mp = 5. We observe that, the plateaus become wider from the high B field side, when V0 is increased, i.e mobility is reduced. Such a behavior is now consistent with the experimental findings. Since the QHPs occur whenever an incompressible strip is formed (somewhere) in the sample and the modulation forces the 2DES to form an incompressible strip at a higher magnetic field, therefore the plateau is also extended up to higher field compared with the (approximately) non-modulated calculation, V0/E0 F < 0.1. Our investigation of the impurities lead us to conclude that, one has to define mobility at high magnetic fields also taking into account screening effects in general and furthermore also the geometric properties of the sample such as the width and depletion length. As an example if we consider an impurity concentration of ≈ %1 the long range part of the potential fluctuation can be approximated to 900 nm. However, note that the amplitude of this fluctuation varies between %5 − 25 of the Fermi energy, considering different separation thicknesses, therefore the wafer changes from low mobility to intermediate one. Another important parameter is the number of modulations within the system: a sample with an extend of 2 µm and V0/EF = 0.1 Disorder in screening theory 16 mobility low intermediate 1 intermediate 2 high mp (10 µm) mp (2 µm) V0/E0 F 0.5 0.5 0.05 0.05 19-20 9-10 19-20 9-10 5-6 2-3 5-6 2-3 Table 2. A qualitative comparison of the mobility in the presence of magnetic field also taking into account self-consistent screening. Mobility also depends on the size of the sample when screening is also considered. is a high mobility sample with the same mp (only 2 maximum), however, sample with a width of 10 µm is low mobility (10 maximum). In the next section we study the plateau widths of different mobility samples, while keeping constant the extend and the amplitude of long range potential fluctuations (i.e. V0 and mp) and short range impurity parameters ((cid:101)Vimp, NI and Rg) under experimental conditions. 4. Discussion:Comparison with the experiments In this final section, we harvest our findings of the previous sections to make quantitative estimations of the plateau widths, considering narrow gate defined samples. Our aim is to show the qualitative and quantitative differences between "high" and "low" mobility samples, by taking into account properties of the single impurity potentials and the resulting disorder potential. The experimental realizations of these samples are reported in the literature [15, 16]. We estimated in Sec. 2.2 that, the range of the potential fluctuations is (cid:46) 500 nm for low mobility (NI > 3300) and is (cid:38) 1 µm at high mobility. Therefore, the modulation period is chosen such that many oscillations correspond to low mobility, and few oscillations correspond high mobility. As an specific example let us consider a 10 µm sample, for the low mobility we choose mp = 19 − 20 and for the high mobility mp is taken as 9 or 10. The amplitude of the disorder potential is damped to %50 of the Fermi energy when considering the effect of spacer thickness, however, including screening this amplitude is further reduced to few percents. In light of this estimations the low mobility will be presented by a modulation amplitude of V0/E0 F = 0.05. Therefore, we have 4 different combinations of the disorder potential parameters yielding four different mobilities considering two sample widths, as tabulated in table 2. The second important aspect of the disorder is the single impurity parameters, for low mobility set we choose Rg = 20 nm and γI = 0.3, whereas for high mobility Rg = 10 nm and γI = 0.05 is set. Remember that, the range of the single impurity is much less important than γI in determining the plateau width (see sec. 3.1). F = 0.5, whereas high mobility corresponds to V0/E0 Figure 8 summarizes our results considering above discussed mobility regimes for two different sample widths. In Fig. 8a, we show the calculated Hall resistances for a sample of 10 microns with the highest mobility (solid (black) line) and intermediate Disorder in screening theory 17 1 mobility (broken (red) line). The solid line is the highest mobility since the range of the fluctuations are at the order of 1 µm and the amplitude of the modulation potential is five percent of the Fermi energy. The broken line presents the intermediate mobility considering a modulation amplitude of fifty percent. We observe that the lower mobility wafer presents a larger quantized Hall plateau, which is now in complete agreement with the experimental results. Moreover, our calculation scheme is free of localization assumptions in contrast to the known literature and we only considered a very limited level broadening, i.e. γI = 0.05. In fact our results also hold for Dirac-delta Landau levels, however, for the sake of consistency we choose the broadening parameters according to the selected disorder parameters. In Fig. 8c, we show two curves for even lower mobilities, the solid line corresponds to the intermediate 2 case, whereas the broken line is the lowest mobility considered here. The potential fluctuation range (i.e. the modulation period) is chosen to present the low mobility wafer. We again see that for the lowest mobility the quantized Hall plateau is enlarged considerably from both edges of the plateau. These results explicitly show that the quantized Hall plateaus become broader if one strongly modulates the electronic system by long range potential fluctuations, either by changing the range or the amplitude of the modulation. Similar results are also obtained for a relatively narrower sample 2d = 3 µm, Fig. 8b and 8d, however, we see that decreasing the range of the potential fluctuation is more efficient in enlarging the quantized Hall plateaus when compared to the effect of the amplitude of the modulation. The last interesting investigation is on the parity of the modulation period, i.e. whether mp is odd or even. Figure 9 presents the different behavior when considering even (a) or odd (b) periods. Here, all the disorder parameters are kept fixed, other than the parity. We see that for the even parity the plateau is shifted towards the high field edge, both for ν = 2 and 4, whereas for the odd parity the plateau is enlarged from both sides. This tendency is also observed for the larger sample (not shown here). We attribute this behavior again to the formation of the incompressible strips, however, this time only to the one residing at the center of the sample, i.e. the bulk incompressible strip. The picture is as follows: If the maxima of the modulation potential is at the center of the sample, the incompressible strip is formed at a higher magnetic field value, whereas, the edge incompressible strips become narrower at the lower field side. Hence, due to the larger incompressible strip at the bulk of the sample the plateau is shifted to the higher field, in contrast, due to the narrower (compared to the unmodulated system) edge strips the plateau is cut off at higher fields. Since, the edge incompressible strip becomes narrower than the extend of the wave function. For the odd parity, the edge incompressible strips become wider, therefore, the plateau extends to the lower B fields. The enhancement at the high field edge results from the two maximum in the proximity of the center. For a better visualization of the incompressible strip distribution we suggest reader to look at Fig.2 of Ref. [24] and Fig.1 of Ref. [43]. Such a shift of the quantized Hall plateaus is also reported in the literature [42] and is attributed to the asymmetrical density of states due to the acceptors in the system [44]. We claim that, Disorder in screening theory 18 Figure 8. Line plots of the Hall resistance as a function of magnetic field considering two sample widths (2d = 10 µm left panels, 2d = 3 µm right panels) and impurity concentrations (∼ %3 (a) and (b), ∼ %20 (c) and (d)). Here the single impurity parameters are calculated from Eq. 7, otherwise other parameters are the same. the shift due to the modulation parity change observed in our calculations overlap with their findings. Note that in our calculations we only consider symmetric DOS, however, replacing a maxima with a minima at the confinement potential corresponds to the acceptor behavior of the dopants. A systematic experimental investigation is suggested to understand the underlying physical mechanism, where the system is doped with small number of acceptors. 5. Conclusion In this work we tackled with the long standing and widely discussed question of the effect of disorder on the quantized Hall plateaus. The distinguishing aspect of our approach relies on the separate treatment of the long and short range of the disorder potential. We show that assuming Gaussian impurities is not sufficient to describe long range potential fluctuations, however, is adequate to give a prescription in defining the density of states broadening and conductivities. The discrepancy in handling the long range potential fluctuations is cured by the inclusion of a modulation potential to the self-consistent calculations. We estimated the range of these fluctuations from our analytical and numerical calculations considering the effect of dielectric spacer and the screening of the Disorder in screening theory 19 Figure 9. Even-odd parity dependency of the Hall plateaus at high impurity concentration. (a) corresponds to a "acceptor" doped wafer, whereas in (b) the ionized impurities are positively charged. 2DES. It is observed that spacer damps the short range fluctuations effectively, whereas the direct Coulomb interaction is dominant in screening the long range fluctuations. Utilizing the estimations of the range and the amplitude of potential fluctuations, we classified mobility in four groups and calculated the Hall resistances within the screening theory of the quantized Hall effect. We found that the Hall plateaus are wider when decreasing the mobility, not surprisingly. However, the most important point of our theory is that, we do not consider any localization assumptions, still obtain correct behavior of the plateau widths. We show that B = 0 and/or short range impurity defined mobility is not adequate to describe the actual mobility at high magnetic fields, moreover, one has to include geometrical properties of the sample at hand. A natural persecutor theoretical investigation of the present work should deal with the activated behavior of the longitudinal resistance within the screening theory. As it is well known, the properties of the localized states, e.g. the localization length, is usually obtained from the activation experiments [45]. Moreover, spin generalization of the screening theory [46] is necessary to describe and investigate odd integer quantized plateaus also considering level broadening, namely disorder. Disorder in screening theory Acknowledgments 20 One of the authors (A.S.) would like to thank E. Ahlswede, S. C. Lok and J. Weiss for the enlightening discussions on the disorder from "an experimentalist" point of view. The Authors acknowledges, the Feza-Grsey Institute for supporting the III. Nano-electronic symposium, where this work has been conducted partially and would like to acknowledge the Scientific and Technical Research Council of Turkey (TUBITAK) for supporting under grant no 109T083. References [1] B. Kramer, S. Kettemann, and T. Ohtsuki. Localization in the quantum hall regime. Physica E, 20:172, 2003. [2] L. Schweitzer, B. Kramer, and A. MacKinnon. Selective probing of ballistic electron orbits in rectangular antidot lattices. Z. Physik B, 59:379, 1985. [3] W. Cai and C. S. Ting. Screening effect on the landau-level broadening for electrons in gaas-gaalas heterostructures. Phys. Rev. B, 33:3967, 1986. [4] T. Ando and Y. Uemura. J. Phys. Soc. Japan, 36:959, 1974. [5] T. Ando, Y. Matsumoto, and Y. Uemura. J. Phys. Soc. Japan, 39:279, 1975. [6] J. A. Nixon and J. H. Davies. Potential fluctuations in heterostructure devices. Phys. Rev. B, 41:7929 -- 7932, April 1990. [7] M. Stopa and Y. Aoyagi. Effect of donor layer ordering on the formation of single mode quantum wires. Physica B Condensed Matter, 227:61 -- 64, September 1996. [8] M. M. Fogler and B. I. Shklovskii. Resistance of a long wire in the quantum hall regime. Phys. Rev. B., 50:1656, 1994. [9] A. L. Efros. Non-linear screening and the background density of 2deg states in magnetic field. Solid State Commun., 67:1019, 1988. [10] M. Buttiker. Four-terminal phase-coherent conductance. Phys. Rev. Lett., 57:1761, 1986. [11] D. B. Chklovskii, B. I. Shklovskii, and L. I. Glazman. Electrostatics of edge states. Phys. Rev. B, 46:4026, 1992. [12] S. Datta. In Electronic Transport in Mesoscopic Systems, Cambridge, 1995. University press. [13] N. Ruhe, J. I. Springborn, C. Heyn, M. A. Wilde, and D. Grundler. Simultaneous measurement of the de Haas-van Alphen and the Shubnikov-de Haas effect in a two-dimensional electron system. Phys. Rev. B, 74(23):235326 -- +, December 2006. [14] Jiri J Mares, Afif Siddiki, Dobroslav Kindl, Pavel Hubik, and Jozef Kristofik. Electrostatic screening and experimental evidence of a topological phase transition in a bulk quantum hall liquid. New Journal of Physics, 11(8):083028, 2009. [15] J. Horas, A. Siddiki, J. Moser, W. Wegscheider, and S. Ludwig. Investigations on unconventional aspects in the quantum Hall regime of narrow gate defined channels. Physica E Low-Dimensional Systems and Nanostructures, 40:1130 -- 1132, March 2008. [16] A. Siddiki, J. Horas, J. Moser, W. Wegscheider, and S. Ludwig. Interaction mediated asymmetries of the quantized Hall effect. ArXiv e-prints:0905.0204[cond-mat.mes-hall], May 2009. [17] E. H. Hwang and S. Das Sarma. Limit to two-dimensional mobility in modulation-doped GaAs quantum structures: How to achieve a mobility of 100 million. Phys. Rev. B, 77(23):235437 -- +, June 2008. [18] S. J. MacLeod, K. Chan, T. P. Martin, A. R. Hamilton, A. See, A. P. Micolich, M. Aagesen, and P. E. Lindelof. Role of background impurities in the single-particle relaxation lifetime of a two-dimensional electron gas. Phys. Rev. B, 80(3):035310 -- +, July 2009. [19] K. Guven and R. R. Gerhardts. Self-consistent local-equilibrium model for density profile and Disorder in screening theory 21 distribution of dissipative currents in a hall bar under strong magnetic fields. Phys. Rev. B, 67:115327, 2003. [20] A. Siddiki and R. R. Gerhardts. Incompressible strips in dissipative hall bars as origin of quantized hall plateaus. Phys. Rev. B, 70:195335, 2004. [21] A. Siddiki. Self-consistent coulomb picture of an electron-electron bilayer system. Phys. Rev. B, 75:155311, 2007. [22] A. Siddiki and R. R. Gerhardts. The interrelation between incompressible srtips and quantized hall plateaus. Int. J. Mod. Phys. B, 18:3541, 2004. [23] R. R. Gerhartds. The effect of screening on current distribution and conductance quantisation in narrow quantum Hall systems. Phys. Stat. Sol.b, 245:378, January 2008. [24] A. Siddiki and R. R. Gerhardts. Range-dependent disorder effects on the plateau-widths calculated within the screening theory of the iqhe. Int. J. of Mod. Phys. B, 21:1362, 2007. [25] T. Ando, A. B. Fowler, and F. Stern. Electronic properties of two-dimensional systems. Rev. Mod. Phys., 54:437, 1982. [26] T. Champel, S. Florens, and L. Canet. Microscopics of disordered two-dimensional electron gases under high magnetic fields: Equilibrium properties and dissipation in the hydrodynamic regime. Phys. Rev. B, 78(12):125302 -- +, September 2008. [27] T. Kramer. a Heuristic Quantum Theory of the Integer Quantum Hall Effect. International Journal of Modern Physics B, 20:1243 -- 1260, 2006. [28] A. Weichselbaum and S. E. Ulloa. Potential landscapes and induced charges near metallic islands in three dimensions. Phys. Rev E, 68(5):056707 -- +, November 2003. [29] S. Arslan, E. Cicek, D. Eksi, S. Aktas, A. Weichselbaum, and A. Siddiki. Modeling of quantum point contacts in high magnetic fields and with current bias outside the linear response regime. Phys. Rev. B, 78(12):125423 -- +, September 2008. [30] A. Siddiki and R. R. Gerhardts. Thomas-fermi-poisson theory of screening for laterally confined and unconfined two-dimensional electron systems in strong magnetic fields. Phys. Rev. B, 68:125315, 2003. [31] K. Lier and R. R. Gerhardts. Self-consistent calculation of edge channels in laterally confined two-dimensional electron systems. Phys. Rev. B, 50:7757, 1994. [32] J. H. Oh and R. R. Gerhardts. Self-consistent thomas-fermi calculation of potential and current distributions in a two-dimensional hall bar geometry. Phys. Rev. B, 56:13519, 1997. [33] R. R. Gerhardts. Path-integral approach to the two-dimensional magneto-conductivity problem ii application ... Z. Physik B, 21:285, 1975. [34] R. B. Laughlin. Gauge gedankenexperiment iqhe. Phys. Rev. B, 23:5632, 1981. [35] M. Buttiker. IBM J. Res. Dev., 32:317, 1988. [36] B. I. Halperin. Self-consistent local-equilibrium model for density profile and distribution of dissipative currents in a hall bar under strong magnetic fields. Phys. Rev. B, 25:2185, 1982. [37] E. Ahlswede, P. Weitz, J. Weis, K. von Klitzing, and K. Eberl. Hall potential profiles in the quantum hall regime measured by a scanning force microscope. Physica B, 298:562, 2001. [38] S. Ilani, A. Yacoby, D. Mahalu, and H. Shtrikman. Unexpected behavior of the local compressibility near the b=0 metal-insulator transition. Phys. Rev. Lett., 84:3133, 2000. [39] G. A. Steele, R. C. Ashoori, L. N. Pfeiffer, and K. W. West. Imaging Transport Resonances in the Quantum Hall Effect. Physical Review Letters, 95(13):136804 -- +, September 2005. [40] K. Slevin and T. Ohtsuki. Critical exponent for the quantum Hall transition. Phys. Rev. B, 80(4):041304(R), July 2009. [41] S. Sakiroglu, U. Erkarslan, G. Oylumluoglu, A. Siddiki, and I. Sokmen. Microscopic theory of the activated behavior of the quantized Hall effect. ArXiv e-prints:0906.0661[cond-mat.mes-hall], June 2009. [42] R. J. Haug, K. V. Klitzing, and K. Ploog. Analysis of the asymmetry in Shubnikov-de Haas oscillations of two-dimensional systems. Phys. Rev. B, 35:5933 -- 5935, April 1987. [43] A. Siddiki and R. R. Gerhardts. Nonlinear thomas-fermi-poisson theory of screening for a hall bar Disorder in screening theory 22 under strong magnetic fields. In A. R. Long and J. H. Davies, editors, Proc. 15th Intern. Conf. on High Magnetic Fields in Semicond. Phys., Bristol, 2002. Institute of Physics Publishing. [44] R. J. Haug, R. R. Gerhardts, K. V. Klitzing, and K. Ploog. Effect of repulsive and attractive scattering centers on the magnetotransport properties of a two-dimensional electron gas. Physical Review Letters, 59:1349 -- 1352, September 1987. [45] J. Matthews and M. E. Cage. Temperature Dependence of the Hall and Longitudinal Resistances in a Quantum Hall Resistance Standard. J. Res. Natl. Inst. Stand. Technol., 110:497, October 2005. [46] A. Siddiki. The spin-split incompressible edge states within empirical Hartree approximation at intermediately large Hall samples. Physica E Low-Dimensional Systems and Nanostructures, 40:1124 -- 1126, March 2008. 

1403.6483

2

1403

Dissipation enhanced vibrational sensing in an olfactory molecular switch

Motivated by a proposed olfactory mechanism based on a vibrationally-activated molecular switch, we study electron transport within a donor-acceptor pair that is coupled to a vibrational mode and embedded in a surrounding environment. We derive a polaron master equation with which we study the dynamics of both the electronic and vibrational degrees of freedom beyond previously employed semiclassical (Marcus-Jortner) rate analyses. We show: (i) that in the absence of explicit dissipation of the vibrational mode, the semiclassical approach is generally unable to capture the dynamics predicted by our master equation due to both its assumption of one-way (exponential) electron transfer from donor to acceptor and its neglect of the spectral details of the environment; (ii) that by additionally allowing strong dissipation to act on the odorant vibrational mode we can recover exponential electron transfer, though typically at a rate that differs from that given by the Marcus-Jortner expression; (iii) that the ability of the molecular switch to discriminate between the presence and absence of the odorant, and its sensitivity to the odorant vibrational frequency, are enhanced significantly in this strong dissipation regime, when compared to the case without mode dissipation; and (iv) that details of the environment absent from previous Marcus-Jortner analyses can also dramatically alter the sensitivity of the molecular switch, in particular allowing its frequency resolution to be improved. Our results thus demonstrate the constructive role dissipation can play in facilitating sensitive and selective operation in molecular switch devices, as well as the inadequacy of semiclassical rate equations in analysing such behaviour over a wide range of parameters.

cond-mat.mes-hall

cond-mat

Dissipation enhanced vibrational sensing in an olfactory molecular switch Agata Chęcińska,1, a) Felix A. Pollock,2, a) Libby Heaney,1 and Ahsan Nazir3, 4 1)Centre for Quantum Technologies, National University of Singapore, Singapore 117543 2)Atomic & Laser Physics, Clarendon Laboratory, University of Oxford, Parks Road, Oxford OX1 3PU, United Kingdom 3)Photon Science Institute and School of Physics & Astronomy, University of Manchester, Oxford Road, Manchester M13 9PL, United Kingdom 4)Centre for Quantum Dynamics, Imperial College London, London SW7 2AZ, United Kingdom (Dated: September 4, 2018) Motivated by a proposed olfactory mechanism based on a vibrationally-activated molecular switch, we study electron transport within a donor-acceptor pair that is coupled to a vibrational mode and embedded in a surrounding environment. We derive a polaron master equation with which we study the dynamics of both the electronic and vibrational degrees of freedom beyond previously employed semiclassical (Marcus- Jortner) rate analyses. We show: (i) that in the absence of explicit dissipation of the vibrational mode, the semiclassical approach is generally unable to capture the dynamics predicted by our master equation due to both its assumption of one-way (exponential) electron transfer from donor to acceptor and its neglect of the spectral details of the environment; (ii) that by additionally allowing strong dissipation to act on the odorant vibrational mode we can recover exponential electron transfer, though typically at a rate that differs from that given by the Marcus-Jortner expression; (iii) that the ability of the molecular switch to discriminate between the presence and absence of the odorant, and its sensitivity to the odorant vibrational frequency, are enhanced significantly in this strong dissipation regime, when compared to the case without mode dissipation; and (iv) that details of the environment absent from previous Marcus-Jortner analyses can also dramatically alter the sensitivity of the molecular switch, in particular allowing its frequency resolution to be improved. Our results thus demonstrate the constructive role dissipation can play in facilitating sensitive and selective operation in molecular switch devices, as well as the inadequacy of semiclassical rate equations in analysing such behaviour over a wide range of parameters. I. INTRODUCTION Characterising the influence of the environment on the transfer of charge and energy in an open quantum sys- tem is a problem of significant current interest.1 -- 8 In particular, resonant (or near resonant) interactions be- tween environmental degrees of freedom and those inher- ent to the system are thought to play an important role in numerous physical processes.9 -- 20 However, a compre- hensive picture of such dynamics is only beginning to emerge due to the complexity of the systems in ques- tion. Here, by focusing on a proposed model for olfaction as a vibrationally-activated molecular switch, we explore the detailed effects of the environment on the dynam- ics of electron transfer (ET) in an open quantum system, aiming to gain physical insight into vibrationally-assisted transport processes more generally. In fact, obtaining a deeper understanding of olfac- tion, the mechanism for which is still being actively debated,21 -- 30 is an important problem in its own right for both fundamental science and industry.31 -- 35 The pre- vailing theory, known as the lock-and-key model, ex- plains how the odorant size and shape can provide discrimination in the receptor.36 However, this theory does not give a straightforward explanation of why it a)These authors contributed equally to the work. may be possible to distinguish the scents of some very similar odorants, for example between those that are deuterated and non-deuterated.26,30,37 It was suggested as early as 1938 that the sensing of vibrational spectra of molecules38 -- 40 -- later proposed to occur via electron transfer21,22 -- could play an important role in olfaction, supplementing (rather than replacing) the existing lock- and-key model. Recent work, focussing on constructing and exploring model systems that capture the impor- tant physical processes,24,27 has shown that this is indeed a viable proposition. The suggested mechanism, which harnesses vibrationally-assisted ET in a similar manner to inelastic electron tunneling spectroscopy,41,42 can be viewed as an example of a molecular switch, wherein specific vibrations of an external molecule actuate the receptor and lead to a pronounced electron flow. Detec- tion via the frequency of vibrations could help to discern two molecules that are otherwise very similar. However, the need for clear discrimination of the difference in ET dynamics in the presence and absence of the odorant im- poses certain design principles on the molecular switch, which may or may not lead to the employment of quan- tum phenomena to optimise performance. Previously, a receptor-odorant spin-boson model was introduced to describe the vibrationally-assisted ET pro- cess at the heart of the proposed mechanism.24,27 Us- ing an analysis based on the semiclassical Marcus-Jortner (MJ) formula for the ET rate,43 -- 47 it was shown that for certain parameter values the rates for ET in the presence 5 1 0 2 b e F 1 1 ] l l a h - s e m . t a m - d n o c [ 2 v 3 8 4 6 . 3 0 4 1 : v i X r a and absence of the odorant can differ drastically. This indicates how the ET process could help in molecular recognition via sensing of vibrational spectra; sensitivity of the switch to the presence or absence of the odorant can be understood as a significant difference in typical timescales, or more generally in the population dynam- ics, for processes with or without the odorant coupled to the receptor. Additionally, in Refs. 27 and 29 further evidence supporting a vibrationally-assisted mechanism was obtained from sophisticated quantum chemistry cal- culations. Inspired by the possibility of vibrational sensing in ol- faction, we focus here on the physical model of a molecu- lar switch tuned for frequency detection. We go beyond the MJ approach to look at both frequency selectivity and detailed dynamics in a variety of regimes in which a semiclassical analysis breaks down. Starting from a microscopic Hamiltonian describing the odorant (as an oscillator), receptor (as a donor-acceptor two-level sys- tem), and environment (as a collection of independent oscillators), we derive a polaron-representation master equation for the ET process.48 -- 58 From this we may ex- tract the relevant ET rates, extending previous analyses to a broader set of parameters and looking in more detail at the assumptions required for the MJ rates to be valid; in fact, these rates arise naturally from our master equa- tion in the semiclassical limit where the temperature is high compared to the energy scales of the environment. In general, we find that the dynamics of donor-acceptor (DA) populations predicted by our master equation can differ considerably from that given by the simpler MJ rates. The main reason for this discrepancy is the as- sumption inherent to the MJ analysis of exponential ET from donor to acceptor, which we find to be invalid for a wide set of parameter values. Our approach also has the crucial advantage of allow- ing us to additionally incorporate the effects of dissipa- tion on the odorant mode, since it is treated explicitly in our formalism. We shall show that by introducing suf- ficiently strong mode dissipation, it is possible to bring the DA dynamics derived from our master equation into a simpler exponential form. However, even within this strongly dissipative regime, our master equation predic- tions can still differ markedly from the semiclassical MJ theory, depending upon the specific nature of the envi- ronment experienced by the DA pair. In particular, we show that while in the MJ case the receptor is very sensi- tive to the presence or absence of the odorant, it is far less so to the specific odorant vibrational frequency. We find, however, that the frequency resolution can be enhanced by considering environments which contain components of similar or larger energy to that set by the ambient tem- perature, i.e. by working outside the semiclassical limit. More generally, we show that when considering odorant mode dissipation, both the sensitivity of the switch to the presence or absence of the odorant, and to the resonance conditions between the odorant and the DA pair, can be significantly amplified in comparison to the dissipation- 2 Figure 1. Relevant DA-odorant energy levels. The three most relevant single-electron joint DA-odorant states and the principal transition rates between them. The rates ΓDnAm and ΓAnDm are derived in Sec. IV, whilst odorant dissipation with rate γ0 is discussed in Sec. V. The notation X, n(cid:105) refers to the joint state of the DA pair, X ∈ {D, A}, and the odorant in Fock state n(cid:105). less case. We thus find that odorant dissipation plays a constructive role in enhancing the vibrational sensing capabilities of our molecular switch. This paper is structured as follows. We begin in Sec- tion II by describing the molecular switch model and giv- ing an estimate of the parameters relevant to our analy- sis. In Section III we briefly outline the derivation of our polaron-representation master equation, with further de- tails given in the Appendix. Sections IV A and B are devoted to showing how the MJ rates naturally emerge in the semiclassical limit of our master equation, in the absence and presence of the odorant, respectively. Sec- tion IV C contrasts the full master equation dynamics with a simpler rate analysis in the illustrative case of a dissipationless odorant mode, while Section V considers the impact of odorant dissipation upon the ET dynamics and the resulting switch frequency resolution. Finally, we summarise our findings in Section VI. II. MODEL OF OLFACTION Few details are presently known about olfactory recep- tors and their properties.32,33,59 Here, following earlier work,24,27,29 we study a simplified model which captures the essential physics of the electron transfer process. We assume that there exist specific electronic states of the re- ceptor that can be identified as a DA pair, with other lev- els well separated in energy. We model the vibrationally- assisted ET process using a spin-boson Hamiltonian. The DA pair is coupled to an environment represented by a bath of harmonic oscillators (which includes the vibra- tional degrees of freedom of the receptor) and to the odorant (when present in the receptor). The odorant is modelled as a single mode harmonic oscillator, though this could also be extended to a set of modes. Our Hamil- tonian thus takes the form H =
DD(cid:105)(cid:104)D +
AA(cid:105)(cid:104)A + V(cid:0)A(cid:105)(cid:104)D + D(cid:105)(cid:104)A(cid:1) (cid:88) (cid:2)ωkb + ω0a†a + (γDD(cid:105)(cid:104)D + γAA(cid:105)(cid:104)A)(a† + a) † kbk + (γkDD(cid:105)(cid:104)D + γkAA(cid:105)(cid:104)A)(b + k + bk)(cid:3). † k (1) Here, X(cid:105), with X = D, A, represents the donor (D) and acceptor (A) electronic state with on-site energy
X, and V refers to the tunnel coupling. The odorant molecu- lar mode, with creation (annihilation) operator a† (a), has frequency ω0 and is coupled to the DA pair via γX. The environmental oscillators, with creation (annihila- † tion) operators b k (bk) for modes of frequency ωk, couple to the receptor electronic sites via γkX. We shall con- sider odorant dissipation in Section V below. Transfer of an electron gives rise to changes in the local electric field and justifies the present form of interaction between the oscillators and the DA pair.1,24,27,29 Fig. 1 shows the most relevant energy levels of the joint system along with transition rates between them (to be derived in later sec- tions). Specific parameter values for the model will be discussed below. In Refs. 24 and 27, MJ formulas were employed to de- fine two types of ET rate: elastic, in the absence of the odorant, and inelastic, in which the odorant is present. These rates were calculated using Fermi's golden rule, wherein the tunnelling term V is treated as a perturba- tion. This naturally gives rise to a representation of the receptor-odorant-environment basis in terms of displaced oscillator states -- as can be seen from Eq. (1) by setting V = 0 -- between which it is assumed that incoherent ET occurs. This is, in fact, a reflection of the polaron picture that we shall discuss in the next section. In Ref. 24, typi- cal ET timescales were estimated to be ∼ 100 ns (elastic) versus ∼ 1 ns (inelastic), predicting the inelastic process to be much faster than the elastic one, as required for discriminating between the presence and the absence of the odorant. Ref. 27 also discusses a wider range of pa- rameters in which a separation of timescales between the elastic and inelastic processes can be obtained. Here, rather than following a Fermi golden rule calcula- tion, we shall look instead at receptor ET dynamics and issues such as odorant vibrational frequency resolution from the perspective of the master equation formalism. This includes the MJ rate analysis as a limiting case, but can also go well beyond the restricted regime of validity of such an approach. A. Estimated parameters 3 Parameter
D −
A ω0 (γD − γA)2/ω2 0 T Value Parameter 200 meV 100 − 300 meV 0.01 300 K V λ ωc kBT Value 1 meV 10 − 60 meV 10 , kBT , 2kBT kB T 25.85 meV Table I. Table of parameter values used in this article. shall, however, explore an effective window of frequen- cies away from resonance to study the issue of molecu- lar frequency recognition, i.e. we would like to explore whether the switch is sensitive merely to the presence or absence of the odorant, or to its particular vibrational frequency as well. Typical values for ω0 are in the range of 70 − 400 meV, and we choose
D −
A = 200 meV for the DA energy gap. The DA tunnel coupling is estimated to be on the order of V ∼ 1 meV.24 Increasing V acts to enhance the ET rate in the absence of the odorant in the receptor, which is disadvantageous as far as the switching mechanism is concerned. Keeping V small in comparison to other system parameters is therefore well motivated physically. Estimates of the coupling between the DA pair and the odorant mode, and of the reorganization energy of the multi-mode environment, have been given as (γD − γA)2/ω2 0 ∼ 0.01 and λ ∼ 30 meV, respectively. Here, we shall explore variations in the DA-environment coupling strength through the reorganisation energy, fo- cussing on the range λ ∼ 10 − 60 meV. Previously, the detailed properties of the environment, such as the spec- tral density, (cid:88) J(ω) = (γkD − γkA)2δ(ω − ωk), (2) k were not discussed, as they do not enter the semiclassi- cal MJ rates. They are, however, important for our more general analysis. We choose to work with an Ohmic envi- ronment, J(ω) ∝ ω as ω → 0, with a characteristic high frequency cut-off ωc, since in the absence of more detailed information, it straightforwardly reproduces the results of Refs. 24 and 27 in the semiclassical limit (βωc (cid:28) 1, where β = 1/kBT is the inverse temperature). We take the cut-off to be exponential, leading to the following spectral density defined also in terms of the reorganisa- tion energy: J(ω) = λ e−ω/ωc. ω ωc (3) For consistency with the published literature, we fol- low the estimation of parameters presented in Refs. 24 and 27. In common with previous studies, we assume that the energy gap
D −
A is relatively close to res- onance with the odorant vibrational frequency ω0. We We shall vary the ratio ωc/kBT from low to high in order to explore the effect of widening the range of frequencies within the environment that can interact with the recep- tor DA pair. We assume that T = 300 K throughout, and summarise the model parameters in Table I. III. POLARON MASTER EQUATION As noted, for our system to act as an effective molecu- lar switch, the Hamiltonian parameters should be such that unwanted transitions from donor to acceptor are avoided when the odorant is absent. This requires the tunnel coupling V to be small compared to other energy scales in the problem. In this regime, it is convenient to move into a polaron transformed reference frame to remove the linear coupling terms in Eq. (1). Provided V is indeed small, then perturbative expansions in the transformed basis are valid over a much wider range of system-environment coupling strengths than those in the untransformed case.48 -- 58 Let us consider the (unitary) polaron transformation U = e−(SA+SB) acting on our Hamiltonian, where SA =(cid:0)uDD(cid:105)(cid:104)D + uAA(cid:105)(cid:104)A(cid:1)(a† − a), (cid:88) (cid:0)αkDD(cid:105)(cid:104)D + αkAA(cid:105)(cid:104)A(cid:1)(b † k − bk), (4) (5) SB = and k with uX = γX /ω0 and αkX = γkX /ωk. The polaron transformed Hamiltonian takes the form HP = U†HU = HSP + HB + HIP, where HSP =
(cid:48) AA(cid:105)(cid:104)A + ω0a†a, DD(cid:105)(cid:104)D +
(cid:48) (cid:88) HB = k † ωkb kbk, and HIP = V(cid:0)D(cid:105)(cid:104)AA+B+ + A(cid:105)(cid:104)DA−B−(cid:1). (6) (7) (8) k ωkα2 X =
X − ω0u2 Here,
(cid:48) kX is the polaron- shifted on-site energy, and A± = D(±(uD − uA)) and B± = ΠkDk(±(αkD − αkA)) are the new oscillator in- teraction operators, written in terms of the displacement operators D(u) = eua†−u∗a and Dk(αk) = eαkb kbk, re- spectively. Note that the polaron transformation leaves the operators D(cid:105)(cid:104)D and A(cid:105)(cid:104)A, required for calculating DA populations, unchanged. † k−α∗ (cid:90) ∞ Tracing out the environment and treating the pertur- bative term HIP up to second order in the standard Born- Markov approximations,7 we obtain a master equation in the polaron frame interaction picture (with respect to HSP + HB), given by ρSP(t) = −V 2 (cid:8)(cid:0)[A(cid:105)(cid:104)DA−(t),D(cid:105)(cid:104)AA+(t − τ )ρSP(t)]e−i
τ + [D(cid:105)(cid:104)AA+(t),A(cid:105)(cid:104)DA−(t − τ )ρSP(t)]ei
τ(cid:1)C(τ ) + H.c.(cid:9), (9) dτ 0 see the Appendix for further details. Here, ρSP(t) = trB{ χP(t)} is the reduced density operator describing the X −(cid:80) 4 DA pair and odorant mode on tracing out the receptor environment, with χP(t) the total density operator in the polaron frame interaction picture. The bath correlation function, defined as C(τ ) = trB (B±(τ )B∓(0)ρB) with B±(t) = eiHB tBe−iHB t, takes the form of an exponential of the lineshape function: −(cid:82) ∞ D −
(cid:48) C(τ ) = e with
=
(cid:48) 0 dω J(ω) ω2 [(1−cos ωτ ) coth(βω/2)+i sin ωτ ], (10) A now the DA energy gap. IV. ELECTRON TRANSFER DYNAMICS Having derived our polaron master equation, we shall use it below, in Section C, to investigate the receptor ET dynamics beyond the MJ rate analysis employed in earlier work. First, however, we show how the MJ rates arise from our master equation in the semiclassical limit, for both elastic and inelastic processes, and thus how our theory is consistent with previous treatments in this regime. In addition, we discuss other important tran- sition rates which we shall subsequently show are sup- pressed only with the introduction of strong dissipation acting on the odorant mode. A. Odorant absent Let us consider the population transfer from donor to acceptor for the case in which the odorant is absent, meaning that A±(t) = 11 in Eq. (9). We can then eas- ily derive rate equations governing the donor (pD(t)) and acceptor (pA(t)) population dynamics. Our interest lies in the rates that appear in these equations Γ(±
) = dτ e±i
τ C(τ ), (11) (cid:90) ∞ −∞ where, by defining ΓDA = V 2Γ(
) and ΓAD = V 2Γ(−
), we obtain pD(t) = −ΓDApD(t) + ΓADpA(t), pA(t) = −ΓADpA(t) + ΓDApD(t). In the limit that ΓDA (cid:29) ΓAD, these equations define exponential transfer of population from donor to acceptor at (approximately) the rate ΓDA, which we write as (13) For a low-frequency environment in which βωc (cid:28) 1, we may derive a simple form for ΓDA which turns out to be dτ ei
τ e−ϕ(τ ), (cid:90) ∞ (cid:2)(1 − cos ωτ ) coth(βω/2) + i sin ωτ(cid:3). (12) −∞ ΓDA = V 2 (cid:90) ∞ where ϕ(τ ) = dω 0 J(ω) ω2 5 the same as the MJ rate. Taking the spectral density defined in Eq. (3) and expanding coth(βω/2) ≈ 2/βω in Eq. (13), we find (cid:2)ωc(iβ + 2τ ) tan−1(ωcτ ) − ln(1 + ω2 c τ 2)(cid:3) . ϕ(τ ) ≈ λ βω2 c (14) In the regime in which we are presently interested, e−ϕ(τ ) is strongly peaked around τ = 0, such that we may ex- pand ϕ(τ ) to second order in τ to give ϕ(τ ) ≈ iλτ + λτ 2 β . (15) With these assumptions we can write (cid:90) ∞ may or may not act to enhance the rate associated with the process. We return to the master equation [Eq. (9)] to de- rive an expression for the dynamics of the population of the acceptor and a given Fock state n(cid:105) of the odorant, ρAnAn = trS+O(A, n(cid:105)(cid:104)n, AρSP(t)) = (cid:104)n, AρSP(t)A, n(cid:105), where the trace is taken over both the odorant (O) (cid:80) and the DA pair (S) degrees of freedom. Decom- posing the DA-odorant density operator as ρSP(t) = X,X(cid:48),l,m ρXlX(cid:48)m(t)X, l(cid:105)(cid:104)X(cid:48), m, where X, X(cid:48) ∈ {D, A} and l, m are odorant Fock states, allows us to identify the different contributions to the change in population of the state A, n(cid:105). In particular, if we assume that the only donor-odorant state of interest is one with no odorant ex- citations, D, 0(cid:105) -- valid when we initialise the system in this state and energy splittings are large enough to sup- press transitions to any other donor-odorant states -- we may then define a rate of transfer D, 0(cid:105) → A, n(cid:105) as ΓDA ≈ V 2 dτ ei(
−λ)τ e−λτ 2/β −∞ 1√ 4πkBT λ exp − (
− λ)2 4kBT λ = 2πV 2 (cid:20) (cid:21) , which agrees with the elastic rate (odorant absent) pre- sented in Ref. 24 when αkD = −αkA is assumed, and as a consequence
→
D −
A. We apply this constraint on αkX throughout the paper, although little modification would be required to account for the more general case. Choosing λ = 30 meV, and other parameters as out- lined in Table I, we obtain ΓDA = 5.67× 10−6 meV from Eq. (16), which corresponds to τDA = 1/ΓDA = 116 ns as the DA transfer time in the absence of the odorant. Of course, the limit βωc (cid:28) 1 used to derive Eq. (16) may not always be met, in which case we can use the more general form of Eq. (12) to define the DA elastic transfer rate. Importantly, this allows us to discuss lower tem- peratures and larger environmental cut-offs than those to which the MJ rates apply. B. Odorant present A similar rate can also be derived when the odorant is present in the receptor. In this situation, we must deal with a more complex system, as the reduced density operator (after tracing out the environmental degrees of freedom) now encompasses both the two-level DA pair and the odorant harmonic oscillator. Obtaining general equations of motion for the DA populations becomes sig- nificantly more involved. However, this is unnecessary if we instead look at ET rates between specific states of the combined DA-odorant system. Let us assume that we initialize the system in the state D, 0(cid:105) (electron on the donor, odorant in its ground-state 0(cid:105)) and we are inter- ested in the rate of ET to a state of the form A, n(cid:105), where n(cid:105) is an arbitrary Fock (number) state of the odorant. We are thus considering situations in which the receptor population transfer D(cid:105) → A(cid:105) is accompanied by exci- tation of the odorant vibrational mode 0(cid:105) → n(cid:105), which (cid:18) (cid:90) ∞ 0 (16) ΓD0An = ei
τ C(τ )(cid:104)nA−(t − τ )0(cid:105)(cid:104)0A+(t)n(cid:105) dτ (cid:19) + e−i
τ C∗(τ )(cid:104)nA−(t)0(cid:105)(cid:104)0A+(t − τ )n(cid:105) V 2. (17) √ Using (cid:104)nD(α)0(cid:105) = (αn/ n!)e−α2/2, it is straightfor- ward to calculate the expectation values of A±(t). We then arrive at V 2(uD − uA)2ne−uD−uA2 ΓD0An = n! dτ ei(
−nω0)τ e−ϕ(τ ), (18) (cid:90) ∞ −∞ × which generalises Eq. (12) in the presence of the odorant (ΓDA = ΓD0A0). At this point we can again take the limit βωc (cid:28) 1, and follow the same steps that led us from Eq. (12) to Eq. (16) to find ΓD0An ≈ 2πV 2 (uD − uA)2ne−uD−uA2 (cid:21) 4πkBT λ (cid:20) √ − (
− nω0 − λ)2 × exp n! . 4kBT λ (19) Once again, this result for the inelastic ET rates (n > 0) agrees with the MJ form found in Ref. 24 when we assume uD = −uA and αkD = −αkA, such that
→
D−
A. Us- ing the values in Table I, choosing λ = 30 meV and ω0 =
D −
A, we obtain ΓD0A1 = 4.71× 10−4 meV, which cor- responds to a transfer time of τD0A1 = 1/ΓD0A1 = 1.4 ns. This is far shorter than the 116 ns transfer time found above in the absence of the odorant, as required for a vi- able molecular switch. Additionally, if we take the same parameter values and look at the rate for the two-phonon transition we then find ΓD0A2 = 1.24 × 10−13 meV, and a corresponding time τD0A2 = 5.3 s, confirming that the single-phonon process is dominant within this treatment. As before, in situations in which the limit βωc (cid:28) 1 does not apply, we may instead use Eq. (18) to define the in- elastic rates, again generalising the MJ rates to a wider range of parameters. Of course, our master equation also allows us to calcu- late the reverse rates ΓAnD0 (as well as other rates such as ΓDnDm and ΓAnAm, which do not play a major role). Following a derivation similar to that leading to Eq. (18), we find V 2(uD − uA)2ne−uD−uA2 ΓAnD0 = (cid:90) ∞ × n! dτ ei(
−nω0)τ e−ϕ(−τ ), (20) −∞ and, in the limit βωc (cid:28) 1, ΓAnD0 ≈ 2πV 2 (uD − uA)2ne−uD−uA2 (cid:21) √ n! (cid:20) 4πkBT λ − (
− nω0 + λ)2 4kBT λ × exp . (21) Notably, the single-phonon reverse transfer rate, ΓA1D0, is equal to the donor-to-acceptor rate, ΓD0A1, when the odorant is resonant with the receptor (ω0 =
D −
A). We shall see below that this has important implications for the dynamics of the DA pair over a wide range of parameters, often invalidating the treatment of inelastic ET as being a one-way process. C. Master equation dynamics versus ET rates We are now in a position to compare our full polaron master equation [Eq. (9)] with a model in which the donor population simply decays with the ET rates calculated in the previous section. Specifically, we would like to know how reliably we can apply a rate analysis, such as the MJ evaluation employed in previous studies,24,27 to parameter regimes estimated to be relevant to the olfac- tory process. We have already seen that with the odor- ant present, the possibility of transfer from acceptor to donor cannot be neglected when close to resonance. This is especially true when the inelastic process dominates, as the reverse rate in the elastic case is suppressed by a factor ∼ exp [−(
D −
A)λ/kBT ]. Without further as- sumptions, we thus expect exponential decay from donor to acceptor to hold only when the ET is mediated pri- marily by the environment (n = 0), rather than by the odorant itself. This intuition is borne out in Fig. 2, which shows a comparison between the DA population dynamics (in the presence of the odorant) calculated from our master equation (9), and predicted by the ET rates of Eq. (18). The receptor is taken to be initialised in the donor state, with the odorant mode and the environment in thermal states, of HO = ω0a†a and of HB, respectively.60 As can clearly be seen, the ET dynamics can differ consid- erably between the two methods. As expected, the best 6 agreement is found for large reorganisation energies (right column), where elastic transfer due to the environment dominates over inelastic transfer via the odorant. On the other hand, when the odorant does play a significant role in mediating the ET, then the agreement is generally poor, with our master equation predicting DA dynamics that cannot be fitted by a single exponential form.61 This is particular true in the low-frequency environment limit (ωc (cid:28) kBT ) shown in panels (i) and (ii), to which the MJ rates would usually be assumed to apply.62 These discrepancies suggest, in fact, that it is the combination of population accumulation within the odorant and the inherent competition between forward and backward pro- cesses that limits the timescale for complete transfer from donor to acceptor in the master equation. We can thus gain further insight by looking at the pop- ulation dynamics of the odorant mode. In Fig. 3 we plot the evolution of the lowest two odorant states as predicted by our master equation, for the same param- eter sets as in Fig. 2. Note that other levels are never significantly populated, since the total initial energy is insufficient to further excite the mode. In the cases in which the odorant-assisted (inelastic) rate dominates, ex- citation is quickly transferred from the DA pair to the odorant. The electron is then equally likely to be found on the acceptor (having deposited energy to the odor- ant mode) as it is on the donor. Due to the equality of the forward and backward rates on resonance, the only way energy can now exit the system, to allow full ET to the acceptor, is into the environment, the timescale for which is dependent upon the reorganisation energy and cut-off frequency. Since the rate analysis that leads to the dashed curves in Fig. 2 ignores the possibility of reverse acceptor-to-donor transfer, it consequently over- estimates the inelastic transfer rate. It is worth noting that even for the cases in which the master equation dy- namics shown in Fig. 2 appears to reach a (quasi) steady state with the donor and acceptor similarly populated, in fact full ET from donor to acceptor does still eventually occur, albeit on a very long timescale. Of course, in any realistic physical setting, the odor- ant mode would not be isolated from the environment. In MJ theory, for the semiclassical limit to be valid in the presence of the odorant an implicit assumption is made that the odorant vibrational mode dissipates its energy to the environment on a timescale much shorter than that of the ET process. However, within our master equation approach we can explicitly include mode dissipation, and indeed explore the effects of varying the associated rate. Thus, by contrasting the dynamics illustrated here in the absence of odorant dissipation with that in its presence in the next section, we can investigate the detailed role such dissipation plays in vibrationally-assisted ET. We shall show that for strong dissipation we can recover ex- ponential inelastic ET, though generally at a rate that differs from the semiclassical MJ formula. Furthermore, we shall demonstrate that reducing the level of odorant dissipation can in fact act to limit the frequency detection 7 odorant states to ensure numerical convergence) and by the total donor-to-acceptor ET rate(cid:80)3 Figure 2. Donor and acceptor population dynamics. Comparison of the donor, pD(t) (black, pD(0) = 1), and acceptor, pA(t) (red, pA(0) = 0), population dynamics as predicted by the polaron master equation (9) (solid lines, calculated using 3 n=0 ΓD0An from Eq. (18) (dashed lines). The reorganisation energy λ increases from left to right: λ = 15 meV in the left column, λ = 30 meV in the middle column and λ = 60 meV in the right column. The bath cut-off frequency ωc increases from top to bottom, with ωc = kBT /10 for panels (i - iii), ωc = kBT for panels (iv - vi) and ωc = 2kBT for panels (vii - ix). The odorant frequency is chosen to be resonant with the DA pair, i.e. ω0 =
D −
A = 200 meV, and other parameters are as listed in Table I. Note that for each set AA(cid:105)(cid:104)A in the of parameters, the populations eventually tend to a thermal distribution with respect to HDA =
(cid:48) steady state; this is true even for panels (i) and (ii), which equilibrate on a very long time scale. DD(cid:105)(cid:104)D +
(cid:48) properties of the receptor, suggesting that strong dissi- pation may actually be beneficial for obtaining frequency selective switching processes. V. DISSIPATION ASSISTED ELECTRON TRANSFER Given that, when present, the odorant is considered to be in the vicinity of the DA pair, it is reasonable to as- sume that it too interacts with the surrounding environ- ment. Taking a linear coupling form we can model the effects of the resulting mode dissipation in a straight- forward manner, once Born-Markov and rotating wave approximations are made, by introducing an additional Lindblad term to the right-hand-side of our master equa- tion (9):7 Ldiss[ρSP] = − γ0 2 − γ0 2 (cid:0)aa† ρSP − 2a† ρSPa + ρSPaa†(cid:1) (N0 + 1)(cid:0)a†aρSP − 2aρSPa† + ρSPa†a(cid:1). N0 (22) dotted curves), for moderate dissipation (γ0 ∼ ΓD0A1, dashed curves) and for strong dissipation (γ0 (cid:29) ΓD0A1, solid curves). In the limit of large dissipation, we see that the behaviour of the DA populations is consistent with that given by the transfer rates of Eq. (18), and the de- scription of the dynamics as an exponential ET process from donor to acceptor once again becomes valid (note the agreement between the points and solid curves). Fur- thermore, the dynamics only agrees with that predicted by MJ theory (blue squares) in the limit of low cut-off frequency in the bath (top panel). Importantly, we can also see that adding mode dissipation actually assists the transfer of population from donor to acceptor. In fact, the timescale for complete ET is substantially reduced as the dissipation rate is increased, thus ensuring a greater variance between the transfer time in the presence and absence of the odorant. In this manner, mode dissipa- tion is seen to be beneficial for the molecular switching process, allowing easier discrimination in the receptor be- tween the cases with and without the odorant present. Here, γ0 is the odorant dissipation rate and N0 = (eω0/kB T − 1)−1 is the phonon occupation number. In Fig. 4 we investigate the effect of adding dissipa- tion to the odorant. Specifically, we show a comparison of the dynamics in the absence of dissipation (γ0 = 0, A. Frequency resolution at strong dissipation We can think of the olfactory model outlined herein in terms of a vibrational spectrometer, and one of the principle figures of merit for any spectroscopic device is Donor
Acceptorpopulation02460.0.20.40.60.81.
i
02460.0.20.40.60.81.
ii
02460.0.20.40.60.81.
iii
02460.0.20.40.60.81.
iv
02460.0.20.40.60.81.
v
02460.0.20.40.60.81.
vi
02460.0.20.40.60.81.
vii
02460.0.20.40.60.81.
viii
02460.0.20.40.60.81.
ix
02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.02460.0.20.40.60.81.Time
ns
8 Figure 3. Odorant dynamics. Populations of the odorant ground state, p0(cid:105)(cid:104)0(t) (black, p0(cid:105)(cid:104)0(0) = 1), and first excited state, p1(cid:105)(cid:104)1(t) (red, p1(cid:105)(cid:104)1(0) = 0), calculated from Eq. (9). The reorganisation energy λ increases from left to right: λ = 15 meV in the left column, λ = 30meV in the middle column and λ = 60 meV in the right column. The bath cut-off frequency ωc increases from top to bottom, with ωc = kBT /10 for panels (i - iii), ωc = kBT for panels (iv - vi) and ωc = 2kBT for panels (vii - ix). The odorant frequency is ω0 =
D −
A = 200 meV. Populations of higher Fock states are negligible at all times. its frequency resolution. The more sensitive the individ- ual receptors are to resonance conditions between the DA pair and odorant, the better they will be able to distin- guish the vibrational spectra of different molecules. To explore this aspect in our model, in Fig. 5 we map out the ratio r = (Γtot − Γ0)/(Γtot + Γ0) of tunneling rates as a function of odorant frequency ω0 and bath reorganisa- tion energy λ. Here, the definition of Γtot as the tunneling rate in the presence of the odorant and Γ0 as the rate in its absence is made possible by the strong mode dissipa- tion, γ0 = 1000 ns−1, which ensures exponential ET from donor to acceptor as just demonstrated. The top panel corresponds to the regime in which ωc (cid:28) kBT , therefore representing the behaviour predicted by the MJ rates of Eqs. (16) and (19), while the other panels encompass pa- rameter ranges outside the MJ limit. We see that in each plot there clearly exists a frequency window for which the tunneling rate is enhanced in the presence of the odorant above the background. Further- more, this window becomes narrower as the environmen- tal cut-off frequency is increased (middle and bottom panels), suggesting that the structure of the environment can play an important role in tuning the selectivity, and hence frequency resolution, of the receptor DA pair. Cru- cially, this cannot be predicted by the semiclassical MJ rates, which do not depend on the environmental cut- off, and predict a significant enhancement of tunnelling even for odorant vibrational frequencies relatively far off- resonant with the DA energy gap of 200 meV (top panel). The low-frequency environment regime analysed in pre- vious works -- to which we have shown the MJ rates to apply under strong mode dissipation -- is thus very sensi- tive the presence or absence of the odorant, but not nec- essarily to the odorant's specific vibrational spectrum. It is evident from Fig. 5, however, that increased fre- quency resolution comes at a cost. In order to achieve the more selective behaviour apparent with a larger en- vironmental cut-off (and strong odorant dissipation), the coupling to the continuum environment must be reduced. Furthermore, the switch suffers from intrinsically higher levels of noise, since the peak values of the ratio r also be- come smaller as the cut-off is increased. We can therefore conclude from our analysis that a tradeoff exists in which the combination of moderate to high frequency compo- nents in the environment and a lower reorganisation en- ergy is advantageous for odorant vibrational frequency resolution, though not necessarily for discrimination sim- ply between the presence and the absence of the odorant. It is worth noting also that the validity of the results de- rived from our polaron transformed Hamiltonian rely on the assumption λ (cid:29) V . It is not possible, therefore, to extrapolate Fig. 5 to yet lower reorganisation energies within the present analysis. Finally, we stress that if we move out of the regime of strong odorant dissipation, we find that both the fre- quency resolution of the receptor and its sensitivity to the presence or absence of the odorant decreases. This is illustrated in Fig. 6, which shows the ratio of the trans- fer time (to 90% population on the acceptor site) in the absence of the odorant to that in its presence for dif- Odorant
0
,
1
population02460.0.20.40.60.81.
i
02460.0.20.40.60.81.
ii
02460.0.20.40.60.81.
iii
02460.0.20.40.60.81.
iv
02460.0.20.40.60.81.
v
02460.0.20.40.60.81.
vi
02460.0.20.40.60.81.
vii
02460.0.20.40.60.81.
viii
02460.0.20.40.60.81.
ix
Time
ns
9 Figure 4. Effect of odorant mode dissipation. DA population difference pD(t) − pA(t) as predicted by Eqs. (9) and (22) for different odorant dissipation rates: γ0 = 0 (red dotted curves), γ0 = 1 ns−1 (red dashed curves), and γ0 = 1000 ns−1 (red solid curves). Exponential ET dynamics gov- erned by the rates given in Eq. (18), without mode dissipation, are also shown (points) along with the dynamics predicted by MJ theory (blue squares) for comparison. The bath cut-off frequency ωc increases from top to bottom, with ωc = kBT /10 in panel (i), ωc = kBT in panel (ii) and ωc = 2kBT in panel (iii). The reorganisation energy is λ = 15 meV in each case and the odorant frequency is taken to be on resonance with the DA splitting, ω0 =
D −
A = 200 meV. ferent levels of odorant dissipation. We use the transfer time to allow a meaningful comparison here, since there is no single rate that accurately describes the dynamics outside the strongly dissipative case. As the dissipation rate is decreased, the height of the peak also decreases, while its width remains similar (dotted curve) until the dissipation ceases to play a role (dashed curve). Thus, as the mode dissipation is weakened, the receptor frequency resolution is severely compromised, as well as its capa- bility to distinguish between the cases with and without Figure 5. Switch frequency resolution. Contour plots of r = (Γtot−Γ0) (Γtot+Γ0) as a function of odorant frequency ω0 and bath reorganisation energy λ. Top: ωc = kBT /10 (corresponding to the MJ limit). Middle: ωc = kBT . Bottom: ωc = 2kBT . Other parameters are given in Table I. Light regions corre- spond to r → 1 (good odorant discrimination) and the dark- est regions correspond to r → 0 (poor discrimination). The red dashed lines shows the DA energy splitting. the odorant present. VI. SUMMARY We have developed a dynamical theory of a molecular switch and applied it to investigate vibrationally-assisted ET in the context of a proposed olfactory process. We have shown that the dynamics of the olfactory receptor pD(t)-pA(t)■■■■■■■■■■■■●●●●●●●●●●●●0246-0.2-0.4-0.6-0.8-1.0.0.20.40.60.81.(i)■■■■■■■■■■■■●●●●●●●●●●●●0246-0.2-0.4-0.6-0.8-1.0.0.20.40.60.81.(ii)■■■■■■■■■■■■●●●●●●●●●●●●0246-0.2-0.4-0.6-0.8-1.0.0.20.40.60.81.(iii)Time(ns)10 VII. ACKNOWLEDGEMENTS A. C. would like to thank V. Vedral for fruitful discus- sions. A. N. is supported by Imperial College London and The University of Manchester. F. A. P. is supported by the Leverhulme Trust. A. C. was supported by the Na- tional Research Foundation and Ministry of Education, Singapore. APPENDIX: DERIVATION OF THE MASTER EQUATION In this Appendix we outline the general steps required to derive our polaron master equation. We start the anal- ysis by writing the polaron-transformed Hamiltonian as (see Sec. III) HP = U†HU = HSP + HB + HIP, (23) and moving into the interaction picture with respect to HSP and HB. To derive the master equation we focus on the interaction part of the Hamiltonian which takes the form HIP(t) =V(cid:0)D(cid:105)(cid:104)A(t)A+(t)B+(t) + A(cid:105)(cid:104)D(t)A−(t)B−(t)(cid:1) (cid:88) Sj(t)Bj(t), = j with j = {+,−} and S+(t) = V D(cid:105)(cid:104)Aei
tA+(t), S−(t) = V A(cid:105)(cid:104)De−i
tA−(t). (24) (25) By considering the Liouville-von Neumann equation of motion for the density matrix of the whole system in the interaction picture, χP(t), and tracing over the environ- ment, we find: (cid:90) t (cid:110)(cid:104) HIP(t), (cid:104) HIP(s), χP(s) (cid:105)(cid:105)(cid:111) . ρSP(t) = −i d dt dstrB 0 (26) Here, we have substituted in the solution to the full Liouville-von Neumann equation on the right-hand-side and have used tr{[ HIP(t), χP(0)]} = 0. We then make the standard Born-Markov approximations. We assume that the evolution of the system is time local, such that it depends only on its current state, and that in the trans- formed frame the effective interaction 'strength' is small, so that we may write χP(t) (cid:39) ρSP(t) ⊗ ρB at all times, where ρB is a thermal state of the environment. The Markov approximation amounts to replacing the state at time s in the integrand of Eq. (26) with the state at time t, changing variables such that s → t− τ, and taking the 0 ) to that in its presence (t90 Figure 6. Dissipation-assisted electron transfer. The ra- tio of the electron transfer time (to 90% acceptor population) in the absence of the odorant (t90 tot) as a function of odorant frequency, as predicted by Eqs. (9) and (22). We take λ = 10 meV, ωc = kBT , and all other parameters are from Table I. The three curves correspond to different levels of dissipation on the odorant: γ0 = 0 (red, dashed), γ0 = 1 ns−1 (blue, dotted) and γ0 = 1000 ns−1 (grey, solid). The blue points correspond to the same ratio calculated using Eq. (18), considering n = 0, 1. can differ dramatically from that predicted by semiclas- sical MJ theory, even for a low-frequency environment, if vibrational dissipation is weak. The ET dynamics can, however, be brought into agreement with a simpler rate analysis -- though not generally the MJ rates unless the semiclassical limit also applies -- provided that strong dissipation is allowed to act on the odorant vibrational mode. This results in an enhanced switching with respect to the dissipationless case. Furthermore, we have found that low frequency environments, to which the MJ rates apply under strong odorant dissipation, do not provide good odorant frequency resolution in the ET rates. By modifying environmental parameters to move beyond the semiclassical limit it is, nevertheless, possible to substan- tially increase this resolution and thus select for odorants of a particular frequency. While our present model is motivated by the problem of describing olfaction in biological systems, it actually corresponds to a wide variety of physical settings in which vibrationally-assisted transport processes are important. Examples include nano-mechanical oscillators coupled to two-level systems,63 such as quantum dots64 and super- conducting qubits,65,66 as well as other biological systems such as the photosynthetic reaction centre in certain or- ganisms.67,68 Our results may also be especially relevant for the development of artificial molecular sensors, which could use the principles we have described in their design to aid in distinguishing chemical species based on their vibrational spectra.34,69,70 (cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)100150200250300010203040506070Ω0
meV
t090
ttot90(cid:104) HIP(t − τ ), ρSP(t)ρB (cid:105)(cid:105)(cid:111) . (27) 0 dτ (cid:90) ∞ (cid:110)(cid:104) HIP(t), (cid:90) ∞ dτ integral to infinity: ρSP(t) = − i d dt × trB ρSP(t) = −(cid:88) (cid:18) d dt In terms of the operators S±(t) this now reads 0 j,j(cid:48) [Sj(t), Sj(cid:48)(t − τ )ρSP(t)]Cjj(cid:48)(τ ) × + [ρSP(t)Sj(cid:48)(t − τ ), Sj(t)]Cj(cid:48)j(−τ ) (cid:19) , (28) (29) where (cid:0)Bj(τ )Bj(cid:48)(0)ρB (cid:1), Cjj(cid:48)(t) = trB and we have used [ρB, HB] = 0. There are actually two types of correlation function, C±±(τ ) and C±∓(τ ), but one of them vanishes (C±±(τ ) → 0) and the other sat- isfies C±∓(−τ ) = [C±∓(τ )]∗. We therefore arrive at our polaron master equation dτ (cid:90) ∞ (cid:26)(cid:0)[S+(t), S−(t − τ )ρSP(t)]+ + [S−(t), S+(t − τ )ρSP(t)](cid:1)C(τ ) + H.c. × 0 (cid:27) , ρSP(t) = − d dt (30) where we have written C(τ ) ≡ C±∓(τ ). This is equiva- lent to Eq. (9) in the main text. REFERENCES 1A. Nitzan, Chemical Dynamics in Condensed Phases: Relax- ation, Transfer and Reactions in Condensed Molecular Systems (Oxford University Press, 2006). 2V. May and O. Kühn, Charge and Energy Transfer Dynamics in Molecular Systems (Wiley, 2011). 3N. Lambert, Y.-N. Chen, Y.-C. Cheng, C.-M. Li, G.-Y. Chen, and F. Nori, Nature Phys. 9, 10 (2013). 4Y.-C. Cheng and G. R. Fleming, Annu. Rev. Phys. Chem. 60, 241 (2009). 5A. Olaya-Castro and G. D. Scholes, Int. Rev. Phys. Chem. 30, 49 (2011). 6S. F. Huelga and M. B. Plenio, Procedia Chem. 3, 248 (2011). 7H.-P. Breuer and F. Petruccione, The theory of open quantum systems (Oxford University Press, 2002). 8A. Ishizaki, T. R. Calhoun, G. S. Schlau-Cohen, Fleming, Phys. Chem. Chem. Phys. 12, 7319 (2010). 9A. Garg, J. N. Onuchic, and V. Ambegaokar, J. Chem. Phys. 83, 4491 (1985). and G. R. 10C. Olbrich, J. Strümpfer, K. Schulten, and U. Kleinekathöfer, J. Phys. Chem. Lett. 2, 1771 (2011). 11 11S. Shim, P. Rebentrost, S. Valleau, and A. Aspuru-Guzik, Bio- 12M. d. Rey, A. W. Chin, S. F. Huelga, and M. B. Plenio, J. Phys. phys. J. 102, 649 (2012). Chem. Lett. 4, 903 (2013). 13A. W. Chin, J. Prior, R. Rosenbach, F. Caycedo-Soler, S. F. Huelga, and M. B. Plenio, Nature Phys. 9, 113 (2013). 14E. J. O'Reilly and A. Olaya-Castro, Nature Commun. 5, 3012 15E. K. Irish, R. Gómez-Bombarelli, and B. W. Lovett, Phys. (2014). Rev. A 90, 012510 (2014). Phys. 13, 113034 (2011). 16, 053018 (2014). 16G. Ritschel, J. Roden, W. T. Strunz, and A. Eisfeld, New J. 17J. Lim, M. Tame, K. H. Yee, J.-S. Lee, and J. Lee, New J. Phys. 18A. Kolli, E. J. O'Reilly, G. D. Scholes, and A. Olaya-Castro, J. Chem. Phys. 137, 174109 (2012). 19N. Christensson, H. F. Kauffmann, T. Pullerits, and T. Mančal, J. Phys. Chem. B 116, 7449 (2012). 20A. Chenu, N. Christensson, H. F. Kauffmann, and T. Mančal, Sci. Rep. 3, 2029 (2013). 21L. Turin, Chem. Senses 21, 773 (1996). 22L. Turin, J. Theor. Biol. 216, 367 (2002). 23A. Keller and L. B. Vosshall, Nature Neurosci. 7, 337 (2004). 24J. C. Brookes, F. Hartoutsiou, A. P. Horsfield, and A. M. Stone- ham, Phys. Rev. Lett. 98, 038101 (2007). 25T. P. Hettinger, PNAS 108, E349 (2011). 26M. I. Franco, L. Turin, A. Mershin, and E. M. C. Skoulakis, PNAS 108, 3797 (2011). 27I. A. Solov'yov, P.-Y. Chang, and K. Schulten, Phys. Chem. Chem. Phys. 14, 13861 (2012). 28P. Kovacic, J. Electrostat. 70, 1 (2012). 29E. R. Bittner, A. Madalan, A. Czader, and G. Roman, J. Chem. Phys. 137, 22A551 (2012). 30S. Gane, D. Georganakis, K. Maniati, M. Vamvakias, N. Ragous- sis, E. M. C. Skoulakis, and L. Turin, PLoS ONE 8, e55780 (2013). 31D. J. Rowe, Chemistry and technology of flavors and fragrances (Blackwell, Oxford, 2005). 32R. Axel, Angew. Chem. Int. Ed. 44, 6110 (2005). 33L. B. Buck, Angew. Chem. Int. Ed. 44, 6128 (2005). 34S. H. Lee, O. S. Kwon, H. S. Song, S. J. Park, J. H. Sung, J. Jang, and T. H. Park, Biomaterials 33, 1722 (2012). 35R. H. Farahi, A. Passian, L. Tetard, and T. Thundat, ACS Nano 6, 4548 (2012). 36R. B. Silverman, The Organic Chemistry of Enzyme-catalyzed Reactions (Academic Press, London, 2002). 37L. Haffenden, V. Yaylayan, and J. Fortin, Food Chem. 73, 67 (2001). 38G. M. Dyson, Chem. Ind. 57, 647 (1938). 39R. H. Wright, J. Theor. Biol. 64, 473 (1977). 40R. H. Wright, The sense of smell (CRC Press, 1982). 41J. Lambe and R. C. Jaklevic, Phys. Rev. 165, 821 (1968). 42C. J. Adkins and W. A. Phillips, J. Phys. C 18, 1313 (1985). 43R. A. Marcus, J. Chem. Phys. 24, 966 (1956). 44R. A. Marcus, Annu. Rev. Phys. Chem. 15, 155 (1964). 45R. A. Marcus, J. Chem. Phys. 43, 679 (1965). 46R. Marcus and N. Sutin, Biochim. Biophys. Acta. 811, 265 (1985). 47M. Bixon and J. Jortner, Adv. Chem. Phys. 106, 35 (1999). 48T. Holstein, Ann. Phys. 8, 325 (1959). 49T. Holstein, Ann. Phys. 8, 343 (1959). 50B. Jackson and R. Silbey, J. Chem. Phys. 78, 4193 (1983). 51A. Nazir, Phys. Rev. Lett. 103, 146404 (2009). 52S. Jang, Y.-C. Cheng, D. R. Reichman, and J. D. Eaves, J. Chem. Phys. 129, 101104 (2008). 53S. Jang, J. Chem. Phys. 131, 164101 (2009). 54S. Jang, J. Chem. Phys. 135, 034105 (2011). 55D. P. S. McCutcheon and A. Nazir, New J. Phys. 12, 113042 (2010). 56D. P. S. McCutcheon and A. Nazir, Phys. Rev. B 83, 165101 57A. Kolli, A. Nazir, and A. Olaya-Castro, J. Chem. Phys. 135 58D. P. S. McCutcheon and A. Nazir, J. Chem. Phys. 135, 114501 59L. Buck and R. Axel, Cell 65, 175 (1991). 60For ω0 = 200 meV and T = 300 K this corresponds to the odorant initial state being essentially the ground state, and is therefore consistent with the initial condition in the rate analysis. 61The DA dynamics is generally well fitted by a biexponential form. 62Note that the ET rates used to calculate the dashed curves in panels (i - iii) of Fig. 2 do agree with the MJ rates. 63L. G. Remus and M. P. Blencowe, Phys. Rev. B 86, 205419 (2011). (2011). 154112 (2011). (2012). 12 64N. Lambert and F. Nori, Phys. Rev. B 78, 214302 (2008). 65A. D. Armour and M. P. Blencowe, New J. Phys. 10, 095004 66A. D. Armour and M. P. Blencowe, New J. Phys. 10, 095005 (2008). (2008). 67D. Xu and K. Schulten, Chem. Phys. 182, 91 (1994). 68R. E. Blankenship, Molecular mechanisms of photosynthesis 69S. H. Lee, H. J. Jin, H. S. Song, S. Hong, and T. H. Park, J. (Blackwell, Oxford, 2002). Biotechnol. 157, 467 (2012). 70L. Du, C. Wu, Q. Liu, L. Huang, and P. Wang, Biosens. Bio- electron. 42, 570 (2013). 

1803.11085

1

1803

Non-Gaussian Current Fluctuations in a Short Diffusive Conductor

We report the measurement of the third moment of current fluctuations in a short metallic wire at low temperature. The data are deduced from the statistics of voltage fluctuations across the conductor using a careful determination of environmental contributions. Our results at low bias agree very well with theoretical predictions for coherent transport with no fitting parameter. By increasing the bias voltage we explore the cross-over from elastic to inelastic transport.

cond-mat.mes-hall

cond-mat

a Non-Gaussian Current Fluctuations in a Short Diffusive Conductor Edouard Pinsolle,1 Samuel Houle,1 Christian Lupien,1 and Bertrand Reulet1 1Institut Quantique, D´epartement de Physique, Universit´e de Sherbrooke, Sherbrooke, Qu´ebec J1K 2R1, Canada. (Dated: June 30, 2021) We report the measurement of the third moment of current fluctuations in a short metallic wire at low temperature. The data are deduced from the statistics of voltage fluctuations across the conductor using a careful determination of environmental contributions. Our results at low bias agree very well with theoretical predictions for coherent transport with no fitting parameter. By increasing the bias voltage we explore the cross-over from elastic to inelastic transport. Over the years the study of current fluctuations, and in particular their variance (or second moment) (cid:104)δI 2(cid:105), has given new insights in the properties of quasi-particles in conductors. The study of the symmetry of the Kondo state in carbon nanotubes [1] or the observation of heat quantization [2] are recent examples of such measure- ments. Despite this success there has been only a few at- tempts to push deeper the study of current fluctuations in mesoscopic conductors by tackling the measurement of higher order moments such as the third one (cid:104)δI 3(cid:105). As far as we know, such measurements have been per- formed only in tunnel junctions [3 -- 6], quantum dots [7 -- 10] and avalanche diodes [11], revealing physics hidden in the study of the second moment such as the coupling with the environment [3], the dynamics of quantum noise [5] or the ordering of operators in a quantum measure- ment [6]. Higher order moments of current fluctuations in even the simplest system, an electrical wire, has never been probed experimentally. This letter reports such a measurement. The variance of current fluctuations in a diffusive wire has been calculated using many techniques [12 -- 14], all providing the same answer for the spectral density of cur- rent fluctuations SI 2 measured at temperature T with a voltage bias V : SI 2 = 2 3 2kBT R + 1 3 eV R coth eV 2kBT , (1) where R is the sample resistance, e the electron charge and kB the Boltzmann's constant. This result indi- cates the existence of shot noise with a Fano factor F2 = e−1dSI 2/dI = 1/3 at large bias V (cid:29) kBT /e, which has been confirmed experimentally [15]. The reduction of the Fano factor as compared to that of a tunnel junc- tion, F2 = 1, is interpreted in the quantum theory as stemming from the existence of well transmitting chan- nels and in the semi-classical theory from the existence of a position-dependent distribution function. The third moment of current fluctuations has also been calculated by several theories [14, 16 -- 20] which at low frequency all yield to the same spectral density SI 3 given by: SI 3 = 1 15 e2I + 12 5 kBT dSI 2 dV . (2) This result differs from that of a tunnel junction SI 3 = e2I on two main factors: first it depends on temperature; second it has a much lower Fano factor at high voltage, F3 = e−2dSI 3/dI = 1/15 instead of F3 = 1 for the tun- nel junction. While this prediction comes in the quantum regime again from the statistical distribution of transmis- sions, the semi-classical prediction involves a "cascade" or feedback mechanism similar to that explaining envi- ronmental contributions. Eq. (2) corresponds to a measurement performed with a noiseless voltage bias and an ammeter, i.e. an appa- ratus with an input impedance much lower than that of the sample. This situation can be achieved with a high impedance sample, but a typical metallic wire has a low impedance and one has to consider the effects of both the finite impedance of the environment, here a resis- tance RA (the input impedance of the amplifier), and the current noise experienced by the sample, here gener- ated by the amplifier used to detect current fluctuations and described by a noise spectral density SA. These con- tributions to the spectral density of the variance of volt- age fluctuations across the amplifier's input resistance are simply given by: SV 2 = R2 D (SI 2 + SA) , (3) with RD = RRA/(R + RA). Here the environment only renormalizes the noise generated by the sample and adds a contribution which does not depend on the tempera- ture or bias voltage. In contrast, environmental effects are much more subtle on the third moment of voltage fluctuations. They have been thoroughly studied both theoretically [21, 22] and experimentally [3, 4] and obey: SV 3 = −R3 DSI 3 + 3R4 D(SA + SI 2) dSI 2 dV . (4) As a consequence, a reliable way to characterize all the environmental terms is required to extract the intrinsic third moment of current fluctuations SI 3. In the following we show the measurement of SI 3 for a short metallic wire placed at very low temperature. We describe the experimental setup and the results for SV 3, the calibration of the environmental contributions using 2 FIG. 1. Schematics of the experimental setup. A/D repre- sents a 14 bits, 400 MSample/s digitizer. a tunnel junction and the results for SI 3. We compare these results with the theoretical prediction of Eq. (2) in the elastic transport regime. We also report measure- ments performed at high bias that explore the crossover to the inelastic regime. Experimental setup. The sample is a 1 µm long, 10 nm wide, 165 nm thick Aluminum (Al) wire of resistance Rw = 30.5 Ω. Its contacts, also made of Al, are much larger (400 µm×400 µm) and thicker (200 nm) to make sure they behave as good electron reservoirs [23]. An Al tunnel junction of resistance Rj = 34 Ω is used as a reference to calibrate the setup. Both samples have been made by e-beam lithography and the metal has been deposited by double angle evaporation [24]. The experimental setup is presented in Fig. 1. The samples are placed on the 7 mK stage of a dilution refrigerator. They are kept in their normal, non superconducting state with the help of a strong Neodymium permanent mag- net. The two samples are connected to a cryogenic mi- crowave switch which allows us to measure either of them without changing anything in the detection circuit. They are dc current biased through the dc port of a bias-tee and ac coupled to a cryogenic microwave amplifier in the range 40 MHz-1 GHz. The use of a cryogenic amplifier both optimizes the signal to noise ratio and minimizes the noise experienced by the sample which leads to en- vironmental contributions. The signal is further ampli- fied at room temperature in order to achieve a level high enough for digitization. Non-linearities in the detection are very detrimental since they lead to strong artifacts. Despite the use of ultra-linear amplifiers, non-linearities still give rise to a contribution which is an even function of I in the sample. We simply remove this by considering [SV 3(I) − SV 3(−I)]/2. After amplification the signal is digitized by a 14 bit, 400 MS/s digitizer with a 1 GHz analog bandwidth. We measure real-time histograms of FIG. 2. Third moment of voltage fluctuations SV 3 for the wire (purple) and the tunnel junction (orange) vs. I. Symbols are experimental data; the solid line is a fit for the tunnel junction using Eq. (4) with SI3 = e2I, from which the environmental parameters are deduced. Inset: the second moment of current fluctuations SI2 vs. I (the noise of the amplifier has been subtracted for the sake of clarity). the signal from which moments are computed. Results: elastic transport. In the inset of Fig. 2 we present the measurement of SV 2 for the tunnel junction (orange symbols) and the wire (purple). From the high current slope of SV 2 vs. I for the tunnel junction we find the gain of the setup. Then, we deduce the Fano factor of the wire F2 = 0.35± 0.02, in good agreement with the theoretical value of 1 3 in Eq. (1). This ensures that elec- tron transport is elastic in the sample, in agreement with other measurements of similar wires [15, 25]. From SV 2 we also deduce the electron temperature for the wire and for the tunnel junction, as well as the noise temperature of the amplifier Ta (cid:39) 7.5 K. Values of the temperature indicated in the various figures correspond to electronic temperatures deduced from the measurements of SV 2. Fig. 2 presents the measurement of SV 3 vs. I for the tun- nel junction and the wire at a temperature around 650 mK (we have performed similar measurements down to 120 mK). As a consequence of small Fano factors F2 and F3 the signal is much weaker for the wire, but the signal- to-noise ratio clearly allows for a reliable comparison with theory (each point is averaged for 20 min). Following the procedure of [26], we use the measurements performed at all temperatures on the tunnel junction to extract the parameters that characterize the environment, i.e. the amplifier impedance RA = 44.8 Ω and the effective envi- ronmental noise temperature T ∗ 0 = 0.54 K. An example of fit on the tunnel junction SV 3 is provided as a solid line on Fig. 2. From the knowledge of the environmen- tal parameters we can extract the intrinsic third moment of current fluctuations in the wire using Eq. (4). The JunctionWireA/D40MHz-1GHz201510505101520Current (µA)1.00.50.00.51.0SV3 (×10−39 V3/Hz2)Junction T=640 mKWire T=660 mK20020Current (µA)024SV2 (×10−21 V2/Hz)3 FIG. 3. Intrinsic third moment of current fluctuations SI3 vs. I for the wire at different temperatures. Symbols are experimental data, lines are the theoretical expectations of Eq. (2). Inset: SI3 for the tunnel junction (orange) and the wire (purple) for higher currents up to 0.3 mA at T ∼ 640 mK. corresponding results are plotted in Fig. 3. The theo- retical predictions of Eq. (2) are plotted as solid lines with no fitting parameters. A clear agreement between experiment and theory is achieved at all temperatures for the current range explored. At low current (eV < kBT ) we observe that all curves merge, which corresponds to a Fano factor F3 = 1/3 independent of temperature (the fact that the Fano factor of SI 3 at low current is the same as that of SI 2 at high current has been predicted to come from the Pauli principle [14, 19]). At high current, we ob- serve that SI 3 grows linearly with current with a slope F3 = 1/15 that is temperature independent. However there is a constant shift which increases with tempera- ture. In this limit, both SI 2 and SI 3 increase linearly with temperature, while for a tunnel junction none of them do. From elastic to inelastic transport. Electronic trans- port in short samples at very low energy (voltage and temperature) is elastic. When energy is increased, inelas- tic processes are more and more effective, starting with electron-electron interactions which tend to thermalize the electrons among themselves, followed by electron- phonon interactions which tend to thermalize the elec- trons to the phonon bath of the substrate. This man- ifests itself in the variance of current fluctuations in a wire by the Fano factor F2 going from 1/3 (cid:39) 0.33 in 3/4 (cid:39) 0.43 in the hot electron the elastic regime to regime (strong electron-electron, no electron-phonon in- teraction) and decaying to zero as the electron-phonon interaction becomes effective [27]. We show in Fig. 4(a) a lin-log plot of F2 for our wire vs I. F2 is obtained by taking the numerical derivative of SI 2 vs. I after smoothing of the experimental data. After a sharp rise √ FIG. 4. a) Fano factor of the second moment of current fluc- tuation F2 = e−1dSI2 /dI for the wire vs. I. b) Fano factor of the third moment of current fluctuations F3 = e−2dSI3 /dI for the wire vs. I. Different symbols correspond to different tem- peratures ranging from 130 mK to 660 mK. The black lines correspond to the theoretical prediction in the elastic regime. Dashed (dotted dashed) line corresponds to the current for which Le−e = L (Le−ph = L). when eV < kBT , F2 has a short plateau at ∼ 0.35. This corresponds to the elastic regime described by Eq. (1), shown as solid lines in the figure. At higher bias F2 slowly increases up to ∼ 0.37 at I ∼ 80 µA, followed by a de- cay down to 0.28 at I = 0.3 mA. We could not apply a stronger current in the sample without taking the risk to damage it. We do not observe a plateau at F2 (cid:39) 0.43 that would correspond to inelastic scattering being dominated by electron-electron interactions. Inelastic processes should also affect the third moment of current fluctuations. We show in Fig. 4(b) the Fano factor F3 vs. I, deduced from SI 3 . The theoretical pre- diction of Eq. (2) (elastic transport) corresponds to F3 going from 1/3 at eV < kBT to 1/15, as shown as a solid line in Fig. 4(b). We clearly observe this transi- 201510505101520Current (µA)3210123SI3/e2 (µA)130 mK270 mK470 mK660 mK0100200300Current (µA)0100200SI3/e2 (µA)junctionwire100101102Current (µA)0.000.050.100.150.200.250.30F30.000.050.100.150.200.250.300.350.400.45F2Le−e=LLe−ph=L0.430.25b)a)Le−e=LLe−ph=L0.430.25b)a)Le−e=LLe−ph=L0.430.25b)a)Le−e=LLe−ph=L0.430.25b)a)130mK270mK470mK660mKtion in the experimental data, even though we explore the regime eV < kBT deep enough only at the high- est temperature, where the plateau F3 = 1/3 is visible. F3 in the hot electron regime has been predicted to be given by 8/π2 − 9/16 (cid:39) 0.25 while electron-phonon in- teraction is expected to lead to a vanishing F3 at high enough energy. We indeed observe an increase of F3 up to ∼ 0.14 at I ∼ 150 µA followed by a slow decrease down to F3 ∼ 0.10 at I = 0.3 mA. The crossover between the different regimes should occur when the length of the sample is of the order of the corresponding inelastic length, Le−e for electron- electron interaction or Le−ph for electron-phonon inter- action. Both decrease when the bias voltage or the elec- tron temperature is increased. Thus an increase of the current flowing through the sample tends to decrease the inelastic lengths. Two vertical lines in Fig. 4 represent the calculated current for which L = Le−e (left) and L = Le−ph (right). Hence the left part of the plots cor- responds to the elastic regime (L > Le−e, Le−ph), the region between the vertical lines correspond to the hot electron regime (Le−e < L < Le−ph) and the right re- gion to the phonon cooled regime (L > Le−e, Le−ph). The intermediate region where electron-electron interac- tion dominates is clearly very narrow, which explains why we never observe the expected values of the Fano factors calculated in the hot electron regime. To our knowledge the crossover between the elastic and electron-electron regimes has never been theoretically studied. Deep in the electron-phonon regime, the noise in a wire is under- stood as Johnson-Nyquist noise of a sample at a uniform electronic temperature T determined by the balance be- tween Joule heating and phonon cooling, i.e.: (5) RI 2 = ΣΩ(cid:0)T n − T n (cid:1) , ph where Tph is the phonon temperature, Σ the electron phonon coupling constant, Ω the sample volume and n a number ranging from 4 to 6 depending on the sample purity. Thus the noise spectral density SI 2 = 2kBT /R increases with current with a decaying Fano factor F2 ∝ I 2/n−1 at large current. However SI 3 in the electron- phonon regime has been calculated to decrease to zero at high current (or zero phonon temperature Tph = 0) as [18, 28]: SI 3 ∝ k2 BR2/n−2 (ΣΩ)2/n I 4/n−1. (6) This result can be understood using the following sim- ple model, which gives the same result as a full calcula- tion using cascaded Boltzmann-Langevin equations with a position-dependent distribution function for the elec- trons. Let us consider the wire of length L (cid:29) Le−ph as many wires connected in series. Each wire exhibits thermal noise with zero third moment. However for each wire, all the others play the role of an electromagnetic 4 environment that has voltage-dependent noise, i.e. leads to environmental contributions. This immediately leads to: SI 3 = 3SI 2 dSI 2 dI , (7) which gives Eq. (6) for Johnson-Nyquist noise at a tem- perature T given by Eq. (5). In contrast with SI 2, SI 3 is expected to exhibit a maximum in current, then a de- cay at high current. We show in the inset of Fig. 3 our measurements of SI 3 vs. I up to 0.3 mA. We observe that SI 3 for the tunnel junction is strictly linear up to the highest current, as expected. The wire deviates from the linear behavior of Eq. (2) and barely shows any sign of saturation (but F3 decays slightly for the highest cur- rent, see Fig. 4(b)). For this sample Σ and n have been measured [29] and we expect to observe F3 ∝ I−6/5 at a current of I >∼ 0.5 mA, which we did not reach. It is noteworthy that taking n = 4 in Eq. (5) leads to a third moment of current fluctuations saturating at high current, and not decaying. Such a value of n has been observed in thin Au wires [30]. A saturation of SI 3 would however not violate the central limit theorem, neither when considering a sample of arbitrary length or I 2 ∝ I−3/4L−1/2. an arbitrary large current since SI 3 /S3/2 Conclusion. We have measured the third moment of current fluctuations in a wire, thus demonstrating that even the simplest conductor exhibits non-Gaussian noise. Our data at low voltage are in very good quantitative agreement with the theory. In particular we have found a Fano factor F3 = 1/15 characteristic of elastic transport in diffusive conductors. At higher current we recorded a clear deviation from the elastic regime, when inelas- tic scattering lengths become of the order of the sample length. Third moment of current fluctuations, whether intrinsic or environmentally induced, have been shown to affect decoherence in quantum dots [31] and should be taken into account to explain observed coherence times. Our experiment also demonstrates the possibility to mea- sure the third moment of current fluctuations with a great bandwidth ∼ 1GHz (and thus a great sensitiv- ity) in conductors of moderate resistance, while also get- ting rid of environmental contributions thanks to a cryo- genic calibration procedure. This opens the possibility to study many more systems where the statistics of electron transport is more complex. For example in the presence of proximity effect where multiple Andreev reflections have been predicted to lead to a diverging third moment [32, 33], in the vicinity of a phase transition [34] or in conductors where electron-electron interactions are more prominent. We acknowledge technical help of G. Lalibert´e. This work was supported by the Canada Excellence Research Chair program, the NSERC the MDEIE, the FRQMT via the INTRIQ, the Universit´e de Sherbrooke via the EPIQ and the Canada Foundation for Innovation. [1] M. Ferrier, T. Arakawa, T. Hata, R. Fujiwara, R. Dela- grange, R. Weil, R. Deblock, R. Sakano, A. Oguri, and K. Kobayashi, Nature Physics 12, 230 (2015). [2] S. Jezouin, F. D. Parmentier, A. Anthore, U. Gennser, A. Cavanna, Y. Jin, and F. Pierre, Science 342, 601 (2013). [3] B. Reulet, J. Senzier, and D. E. Prober, Phys. Rev. Lett. 91, 196601 (2003). [4] Y. Bomze, G. Gershon, D. Shovkun, L. Levitov, and M. Reznikov, Phys. Rev. Lett. 95, 176601 (2005). [5] J. Gabelli and B. Reulet, Phys. Rev. Lett. 100, 026601 (2008). [6] J. Gabelli and B. Reulet, Journal of Statistical Mechan- ics: Theory and Experiment 2009, P01049 (2009). [7] S. Gustavsson, R. Leturcq, B. Simovic, R. Schleser, and T. Ihn, P. Studerus, K. Ensslin, D. C. Driscoll, A. C. Gossard, Phys. Rev. Lett. 96, 076605 (2006). [8] S. Gustavsson, R. Leturcq, M. Studer, I. Shorubalko, T. Ihn, K. Ensslin, D. C. Driscoll, and A. C. Gossard, Surface Science Reports 64, 191 (2009). [9] E. V. Sukhorukov, A. N. Jordan, S. Gustavsson, R. Leturcq, T. Ihn, and K. Ensslin, Nature Physics 3, 243 (2007). [10] S. Gustavsson, R. Leturcq, T. Ihn, K. Ensslin, M. Rein- wald, and W. Wegscheider, Phys. Rev. B 75, 075314 (2007). [11] J. Gabelli and B. Reulet, Phys. Rev. B 80, 161203 (2009). [12] C. W. J. Beenakker and M. Buttiker, Phys. Rev. B 46, 1889 (1992). [13] K. E. Nagaev, Physics Letters A 169, 103 (1992). [14] P.-E. Roche, B. Derrida, and B. Dou¸cot, Eur. Phys. J. B 43, 529. [15] M. Henny, S. Oberholzer, C. Strunk, and 5 C. Schonenberger, Phys. Rev. B 59, 2871 (1999). [16] K. E. Nagaev, Phys. Rev. B 66 (2002). [17] J. Salo, F. W. J. Hekking, and J. P. Pekola, Phys. Rev. B 74, 125427 (2006). [18] D. B. Gutman and Y. Gefen, Phys. Rev. B 68, 035302 (2003). [19] P.-E. Roche and B. Dou¸cot, Eur. Phys. J. B 27, 393 (2002). [20] H. Lee, L. Levitov, and A. Y. Yakovets, Phys. Rev. B 51, 4079 (1995). [21] C. W. J. Beenakker, M. Kindermann, and Y. V. Nazarov, Phys. Rev. Lett. 90, 176802 (2003). [22] M. Kindermann, Y. V. Nazarov, and C. W. J. Beenakker, Phys. Rev. B 69, 035336 (2004). [23] F. Giazotto, T. T. Heikkila, A. Luukanen, A. M. Savin, and J. P. Pekola, Rev. Mod. Phys. 78, 217 (2006). [24] M. O. Reese, D. F. Santavicca, L. Frunzio, and D. E. Prober, IEEE Transactions on Applied Superconductiv- ity 17, 403 (2007). [25] H. Pothier, S. Gu´eron, N. O. Birge, D. Esteve, and M. H. Devoret, Phys. Rev. Lett. 79, 3490 (1997). [26] J. Gabelli, L. Spietz, J. Aumentado, and B. Reulet, New J. Phys. 15, 113045 (2013). [27] A. H. Steinbach, J. M. Martinis, and M. H. Devoret, Phys. Rev. Lett. 76, 3806 (1996). [28] B. Huard, H. Pothier, D. Esteve, and K. E. Nagaev, Phys. Rev. B 76, 165426 (2007). [29] E. Pinsolle, A. Rousseau, C. Lupien, and B. Reulet, Phys. Rev. Lett. 116, 236601 (2016). [30] S. I. Dorozhkin, F. Lell, and W. Schoepe, Solid State Communications 60, 245 (1986). [31] Y. M. Galperin, B. L. Altshuler, J. Bergli, and D. V. Shantsev, Phys. Rev. Lett. 96, 097009 (2006). [32] S. Pilgram and P. Samuelsson, Phys. Rev. Lett. 94, 086806 (2005). [33] J. C. Cuevas and W. Belzig, Phys. Rev. Lett. 91, 187001 (2003). [34] D. Bagrets and A. Levchenko, Phys. Rev. B 90, 180505 (2014). 

1412.2607

2

1412

Theory of Weyl orbital semimetals and predictions of several materials classes

Graphene, topological insulators, and Weyl semimetals are three widely studied materials classes which possess Dirac or Weyl cones arising from either sublattice symmetry or spin-orbit coupling. In this work, we present a theory of a new class of bulk Dirac and Weyl cones, dubbed Weyl orbital semimetals, where the orbital polarization and texture inversion between two electronic states at discrete momenta lend itself into protected Dirac or Weyl cones without spin-orbit coupling. We also predict several families of Weyl orbital semimetals including V$_3$S$_4$, NiTi3S6, BLi, and PbO$_2$ via first-principle band structure calculations. We find that the highest Fermi velocity predicted in some of these materials is even larger than that of the existing Dirac materials. The synthesis of Weyl orbital semimetals will not only expand the territory of Dirac materials beyond the quintessential spin-orbit coupled systems and hexagonal lattice to the entire periodic table, but it may also open up new possibilities for orbital controlled electronics or `orbitronics'.

cond-mat.mes-hall

cond-mat

Theory of Weyl orbital semimetals and predictions of several materials classes Kapildeb Dolui1, and Tanmoy Das1,2 1Graphene Research Center and Department of Physics, National University of Singapore, 2 Science Drive 3, 117542, Singapore 2Department of Physics, Indian Institute of Physics, Bangalore 560012, India. (Dated: August 30, 2018) 5 1 0 2 r p A 0 2 ] l l a h - s e m . t a m - d n o c [ 2 v 7 0 6 2 . 2 1 4 1 : v i X r a Graphene, topological insulators, and Weyl semimetals are three widely studied materials classes which pos- sess Dirac or Weyl cones arising from either sublattice symmetry or spin-orbit coupling. In this work, we present a theory of a new class of bulk Dirac and Weyl cones, dubbed Weyl orbital semimetals, where the orbital polar- ization and texture inversion between two electronic states at discrete momenta lend itself into protected Dirac or Weyl cones without spin-orbit coupling. We also predict several families of Weyl orbital semimetals including V3S4, NiTi3S6, BLi, and PbO2 via first-principle band structure calculations. We find that the highest Fermi ve- locity predicted in some of these materials is even larger than that of the existing Dirac materials. The synthesis of Weyl orbital semimetals will not only expand the territory of Dirac materials beyond the quintessential spin- orbit coupled systems and hexagonal lattice to the entire periodic table, but it may also open up new possibilities for orbital controlled electronics or 'orbitronics'. Keywords: Condensed Matter Physics, Weyl Orbital Semimetal, Orbitronics I. INTRODUCTION Dirac fermions emerge in solid state systems when the band structure is embedded with any two component quan- tum degrees of freedom such as spin or pseudospin under specific symmetry considerations.[1 -- 6] In graphene, gapless Dirac cones arise due to the sublattice symmetry of the hon- eycomb lattice when the sample dimension is reduced ex- actly to an atomically thin two-dimensional (2D) sheet.[1] In topological insulators, an 'inverted' bulk bandgap, typi- cally opened by spin-orbit coupling (SOC), renders Dirac ex- citations on the surface or boundary of the sample as long as time-reversal symmetry is held.[2, 3] Some of these spe- cial conditions required for the formation of Dirac fermions in graphene and topological insulators are lifted in the Weyl semimetal framework. In the latter family, Weyl cone is formed in the bulk band structure where bulk conduction and valence bands meet only at discrete momenta, and is protected by lattice symmetry.[4 -- 6] Because Weyl nodes are easily ac- cessible for room-temperature applications, tremendous re- search activities have been devoted in recent years for the pre- diction, discovery, and engineering of new Weyl semimetals families. The predicted materials so far include Iridates,[7] HgCr2Se4,[8, 9] A3Bi (A=Na, K, Rb),[10] β-cristobaline BiO2,[11] and also in engineered heterostructures[12, 13]. Experimentally, Cd3As2 [14 -- 16] and Na3Bi [17, 18] have been successfully synthesized to date as bulk Dirac materials. Because SOC is a common ingredient for the formation of Weyl cone in the existing Weyl semimetals and topo- logical insulators, the corresponding materials selection is limited to materials with heavy elements and non-magnetic ground state. However, for spintronics and other transport re- lated purposes, heavy-elements are less effective because they are prone to strong many-body interaction and magnetism which significantly enhance Dirac mass via renormalization and/ or band gapping.[19] Some of the other possibilities are pseudospintronics,[1, 20] and valleytronics [21 -- 23] where the quantum control of electric current is achieved by the definite polarization of the sublattice and 'valley' degrees of freedom, respectively. Here the precise lattice symmetry and structural confinement play the central role to achieve high degree of phase coherence for the electron transport. In this paper, we explore a different idea for the forma- tion of bulk Dirac cones and Weyl orbital semimetals with- out the need of SOC or structural confinement. We develop a generic low-energy theory for the Weyl orbital semimetals for various combinations of different orbitals, such as even and odd orbitals pair or bonding and antibonding states or pair of even or pair of odd orbitals or two different basis of same orbital, in variety of 3D lattices. The key ingredient in our theory roots in finding the optimum conditions in which the two orbitals states commence inverted band structure, and at the same time their inter-orbital electron hoppings obtain an odd function of energy-momentum dispersion, such that its low-energy Hamiltonian can be reduced to an effective k · p - type Hamiltonian. The resulting 3D Dirac or Weyl nodes at the orbital degenerate points are protected by orbital symme- try. By using density functional theory (DFT), we predict four distinct classes of materials which exemplify different mecha- nisms for generating Weyl orbital nodes. The predicted mate- rials have orbital components stemming from the weakly cor- related p or d orbitals, which, thereby, stipulates very high Weyl fermion velocity. For BLi, our first-principle estimation of Weyl fermion velocity is ∼ 2 × 106 m/s, which is even larger than graphene. Finally, we discuss the possibilities of obtaining orbitronics and quantum orbital Hall effect in the Weyl orbital semimetals. II. THEORY We derive the generalized theory of Weyl orbital semimet- als starting from a multiband tight-binding model without in- 2 j=1 d2 spectrum of Hk is E±(k) = ξ+(k) ±(cid:113)(cid:80)4 where ξ±(k) = (ξ1(k) ± ξ2(k))/2. ξ− k corresponds to d4 components, and the remaining dj vectors consist of any com- bination of the inter-basis hopping ξnm. Given that the energy j (k), pro- tected Dirac or Weyl nodes commence when all four d1−4 components vanish simultaneously at discrete momenta (ξ+ acts as the chemical potential shift to the overall band struc- ture, and is not 'in principle' required to vanish together). Around the nodal points, the above Hamiltonian can therefore be reduced to a general Weyl Hamiltonian[24] as Hqψ± q = ±vF q.Γψ± q along three momentum directions, where vF is the Weyl Fermi velocity, q to be measured from the nodal point. Here ψ± corresponds to Weyl fermions possessing op- posite chirality or winding numbers, and its net value vanishes in the whole Brillouin zone.[4 -- 6] For brevity, we denote dispersion ξj with numeric sub- script as intra-orbital term (ξ1,2), while English alphabetic subscripts give inter-orbital hopping terms (ξnm = ξa/b/c). For any pair of orbitals, the intra-orbital bands ξ1,2 have parabolic dispersions, and the Weyl nodes can appear at the loci of ξ1 = ξ2. This makes the Dirac mass (d4) to van- ish at a constant energy contour as shown in Fig. 1(a), which occurs in materials due to band inversion, and thus it is pa- rameter dependent. The remaining three d-vector components not only have to simultaneously vanish at discrete momenta on the same contour, but should also contain odd-parity inter- orbital hopping term and/or imaginary hopping term. Under such circumstances, the effective Hamiltonian near the nodal points can be expressed via Weyl Hamiltonian with linear dis- persion, as demonstrated in Fig. 1(b). When all di vanish at high-symmetry k-points, the corresponding odd-number of cone is called 3D Dirac cone, and when cones form at non- high-symmetry k-points, they come in pairs with opposite or- bital Chern numbers, which are called Weyl cones. The spinor basis for the Weyl fermions can be sought from multiple degrees of freedom, with one pair of different orbitals containing the above-mentioned inter-orbital hopping term, while the other two basis can arise from the sublattice symme- try, or from spin, or from the multiplets of the same orbitals. For example, in a C4 symmetric lattice, s and p orbitals can form a Weyl orbital pair in which pi multiplets fill in the re- maining basis, or between p and d orbitals in which px/y and dxz/yz are all degenerate along the zone diagonal direction, and so on. Similarly, owing to crystal inversion symmetry, a bonding state of two orbitals, and antibonding state of the same or different two orbitals can form a Weyl orbital pair. In Fig. 2, we demonstrate several representative combinations of orbital and crystal symmetries for 3D Dirac or Weyl orbital nodes. Subsequently, four predicted materials exemplify dif- ferent combinations of orbital symmetries which can lead to Weyl node formations. A. Specific examples FIG. 1. Schematic description of the formation of the Weyl orbital semimetal. (a) Contour of band touchings between two parabolic bands (without the inter-orbital hoppings). (b) As the odd-parity inter-orbital hoppings are turned on, Weyl nodes form at discrete mo- mentum points. voking SOC: H = −(cid:88) (cid:2)tn − (cid:88) s,n s,ν,n(cid:54)=m s cn† s±1 + µncn† s cn s s cn ν(cid:2)tnm 1s cn† s cm s+ν1 + tnm s cm s+ν2 (cid:3) . (1) (cid:3) 2s cn† s (cn Here cn† s ) is the creation (annihilation) fermionic opera- tor at a lattice site s, and n, m are orbital indices. The first term is for intra-orbital hoppings which consists of only near- est neighbor hopping, and the second term is the chemical potential. For the inter-orbital hoppings, we consider both the nearest neighbor (third term) and the next-nearest neighbor hopping (fourth term). This is because, for various orbital and lattice geometries, either of them can be the leading term, and can give rise to linear dispersion. The symmetry of the Hamiltonian will remain unchanged when higher order hop- ping parameters are included as it will be evident later. Since we are only interested in the inter-orbital hoppings which can give rise to linear dispersion, we restrict our dis- cussion to the odd parity hopping between different orbitals. For this purpose, we have introduced an index ν = ± which changes sign when the direction of hopping is reversed. This is the crucial part of our theory which can give imaginary hop- ping without SOC. With a Fourier transformation to the mo- mentum space, we obtain Hk = ξn(k)cn† k cn k + ξnm(k)cn† k cm k . (2) (cid:88) m(cid:54)=n (cid:88) n The first term in the above equation is the intra-orbital band dispersion for free fermions, while the second term consists of inter-orbital hopping integrals. Without loosing generality, Eq. 2 can be written in the basis of Dirac matrices Γj as[13] Hk = ξ+(k)Γ0 + ξ−(k)Γ4 + dj(k)Γj, (3) The eral intra-orbital case with band nearest dispersions neighbor for hopping the gen- become 3(cid:88) j=1 Orb.1 Orbital inversion at k-contour at d4=0 (a) (b) Orbital inversion at discrete k at d1-4=0 Orb. 2 3 FIG. 2. Various possible mechanisms and dispersion properties of Weyl orbital semimetals. (a, b) Even and odd orbitals or equivalently bonding and antibonding states lying along the zone boundary or diagonal directions, respectively, producing inter-orbital hoppings ξa/b. (c, d) For a pair of even or odd orbitals, a linear-in-momentum hopping term ξc develops. The choices of orbitals for even and odd symmetries in this figure are representative and more such combinations can be easily thought of. Red and blue arrows depict same phase and out-of phase electron tunnelings, respectively. (e-j) Various illustrative cases of the formation of Dirac or Weyl nodes. In (e), both inversion and Mirror symmetry are held which produces a single 3D Dirac cone at the Γ-point. ξa/b individually produce pairs of Weyl cones in (f, g), while ξc produces four Weyl cones in (h, i, j). In (j), we find that Weyl cones are possible along the non-high-symmetric directions. kj and kl are any reciprocal lattice axes. Blue to green color scale depicts the orbital weight in the dispersion spectrum. ξ1,2(k) = −2(cid:80) 2 a2 j k2 the lattice momentum. Equivalently, when even and odd states are placed diago- nally in a 2D kj − kl plane (similar situation also arises in the 3D case) as shown in Fig. 2(b), the resulting inter-band dispersion turns out to be linear along one momentum direc- tion (which is odd under reflection, say kl) as ξb(kj, kl) = 4itjl cos (kjaj) sin (klal) ∼ 4itjl(1 − 1 j )alkl. On the other hand, when we consider a pair of orbitals with same symmetry, their symmetry property prohibits the formation of linear dispersion along the zone boundary direc- tion. However, when either two even or two odd orbitals are placed equidistantly along the diagonal directions, anisotropic and odd functional electronic hybridization may arise between them as demonstrated in Figs. 2(c)-2(d). For a pair of even orbitals, the linear hybridization can occur between an s or- bital and any of the t2g state of the d orbital, or between the t2g and eg orbitals which are split in energy and mo- mentum by either different occupation numbers, or crystal field splitting or other effects as applicable. Since t2g or- bitals intrinsically break crystal rotational symmetry, its or- bital overlap term with an eg or s state becomes an odd func- tion of momentum. This is shown in Fig. 2(c) for an exam- j j=x,y,z t1,2 cos (kjaj) − µ1,2, and aj are the corresponding lattice constants. Here we discuss various different conditions under which the inter-orbital hopping terms ξa/b/c can obtain linear dispersion purely from the angular dependence of the orbital symmetry, without any specific sublattice symmetry, or SOC. First we discuss a combination of even and odd symmetric states (under spatial inversion) placed either along the Bril- louin zone boundary direction in Fig. 2(a), or along the diag- onal direction in Fig. 2(b). Even and odd states can arise from s or d, and p or f orbitals, respectively, or from bonding and antibonding combinations of any two orbitals or same orbitals from different sublattices (we use the case of the bonding and antibonding states for s orbitals in the discussion because of its simplicity). In both cases, the odd parity orbitals give odd functional hoppings to the even state sitting at the center un- der reflection (ν = ± in Eq. 1). The combination in Fig. 2(a) guarantees the corresponding inter-orbital hopping matrix el- ement to be ξa(kj) = 2itj sin (kjaj) ∼ 2itjajkj (near the nodal points). Here i is the imaginary number, j is the bond- ing direction, and tj is the hopping amplitude between the even and odd states separated by a distance of aj, and kj is - + - + + - + - - + + - + - + - + - + - + - + - s pi dxy s px py - + - + - + + + + - + + + - σg σu (a) (c) (d) (i) (j) Orb.1 Orb. 2 (b) (g) (f) (e) (h) jlajalkjkl. jl sin (kjaj) sin (klal) ∼ 4t(cid:48) ple case of dxy and s orbitals. Such a combination yields ξc(kj, kl) = 4t(cid:48) In- terestingly, for the same multiorbital setup along the diagonal direction, two orthogonal odd orbitals, such as px, py are al- lowed to hybridize. The corresponding dispersion, ξc, comes from a linear combination of π and σ bonds. It should be noted that dxz and dyz orbitals can mimic odd parity orbitals under inversion when projected onto the x -- y plane, or two an- tibonding states would also give a similar linear hopping term. Different combinations of ξa,b,c associated with different Dirac matrices Γ1,2,3 determine the number and location of possible Dirac and Weyl cones. ξa helps create nodal points along its propagating axis, while ξb simultaneously produces massless and massive Dirac/Weyl terms along the kl, and kj directions, respectively. The resulting Weyl nodes always come in multiple of two or merge into a 3D Dirac cone at the Γ point. On the other hand, ξc term produces Weyl nodes in multiple of n, in systems with Cn rotational symmetry, and maintain the translational symmetry of the lattice. Some other combinations of ξa,b,c, however, can sometimes gap out each other's nodal states. Some examples of such large possibility of Weyl orbital nodes in various setup are shown in Fig. 2 (lower panel). In Fig. 2(e), we present a single 3D Dirac cone at the Γ point for an illustrative combination of dj = −iξa(kj) (j = x, y, z). The 3D Dirac cone splits into pairs of Weyl points as the inversion symmetry is lifted. Pair of Weyl cones can be created by using either both ξa/b terms (Fig. 2(f)), or one of them and combine it with ξc (Fig. 2(g)) which results in con- trasting orbital texture inversions. Four Weyl cones can be created with various combinations; for example, for a combi- nation of ξa and ξc, and so on, as shown in Fig. 2(h). These four Weyl nodes can be rotated from the bonding directions toward the zone diagonal ones by breaking the correspond- ing symmetry assigned with the Γ3 matrix. Using the same combination as in Fig. 2(e), we get four Weyl cones along the zone diagonal directions when d1 = ξa(kn), d2 = ξa(kj), and d3 = ±iξa(kl) (kn is perpendicular direction to the kj, kl plane), see Fig. 2(i). Finally, Weyl cones can appear along non-high-symmetry directions, Fig. 2(j), for a choice of d1 + id2 = −(i/2)ξc(kj, kl) and d3 = ±i[ξa(kj) − ξa(kl)]. However, such Weyl nodes are not protected and can be gapped by disorder or perturbations. The details of the param- eters used in the above presentation are listed in the Appendix. The above examples demonstrate how two entangled de- grees of freedom arise purely from the orbital texture inver- sion at discrete momentum points for spinless fermions. Due to the conservation of orbital angular momentum across the Weyl nodes, they remain time-reversal invariant. In these senses, our theory of Weyl orbital semimetal is different from the hexagonal symmetry related graphene,[1] or one- dimensional polyacetylene,[25] or from mirror symmetry in- duced topological crystalline insulator.[26] Opening a band gap at the Weyl orbital points can lead to Weyl orbital topolog- ical insulators, which is reminiscence of weak Z2 topological insulator in time-reversal invariant systems.[2, 3] 4 III. AB-INITIO CALCULATIONS AND MATERIALS PREDICTIONS Electronic structure calculations are carried out by DFT method with the generalized gradient approximation (GGA) in the parametrization of Perdew, Burke and Ernzerhof (PBE) [27] as implemented in the Vienna ab-initio simulation package (VASP) [28]. Projected augmented-wave (PAW) [29] pseudo-potentials are used to describe core electrons. The conjugate gradient method is used to obtain relaxed geome- tries. We include Hubbard U corrections to the GGA calcula- tions (GGA+U) and consider the spherically averaged form of the rotationally invariant effective U parameter with U = 3.0 eV, 5.0 eV and 4.0 eV on the correlated V 3d, Ni 3d and Ti 3d orbitals, respectively. In the cases where GGA+ U is not considered, i.e. for BLi and PbO2, the calculation is repeated with a different Heyd-Scuseria-Ernzerhof (HSE06) [30] hy- brid functional to check for the band crossings, and found that the gapless Weyl cones are intact. In the HSE06 calculations, the same geometry as the PBE cases are used. For each sys- tem, we have explicitly checked that the SOC does not gap out the bulk Weyl nodes. Both atomic positions and cell parame- ters have been allowed to relax, until the forces on each atom are less than 5 meV/ A. The optimized lattice vectors and in- ternal coordinates of all the atoms are listed in supplementary Table SI. We chose both the electronic wavefunction-cutoff energy and k-mesh (for Brillouin zone sampling) such that the accuracy of a total energy convergence is less than 10−4 eV/unit-cell. The structural stabilities of these materials are investigated by calculating the phonon dispersion and the cohesive energy, Ecoh. Force constants are calculated for a 2×2×2 super- cell within the framework density functional perturbation the- ory [31] using the VASP code. Subsequently, phonon dis- persions are calculated using phonopy package [32], and the results are shown in supplementary Fig. S1. Finally, we find that all materials have negative cohesive energy, implying that structure can exist in bound state. A. 3D 'graphene' from sublattice symmetry We first discuss the band structure of the existing V3S4 compound,[33, 34] and the origin of bulk Weyl cones in this material. V3S4 has a monoclinic phase in the space group of C2/m (No. 12), Fig. 3(a). A susceptibility measurement in the polycrystalline sample of V3S4 reported an antiferromag- netic (AFM) transition with N´eel temperature TN ∼ 9 K and magnetic moment of ∼ 0.2 µB per V atom, and also a ferro- magnetic (FM) phase below T ∼ 4.2 K.[35]. Our GGA+U calculation shows that the FM phase has the lowest ground state energy (by 36 meV/f.u. with a magnetic moment of 2.1 µB per V atom). In the FM phase, our band structure cal- culation in Fig. 3(c) shows that all other kz-planes are 2D band insulators, except the kz = π/c one, which has a well-defined Weyl cone for the same spin state. Two inequivalent vanadium atoms are placed in the corner and interior basis of the lattice, as shown in Fig. 3(a), which obtain equal electron occupation 5 FIG. 3. Band structure and DOS of V3S4. (a) The crystal structure of V3S4, and (b) the corresponding reciprocal space. (c) Electronic band structure is shown along the high-symmetry momentum directions, with a blue to green color gradient map which dictates the corresponding orbital weight from two inequivalent V atoms. (d), Full dispersion of the two low-lying bands on the 'Z' plane, exhibit a single Weyl point along Z-I1 direction in the first quadrant. The blue to red color map has not meaning in this figure. (e), Low-energy DOS (black color), and partial DOS (blue and green) for the same two orbital states. The DOS is given in the unit of number of states per unit eV. number. Therefore, their same d-orbitals osculate near the Fermi level, which promote their inter atomic hybridization to follow the even-even orbital hopping principle described in Fig. 2(c) above. Interestingly, the characteristic band dis- persion of V3S4 resembles 2D graphene,[1] despite the differ- ences in the crystal structure and orbital contributions between them. Another characteristic difference between the 2D Dirac cone and bulk Weyl node is that in the former case, the cor- responding density of states (DOS) is linear-in-energy across the Dirac cone, while here it is quadratic in energy.[5] This is indeed evident in the computed DOS of V3S4, presented in Fig. 3(e), overlayed with the partial DOSs to demonstrate how the DOSs of the two V atoms change across the Weyl node. B. Weyl node induced by two even orbitals Our next example is a ternary transition metal sulphide NiTi3S6, in which a unusual anisotropic crystal field splitting of the Ti d-orbitals vanishes at discrete momenta and forms Weyl nodes, see Fig. 4. NiTi3S6 is a known material with a lattice structure belonging to the rhombohedral lattice with space group of R-3H (No. 148).[36] Ti has two electrons in its 3d orbitals, which are shared between the conduction t2g and valence eg orbitals, separated by the crystal field split- ting of the rhombohedral lattice. To delineate the origin of odd functional dispersion from the overlap matrix-element be- tween the two even orbitals, let us focus on the a − b plane of the lattice. The projected dxz orbital on the x -- y plane mim- ics the px-type orbital in the sense that its phase changes sign on both sides of the center position. On the other hand, dz2 orbital acts as a purely isotropic orbital on this plane. There- fore, the inter-orbital electron tunneling between them follows the principle depicted in Fig. 2(a), and creates orbitally polar- ized Weyl nodes. Our predicted Weyl cones in this material may perhaps be less compelling for functional use, however, it exemplifies a new mechanism of the Weyl orbital semimetal arising from the least expected momentum dependent crystal field splitting. However, strain or pressure can be applied to remove the additional electron/hole pockets from the Fermi level, if needed. C. Weyl node induced by two odd orbitals Finally, we discuss two materials BLi and PbO2 in Fig. 5 which show multiple Weyl cones, forming at the crossing points of two p orbitals states in their native phases. Both materials form orthorhombic lattice.[37] Owing to the or- thorhombic structure, the orthogonality between the px/y and pz orbitals is lifted and the inter-orbital electronic interaction is turned on. The bonding between the two orthogonal orbitals stems from a linear combination of both π and σ bondings, and thus is generally robust against strain. Our band structure calculations show that multiple Weyl cones are formed along the Brillouin zone boundary at kz = 0 and kz = π planes. The projected pz states on the x -- y plane appears isotropic to the px/y, which therefore promotes the corresponding inter- orbital hopping to follow the scenario described in Fig. 2(a) and/ or Fig. 2(b). In BLi, there are two Weyl cones almost overlapping with each other along the Γ-X direction, which are thus less sta- ble to impurity scattering. In PbO2, Weyl cones arise from O atoms and thus remain ungapped even after including the SOC. Because of the small atomic number of B and O atoms, the corresponding Weyl states are highly dispersive, and the 0 1 -1 V2 dxz V1 S a 0 0.04 DOS E (eV) b c Γ Y F II1 Z XX1 Y L Γ V1 dxz V2 V3S4 (a) (b) (c) 0 π π 0 -0.2 0.2 (d) E (eV) kx ky Tot. -50 50 0 E (meV) (e) 6 FIG. 4. Band structure and DOS of NiTi3S6. (a)-(b), Real space and reciprocal space representations of the crystal structure. (c), (d), (e), Corresponding electronic band structure along high-symmetry directions, in 2D plane for the two lowest energy bands, and low-energy DOS, respectively. corresponding DOSs are reduced, see Fig. 5(e) and 5(k). We find that the extra electron/hole-pockets from the Fermi level in BLi are removed using HSE06 functional (not shown). Fur- thermore, the band inversion strength for PbO2 is relatively weak. Using relaxed lattice constants, we find Weyl nodes in the present GGA calculation, while HSE06 functional gives a 3D Dirac cone, and the MBJ functional looses the band in- version. However, with small strain of about 3%, we find that the latter two functionals also give Weyl nodes along the same momentum directions as in GGA band structure. The bulk-boundary correspondence associated with the general form of Weyl Hamiltonian in transnational invariant lattice dictates that the Fermi surface on the edge state be- comes disconnected at the points where the bulk and edge states merge, creating truncated Fermi surface (s) or the so- called 'Fermi arc'.[5, 6] To ascertain that our predicted ma- terials indeed belong to the Weyl semimetal class, we have computed the surface state for a representative material PbO2 because of its structural simplicity. The slab of PbO2 are mod- eled with 001 surface containing 16 atomic layers of Pb. A 14 Avacuum is place at the surface to avoid the interaction between two consecutive supercell. The spectral weight map of the Fermi surface of the corresponding surface state indeed shows a 'Fermi arc' in Fig. 5(g). IV. WEYL FERMI VELOCITY AND POSSIBILITY OF ORBITRONICS In Fig. 6, we compare the velocity of all four Weyl or- bital semimetals predicted here, with the other known Dirac or Weyl materials. All Dirac materials having Dirac cones gener- ated as a manifestation of the SOC consist of 'post-transition metals' with higher atomic numbers. With increasing atomic number, the bandwidth decreases and the effects of correla- tion increases, which all conspire in a gradual reduction of the Fermi velocity. On the other hand, the Weyl orbital semimet- als are applicable to any combinations of orbitals, and does not depend on the lattice or atomic properties such as bulk band gap or SOC. This flexibility greatly helps expanding the territory of Dirac materials to very light atoms such as Li, B, C, O, S, V, Ti and Ni with Weyl fermion velocity larger by an order of magnitude. In fact, the Fermi velocity for BLi is found to be ∼2×106m/s, which is the highest among all Dirac materials known to date. Any two-component quantum degrees of freedom of elec- tron in a lattice can mathematically behave in the same way as spin does, and therefore they can be viewed as pseudospin. In this sense, two interlocked orbital degrees of freedom with texture inversion enter into the low-energy Dirac equation in precisely the same manner that real spin produces Dirac equation in topological insulators, or Weyl semimetals, or the pseudospin behaves in the graphene's Dirac equation. There- fore, the physical concepts relating to the spintronics, spin- orbitronics or pseudospintronics application also apply to the Weyl orbital semimetals. We expect orbitally polarized charge current or orbital current in the bulk of these materials, which are protected by a momentum-dependent phase difference φk (see inset to Fig. 6), generic to any Dirac electron motions.[1 -- 3, 20] The corresponding Berry curvatures for the two orbitals pick up opposite values, which can lead to quantum orbital Hall effect. The nature of impurity scattering protection for different quantum orbitronics cases is characteristically dif- ferent. In the present family, an electron can only scatter from one orbital state to another when the impurity vertex contains a corresponding anisotropic orbital-exchange matrix-element or if the electron dynamically passes through the momentum and energy of the Dirac cone. Another advantage of the Weyl 0 2 -2 Γ L B1B XQ P1 Z Ti dz2 Ti Ni a 0 0.06 DOS E (eV) b c (a) (b) (c) (d) Z Γ Ti dxz S F NiTi3S4 π2 π π -0.5 kx ky 0.5 0 E (eV) -50 50 0 E (meV) (e) -100 7 FIG. 5. Band structures and DOSs of isostructural BLi and PbO2. Upper panel shows results of BLi, while the bottom panel gives the same results for PbO2. In both systems, the orbital characters interchange between the px/y and pz orbitals of Li and O atoms, respectively, via protected Weyl points. Large dispersions of these light elements give rise to vanishingly small DOS, which is almost zero above the Fermi level in PbO2 in (e). (g) Spectral weight map of the surface state of PbO2 in (001) cut exhibits 'Fermi arc' behavior. orbital semimetal is that here the Dirac cone is even immune to time-reversal symmetry breaking, and a bulk gap can be en- gineered by the lattice distortion. Therefore, the generation, transport and detection of orbitally protected electric current may lead to new opportunities for orbitronics. Chiral orbital current in the Weyl semimetals can be detected by Kerr effect. Possibilities of orbitronics were predicted earlier in p-doped silicon[43] and p-doped graphene,[44] which are yet to be ob- served. V. CONCLUSIONS The presented new family of Weyl orbital semimetal has the ability to bypass many limitations imposed in other Dirac ma- terials including atomically thin graphene, topological insula- tors and Weyl semimetals under time-reversal invariance and SOC. An important advantage of the 3D Dirac cone than its 2D counterpart is that in the former setup the DOS is quadratic in energy which will be beneficial for designing faster transis- tors and hard drive with low energy consumption. The ac- cessibility of the Dirac fermions in topological insulators is subjected to the bulk band gap, which can be filled by ther- mal broadening. Such a limitation is not present in Weyl semimetal families.[4, 5] Due to the absence of these con- straints, the possible materials classes for the Weyl orbital states are, in principle, expanded to the entire periodic ta- ble. Moreover, both mechanical and chemical tuning induced band gap in the Weyl node are appealing features which can be useful to exciton condensation, photovoltaics and solar cell applications, and optoelectronic technology.[45] Finally, the discovery of Weyl orbital semimetals will open the door for a new field of orbitronics. Acknowledgments: We thank Hsin Lin, Su Ying Quek, Matthias Graf for valuable discussions and hospitality. The work is facilitated by NERSC computing allocation in the USA and the the GRC high-performance computing facility in Singapore. 0 2 -2 Li B a 0 0.01 Energy (eV) b c (a) (c) (d) Γ Γ X S Z U R Y T 0 1 -1 O px/y O pz Pb O a 0 1 DOS b c (b) (f) (h) (i) B pz B py (g) -0.4 0.4 -0.4 0.4 kx [π/a] ky [π/a] BLi PbO2 Γ X S Z U R Y T Γ 0 π π 0 -2 2 E (eV) kx ky -0.2 0.2 0 E (eV) DOS 0 π π kx ky 0 π kz -0.2 0.2 0 E (eV) (e) (j) (k) -50 50 0 E (eV) 8 j jl = ±150 meV, and tn(cid:54)=m For the demonstration of the emergence of Dirac or Weyl ferminons, we take a simple and minimal set of parameters for tn, µn, and tnm: tn=1,2 = 150 meV is taken to be same for all orbitals n, m and along any di- rections j, l. The chemical potential can be chosen in a way k banishes at the Γ point (µn = −6tn) or at any other that ξ− discrete momenta (µn = −6tn ± δ, where δ is a tunable num- In Fig. 1 of main text, we take µ1,2 = ∓0.9 eV for ber). the Dirac point at the Γ, and µ1,2 = ∓0.7 eV otherwise. All tight-binding parameters are kept same for all plots in Fig. 1. We explicitly write down the combinations of ξa,b,c chosen in Fig. 1 of the main text. In the following cases, we assume Dirac or Weyl cones are present in the kj and kl plane, and kn is the perpendicular axis. For Fig. 1E, the d-vectors are taken to be dj = −iξa(kj), where j = 1, 2, 3 corresponding to kj, kl and kn direction, or their various combinations. The choice of d-vector components are For Fig. 1(f) : d1 + id2 = 1 2 ξb(kj, kl), d3 = 1 2 ξc(kj, kl), or d3 = −iξa(kn), or d3 = − i 2 [ξb(kj, kl) + ξb(kn, kl)] . For Fig. 1(g) : d1 + id2 = ξa(kl), d3 = ξc(kj, kn). 1 2 1 2 For Fig. 1(h) : d1 + id2 = ξa(kn), d3 = ξc(kj, kl), ξc(kj, kl), d3 = −iξa(kn). (A1) or d1 + id2 = i 2 The above three cases give Weyl cones along the zone axis. We also provide two other cases, where Weyl cones appear along other directions when a point-group symmetry is bro- ken. In these cases, both inter-basis hoppings between 1 to 3 and 2, 3 are taken to have same sign, violating the symmetry associated with the Γ3 term. Such Weyl cones are probably not as stable as others. For Fig. 1(i) : d1 + id2 = [ξa(kn) + iξa(kj)] , d3 = ±iξa(kl). For Fig. 1(j) : d1 + id2 = − i 2 ξc(kj, kl), d3 = ±i [ξa(kj) − ξa(kl)] . (A2) FIG. 6. Fermi velocity of various classes of Dirac materials. Com- puted Fermi velocity at the Dirac cone (averaged over the two in- tersecting linear-dispersion) of the four Weyl orbital semimetals pre- dicted here are compared with various other experimentally verified Dirac materials. All SOC induced Dirac fermions in heavy-elements have Fermi velocity almost an order of magnitude lower than that of the Weyl orbital semimetals, and graphene. The horizontal coordi- nate gives the average atomic number ( ¯Z) of the elements contribut- ing to the Dirac cone. Gray and yellow shadings separate the two families of Dirac materials without and with SOC, respectively. The Fermi velocity data are taken for the surface states of the 2D topo- logical insulator HgTe/CdTe from Ref. [38], for the 3D topological insulator Bi2Se3 from [39], and for the topological crystalline insu- lators (Pb,Sn)Te from Ref. 40 and 41, β-Ag2Te from Ref. 42. The Fermi velocity at the 3D Dirac cone of the Weyl semiletals Cd3As2 is taken from Ref. 14 and 15, and for Na3Bi from Ref. 17 and 18. The data for the non-SOC induced Dirac cone in graphene is taken from Ref. 1. The inset figure schematically shows the possibility of obtaining orbitally polarized electronic current with an anisotropic phase difference, φk, protecting their quantized currents. Appendix A: Parameter sets for Fig. 2 We use Dirac matrices of the form Γ1,2,3 = σ1⊗σ1,2,3, and Γ4 = I ⊗ σ3, where σi are the Pauli matrices and I is 2×2 unity matrix. Appendix B: Cohesive energy calculation Cohesive energy of a composition, M=AxByCz, is defined as Ecoh = EM − xEA − yEB − zEC. (B1) EM is the total energy of the primitive cell of bulk M, while EA and EB and EC are the total energy per atoms of A, B, and C species, respectively, in their bulk form. x, y, and z are the numbers of A, B and C atoms, respectively, assembled in the primitive cell of M. In case of a binary material M=AxBy the last term in Eq (B1) is omitted. Cohesive energy of considered materials are listed in supplementary Table SII. [1] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, A. K. Geim, The electronic properties of graphene,, Rev. Mod. Phys. 81, 109 (2009). > + 𝑒𝑖φ𝑘 > BLi Graphene PbO2 NiTi3S6 V3S4 Cd3As2 Na3Bi HgTe Bi2Se3 (Pb,Sn)Te 0 1 2 0 30 60 90 𝑍 vF [×106 m/s] β-Ag2Te 9 FIG. 8. Supplementary Figure S2: Phonon spectrum. Since some of the materials we predict here are yet to be synthesized experi- mentally, we carry out the phonon spectrum calculation to study the stability of the lattice. For all materials, we find a stable phonon spectrum. [12] A. A. Burkov and L. Balents, Weyl Semimetal in a topological insulator multilayer, Phys. Rev. Lett. 107, 127205 (2011). [13] T. Das, Weyl semimetal and superconductor designed in an orbital-selective superlattice, Phys. Rev. B 88, 035444 (2013). [14] S. Borisenko et al., Experimental Realization of a Three- Dimensional Dirac Semimetal, arXiv:1309.7978. [15] M. Neupaneet al., Observation of a topological 3D Dirac semimetal phase in high-mobility Cd3As2 and related materi- als, arXiv:1309.7892. [16] M. N. Ali et al., The crystal and electronic structures the three-dimensional electronic analogue of of Cd3As2, graphene, Inorganic Chemistry 53, 4062 (2014). [17] Z. K. Liu et al, Discovery of a three-dimensional topological Dirac semimetal, Na3Bi, Science 343, 864 (2014). [18] S. Y. Xu, et al., Observation of a bulk 3D Dirac multi- plet, Lifshitz transition, and nestled spin states in Na3Bi, arXiv:1312.7624. [19] L. A. Wray et al., A topological insulator surface under strong Coulomb, magnetic and disorder perturbations, Nat. Phys. 7, 32 (2001). [20] D. Pesin and A. H. MacDonald, Spintronics and pseudospin- tronics in graphene and topological insulators, Nat. Mater. 11, 409 (2012). [21] A. Rycerz, J. Tworzydo, and C. W. Beenakker, J. Valley filter and valley valve in graphene, Nat. Phys. 3, 172 (2007). [22] D. Xiao, W. Yao, and Q. Niu, Valley-contrasting physics in graphene: Magnetic moment and topological transport, Phys. Rev. Lett. 99, 236809 (2007). [23] C. E. Nebel, Valleytronics: Electrons dance in diamond, Nat. Mater. 12, 690 (2013). [24] H. Weyl, Zeitschrift fur Physik 56, 330 (1929). [25] W. P. Su, J. R. Schrieffer, and A. J. Heeger, Solitons in polyaeetylene, Phys. Rev. Lett. 42, 1698 (1979). [26] L. Fu, Topological crystalline insulator. Phys. Rev. Lett. 106, 106802 (2011). FIG. 7. Supplementary Figure S1: Comparison of band struc- ture with and without spin-orbit coupling. Here we show the same band structure, plotted in the main figure, for all four materials stud- ied in this paper, but with and without including the spin-orbit cou- pling effect. Expectedly, spin-orbit coupling in these materials is small enough to gap the orbital degenerate point at the Weyl nodes, which provide further justification that the Weyl nodes are protected. [2] M. Z. Hasan and C. L. Kane, Colloquium: Topological insula- tors, Rev. Mod. Phys. 82, 3045 (2010). [3] X. L. Qi and S. C. Zhang, Topological insulators and supercon- ductors, Rev. Mod. Phys. 83, 1057 (2011). [4] A. A. Burkov, M. D. Hook, and L. Balents, Topological nodal semimetals, Phys. Rev. B 84, 235126 (2011). [5] A. M. Turner and A. Vishwanath, Beyond band insu- Topology of semi-metals and interacting phases, lators: arXiv:1301.0330. [6] O. Vafek, and A. Vishwanath, Dirac fermions in solids: From high-Tc cuprates and graphene to topological insulators and Weyl semimetals, Ann. Rev. Cond. Mat. Phys. 5, 83 (2014). [7] X. Wan. A. M. Turner, A. Vishwanath, and S. Y. Savrasov, Topological semimetal and Fermi-arc surface states in the elec- tronic structure of pyrochlore iridates, Phys. Rev. B 83, 205101. (2011). [8] G. Xu, H. Weng, Z. Wang, X. Dai, Z. Fang, Chern Semimetal and the Quantized Anomalous Hall Effect in HgCr2Se4, Phys. Rev. Lett. 107, 186806 (2011). [9] C. Fang, M. J. Gilbert, X. Dai, and B. A. Bernevig, Multi-Weyl Topological Semimetals Stabilized by Point Group Symmetry, Phys. Rev. Lett. 108, 266802 (2012). [10] Z. Wang et al., Dirac semimetal and topological phase transi- tions in A3Bi (A=Na, K, Rb), Phys. Rev. B 85, 195320 (2012). [11] S. M. Young et al., Dirac semimetal in three dimensions, Phys. Rev. Lett. 108, 140405 (2012). ΓYFLII1ZΓXY-0.200.2WOSOCWSOCΓXSYΓZURTZ-101ΓXSYΓZURTZ-0.300.3 E - EF (eV)ΓLB1BZΓXQFP1Z-0.500.5(a)(c)(b)(d)V3S4BLiPbO2NiTi3S6ΓYFLII1ZΓXX1Y0510 Frequency (THz)ΓXSYΓURTZ02040Frequency (THz)ΓXSYΓURTZ051015ΓLB1BZΓXQFP1Z0510(a) V3S4(b) BLi(d) NiTi3S6(c) PbO2[27] J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77, 3865 (1996). [28] G. Kresse, and J. Furthmuller, Efficient iterative schemes for ab-initio total-energy calculations using a plane-wave basis set, Phys. Rev. B 54, 11169 (1996). [29] G. Kresse, and D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method, Phys. Rev. B 59, 1758 (1999). [30] J. Heyd, G. E. Scuseria, M. Ernzerhof, Hybrid functionals based on a screened Coulomb potential, J. Chem. Phys. 118, 8207 (2003). [31] X. Gonze, Perturbation expansion of variational principles at arbitrary order. Phys. Rev. A 52, 1086-1095 (1995). [32] A. Togo, F. Oba, I. Tanaka, First-principles calculations of the ferroelastic transition between rutile-type and CaCl2-type SiO2 at high pressures. Phys. Rev. B 78, 134106 (2008). [33] I. Kawada, M. Nakano-Onoda, M. Ishii, M. Saeki, M. Nakahira, Crystal structures of V3S4 and V5S8, J. Sol. Stat. Chem. 15, 246 (1975). [34] C. Mujica J. Llanos, and O. Wittke, Structure refinement of monoclinic V3S4, J. Alloy. Comp., 226, 136 (1995). [35] Y. Tazuke, T. Sato, and Y. Miyako, Susceptibility and specific heat studies on V3S4, J. Phys. Soc. Jap. 51, 2131 (1982). [36] P. Villars et al. (2010) in Structure Types. Part 8: Space Groups (156) P3m1 (148) R-3 (Villars, P. & Cenzual, K.) 612-612 (Springer Berlin Heidelberg). 10 [37] Y. C. Choi et al., Synthesis of crystalline lead di-oxide nanowires and their electron-beam-induced phase transforma- tion to oxygen deficient lead mono-oxide, J. Korean Phys. Soc. 51, 2045 (2007). [38] M. Kronig et al. Quantum spin Hall insulator state in HgTe quantum wells. Science 318, 766-770 (2007). [39] Y. Xia et al. Observation of a large-gap topological-insulator class with a single Dirac cone on the surface. Nat. Phys. 5, 398- 402 (2009). [40] Y. Tanaka et al. Experimental realization of a topological crys- talline insulator in SnTe. Nat. Phys. 8, 800-803 (2012). [41] S. Y. Xu et al. Observation of a topological crystalline insula- tor phase and topological phase transition in Pb1−xSnxTe. Nat Commun 3, 1192 (2012) . [42] S. Lee et al. Single crystalline β-Ag2Te nanowire as a new topo- logical insulator. Nano Lett. 12, 4194-4199 (2012). [43] B. A. Bernevig, T. L. Hughes and S. C. Zhang, Orbitronics: The intrinsic orbital current in p-doped silicon, Phys. Rev. Lett. 95, 066601 (2005). [44] I. V. Tokatly, Orbital momentum Hall effect in p-doped graphane Phys. Rev. B 82, 161404 (2010). [45] H. Wei, S. P. Chao and V. Aji, Excitonic phases from Weyl semimetals, Phys. Rev. Lett. 109, 196403 (2012). [46] Setyawan W, Curtarolo S (2010) High-throughput electronic structure calculations: challenges and tools. Comp. Mat. Sci. 49:299. [47] Curtarolo S, et al. (2012) AFLOWLIB.ORG: a distributed ma- terials properties repository from high-throughput ab initio cal- culations. Comp. Mat. Sci. 58:227. 11 Crystal structure a=6.65987 A, b=6.65987 A, c = 6.19743 A; α = 64.050◦, β = 64.050◦, γ = 30.759◦ a= 6.68709 A, b= 6.68709 A, c = 6.68709 A; α = 52.977◦, β = 52.977◦, γ = 52.977◦ a= 5.11933 A, b = 5.57092 A, c = 6.05971 A; α = 90.000◦, β = 90.000◦, γ = 90.000◦ a = 3.07497 A, b = 5.67574 A, c = 6.13950 A; α = 90.000◦, β = 90.000◦, γ = 90.000◦ Material V3S4 NiT3S6 PbO2 BLi Atoms x y z 0.000 0.000 0.000 V1 0.737 0.737 0.285 V2 0.263 0.263 0.715 V3 0.637 0.637 0.019 S4 0.363 0.363 0.981 S5 0.120 0.120 0.551 S6 0.880 0.880 0.449 S7 0.500 0.500 0.500 Ni1 0.000 0.000 0.000 Ti2 0.671 0.671 0.671 Ti3 0.329 0.329 0.329 Ti4 0.093 0.415 0.747 S5 0.415 0.747 0.093 S6 0.747 0.093 0.415 S7 0.907 0.585 0.253 S8 0.585 0.253 0.907 S9 0.253 0.907 0.585 S10 0.000 0.250 0.177 Pb1 0.500 0.750 0.323 Pb2 0.000 0.750 0.823 Pb3 0.500 0.250 0.677 Pb4 0.734 0.429 0.405 O5 0.766 0.929 0.095 O6 0.266 0.071 0.405 O7 0.234 0.571 0.095 O8 O9 0.266 0.571 0.595 O10 0.234 0.071 0.905 O11 0.734 0.929 0.595 O12 0.766 0.429 0.905 0.750 0.473 0.500 B1 0.250 0.973 0.000 B2 0.250 0.527 0.500 B3 B4 0.750 0.027 1.000 0.750 0.755 0.254 Li5 0.250 0.255 0.246 Li6 0.250 0.245 0.746 Li7 Li8 0.750 0.745 0.754 TABLE I. Supplementary Table SI. DFT relaxed crystal structure and atomic coordinates of different Weyl orbital semimetals. Materials Ecoh (eV) V3S4 B1Li PbO2 NiTi3S6 -9.27 -0.38 -2.48 -12.30 TABLE II. Supplementary Table SII. Theoretically calculated co- hesive energy for different Weyl orbital semimetal classes. 

1703.05678

1

1703

Role of Charge Traps in the Performance of Atomically-Thin Transistors

Transient currents in atomically thin MoTe$_2$ field-effect transistor are measured during cycles of pulses through the gate electrode. The transients are analyzed in light of a newly proposed model for charge trapping dynamics that renders a time-dependent change in threshold voltage the dominant effect on the channel hysteretic behavior over emission currents from the charge traps. The proposed model is expected to be instrumental in understanding the fundamental physics that governs the performance of atomically thin FETs and is applicable to the entire class of atomically thin-based devices. Hence, the model is vital to the intelligent design of fast and highly efficient opto-electronic devices.

cond-mat.mes-hall

cond-mat

Role of Charge Traps in the Performance of Atomically-Thin Transistors Iddo Amit1, Tobias J. Octon2, Nicola J. Townsend1, Francesco Reale3, C. David Wright2, Cecilia Mattevi3, Monica F. Craciun2 and Saverio Russo1† 1Centre for Graphene Science, Department of Physics, University of Exeter, Stocker Road, Exeter, 2Centre for Graphene Science, Department of Engineering, University of Exeter, North Park Road, United Kingdom, EX4 4QL Exeter, United Kingdom, EX4 4QF 3Department of Materials, Imperial College London, United Kingdom, SW7 2AZ †[email protected] Abstract Transient currents in atomically thin MoTe2 field-effect transistor are measured during cycles of pulses through the gate electrode. The transients are analyzed in light of a newly proposed model for charge trapping dynamics that renders a time-dependent change in threshold voltage the dom- inant effect on the channel hysteretic behavior over emission currents from the charge traps. The proposed model is expected to be instrumental in understanding the fundamental physics that governs the performance of atomically thin FETs and is applicable to the entire class of atomically thin-based devices. Hence, the model is vital to the intelligent design of fast and highly efficient opto-electronic devices. 7 1 0 2 r a M 6 1 ] l l a h - s e m . t a m - d n o c [ 1 v 8 7 6 5 0 . 3 0 7 1 : v i X r a 1 The emerging family of atomically thin materials is fueling the development of conceptually new technologies1 in highly-efficient optoelectronics2,3 and photonic applications,4 to name a few. The large variety of band gap values found in layered transition metal dichalcogenides (TMDC)5,6 make these materials especially suited for transistor applications. TMDCs are compounds with the general formula MX2, where M is a transition metal, e.g. Mo and W, and X is an element of the chalcogen group, S, Se and Te. They appear in a layered structure where the metal forms a hexagonal plane and the chalcogenides are positioned over and under this plane in either a trigonal prismatic (2H), as shown in Fig. 1(a), or octahedral (1T) stacking configuration.7 In the semiconducting 2H systems, the compounds show a transition from indirect band gap in bulk materials to direct band gap in single layers.8 Single and few-layers TMDCs have been implemented in a wide range of applications, ranging from thin film transistors,9 digital electronics and opto-electronics,2,10,11 flexible electronics,12 and up to energy conversion and storage devices.13 However, the defect states in TMDCs have an ambiva- lent nature and can have a major positive or negative impact on the performance of atomically-thin devices. The presence of defects in photodetectors can be beneficial since it has been shown to immo- bilize charges at the channel which improves the gain in photodetectors14 and produces non-volatile memory mechanisms.15 On the other hand, large hysteresis caused, for example, by charge traps2 and significant Schottky barriers16 at the metal-semiconductor interface are still a major design chal- lenge for the realisation of novel device architectures. They have been shown to cause degradation in the performance of transistors17 and generate high levels of flicker noise.18,19 To overcome these challenges, hysteresis is usually avoided by encapsulation20,21 or operation under high-vacuum.22,23 Most of the current research into surface states of TMDCs has focused on the chemical origins of charge trapping. A full understanding of their effect on the electrical properties is still lacking, hindering the optimization of functional components. While hysteresis has been shown to correlate with traps generated at the channel-dielectric interface and the channel-ambient interface,24,25 little 2 attention has been given to the mechanisms by which immobile charges affect the conduction charac- teristics of the devices, which is fundamentally different from those experienced in bulk devices. In this communication we present the first study of the role of immobile charges on the electrical transport properties of atomically thin MoTe2. This TMDC is of particular interest since its direct band gap of 1eV26,27 matches the wavelength of maximal solar emission intensity, thus making it a prime candidate for solar energy converters. MoTe2 is intrinsically p-doped, but can exhibit ambipolar behavior,26,28 mobility in the range of 10 -- 30 cm2 V−1s−1,26,29 and on-off ratios of up to 106.29 A stringent quantitative analysis demonstrates that the role of trapped charges in the operation of MoTe2-based electronic components is a change in the threshold voltage of the field-effect transistor (FET), effectively modulating the resistivity of the entire channel. By repeating the charge capture and emission cycles in different drain biases we are able to distinguish between two sources of transient behavior in MoTe2-FETs. One transient is due to emission of charges from traps to the channel, and the other is due to time-dependent capacitive gating of the channel that produces a transient in the effective threshold voltage. Finally, we present a complete analytical model to support our observations. Our findings are applicable to the entire class of atomically thin-based devices and provide a thorough understanding of charge traps and carrier dynamics which is needed to facilitate the intelligent design of fast and highly efficient opto-electronic devices. Few-layers MoTe2 flakes were obtained by mechanical exfoliation of 2H-MoTe2 bulk crystal (HQ graphene) onto highly doped silicon substrates, covered with 290 nm of high-quality thermally grown SiO2. The silicon substrate was used as a global back gate electrode, with the oxide layer acting as a gate dielectric. Standard electron beam lithography procedure was used to pattern electrodes and electrical leads. The contacts were then immediately metalized with 5 nm of Ti adhesion layer, and 50 nm of Au, using an electron beam evaporation system, working at very low pressure (∼ 10−8 mbar) and at long working distance, to achieve high uniformity in the deposition. The devices were then annealed in dry Ar/H2 environment at ambient pressure for 2 hours at 200◦C. Fig. 1(b) shows a 3 Figure 1: Panel (a) shows a three dimensional model of the 2H-MoTe2 crystal structure, with a single layer of the trigonal prismatic stack. Panel (b) shows the schematics of the device architecture and measurement setup. Panel (c) is atomic force microscope image of a typical device, showing the source and drain symmetric electrodes and the MoTe2 flake (outlined in dashed white line). The inset shows a scan profile (taken along the yellow line) from the substrate to the flake. schematic representation, not to scale, of the device and the circuit details. Atomic force microscopy measurements (Fig. 1(c)), and optical contrast (not shown here) of the flakes confirm that the surface of MoTe2 is not visibly contaminated and that the studied flakes consists of 4 number of layers. Low noise electrical measurements were performed in a home-built Farady cage in the dark and in ambient conditions on more than five different devices, all showing a similar behavior. The drain electrode was biased using a low noise voltage source and the source electrode was kept grounded throughout the experiment. The current flowing through the source electrode was measured using a current preamplifier. An independent voltage source-meter was used to apply a bias to the gate elec- trode while measuring the leakage current. The response time of the system was found to be limited only by the minimal rise time of the preamplifier, which is < 5µs. (See Supporting Information) The electronic behavior of multiple devices was characterized by measuring their drain current vs. 4 voltage response (Ids-Vds) and drain current vs. gate voltage transfer (Ids-Vgs) characteristics. Figure 2(a) shows the response curve of a typical MoTe2 transistor. The curve exhibits a slight asymmetry with higher resistivity for negative applied drain bias, indicating that the metal-semiconductor con- tacts form a small Schottky barrier for holes. The origin of this asymmetry about Vds = 0V is in the different electrostatic potential seen by the source and drain electrodes. In the experiment the poten- tial barrier at the MoTe2/source electrode interface is kept constant, as it is pinned by the gate. On the other hand, the biased drain barrier decreases (increases) in height with positive (negative) drain bias.30 Despite the low Schottky barrier, both the linear and the log-scale of the response curve (inset in Fig 2(a)) show that the device is not rectifying and is, in fact, largely Ohmic in higher Vds values (see Supporting Information). The device transfer characteristics are shown in Fig. 2(b), taken at Vds = 1V. The curve matches the expected behavior of an enhancement-mode p-channel transistor, showing an increase in drain cur- rent as the gate bias grows more negative beyond the threshold voltage (Vth). From the transfer curve, we can estimate the device mobility, µp, and sub-threshold swing, SS. Using µp = L(dIds/dVgs)/(WCoxVds) in the linear regime of the curve, where L = 1µm and W = 3µm are the device length and width, re- spectively, and Cox = ε0εr/d = 115µF m−2 is the gate dielectric capacitance, with ε0 the vacuum permittivity and εr the oxide relative permittivity, we find that the mobility is between 0.12 on the forward sweep and 0.14 cm2V−1s−1 on the back sweep. From the sub-threshold part of the curve, we estimate a sub-threshold swing value of 4 V dec−1 using SS =(cid:0)d log10 Id/dVg (cid:1)−1. The low value of the mobility and the high value of the swing are indicative of the presence of mid-gap trap states.14. In line with these findings, the gate sweep measurements also show a hysteretic behavior resulting in a shift in Vth between the forward and backward sweeps, which changes the threshold voltage by about ∆Vth = −4V and the charge neutrality point by about −6V, see Fig. 2(b). To understand the physical origin of the observed changes in threshold voltage we use the well- 5 Figure 2: Panel (a) shows the response (Ids-Vds) curve of a typical field effect transistor, taken with zero gate bias (Vgs = 0). The inset shows the same curve in a semi-logarithmic scale. Panel (b) shows the transfer (Ids-Vgs) curve of the same device taken with 1 V source drain bias (Vds) shown in a linear (solid black) and semi-logarithmic (solid red) scale. The dashed red lines are a linear extrapolation of the linear part of the curve, showing a change of 4 V in threshold voltage. The dashed black lines indicate the change in the position of the charge neutrality point. The arrows indicate the back gate sweep direction. Panels (c) and (d) show schematic energy band diagrams for the emission (c) and capture process (d) when the channel is in the "off state" and "on state", respectively. EC, EV , EF, ET 1 and ET 2 are the conduction band minimum, the valance band maximum, the Fermi energy, the shallow midgap state and deep midgap state energy, respectively. 6 (a)(b)(c)(d)~6V(0)(0)ECET2EFEVET1(+)(+)ECET2EFEVET1"O(cid:31) State" (Vg>Vth)"On State" (Vg<Vth)~4Vknown equation that describes Vth in field-effect transistors: − QT Cox Vth = ΦMS − Qi Cox − ∆EF (1) where ΦMS is the difference between the metal and semiconductor workfunctions when all the ter- minals are grounded, Cox is the gate dielectric capacitance, Qi is the static charge density within the dielectric, QT is the trapped charge density at the interface between the dielectric and the conductive channel and ∆EF is the shift in the Fermi Energy, required to turn the transistor on. From Eq. 1 it is clear that the only parameter that can change during the back gate sweep is the population of midgap traps, QT , indicating that positive charges (holes) are immobilized during the sweep. The process of charge trapping is illustrated in the energy band diagrams of Fig. 2(c+d) using two "donor-type" mid-gap states. In the "off state", where the Fermi level is above the trap levels (ET 1 and ET 2) the traps are occupied by an electron and are neutral. In the "on state", the traps are void of electrons (occupied by a hole) and are positively charged. A priori, the observed hysteresis can be due to charge trapping in the metal-semiconductor in- terface, i.e. localized at the contacts region, or at the entire surface area of the channel, i.e. at the semiconductor-dielectric and -ambient interface. However, the changes in the transfer curve strongly suggest that most of the charge trapping occurs throughout the entire area of the conductive channel, rather than at the metal-semiconductor interface. The noticeable shift in the charge neutrality point with respect to the gate bias (minimal conductivity in the log-scale, red curve) in Fig. 2(b), is indica- tive of a change in effective doping of the channel due to the space charge region generated by the immobilized charges. In contrast, a change in the degree of Fermi-level pinning at the contacts would have manifested primarily in changes in the linear slope of the logarithmic curve (the sub-threshold slope) and by changes in the width of the neutrality point. Assuming that the trapped charges are distributed in the channel, an assumption that is further validated by the analysis of the threshold tran- sients, we can estimate that the difference in trapped charge density between the forward and back sweep is about 4.3× 1011cm−2, using ∆QT = ∆VCox. 7 To gain insight on the dynamics of the charge traps, their effect on the transfer currents and their role in producing hysteretic cycles, we have monitored the transport characteristics while pulsing the gate electrode from "open" (more negative) to a "close" (more positive) value. The drain current was recorded over long periods of time (60-90 minutes) while the gate was repeatedly pulsed between Vgs = −10V to open the channel and Vgs = 0V to close it (Top panel in Fig. 3(a)). As the pulse on gate drives the channel from a close to an open state, a sudden rise of the current in the channel is measured followed by a fast decay. When the gate is pulsed back to the closed state, the current drops down and then slowly begins to recover. The decay in current in the open state is due to the capturing of holes in mid-gap traps that shifts the threshold voltage to a more negative value (red arrows in Fig. 2(d)), effectively closing the channel. On the other hand, the recovery in the off state is due to the holes that are emitted from the traps (blue arrow in Fig. 2(c)) shifting Vth to a less negative value. While the capture process is spontaneous and fast, the emission mechanism is thermally activated and, therefore, significantly slower then the capture rates. The vast majority of models used to quantify the time-dependent behavior of charge emission from mid-gap traps are based on Schottky or asymmetric diode structures.31,32 These models accurately describe the currents, and the resulting transient changes in capacitance, that are associated solely with the emission of charges from traps back into the circuit. However, transient changes in threshold voltage should affect the measured current in a completely different way, which has not yet been studied though it plays a pivotal role for the development of fast opto-electronic applications. To elucidate the fundamental difference in transient behaviors, we must first describe the main aspects of the conventional semiconductor model for current transients. When the emission of charges from depletion regions takes place, the current has a constant (saturation) component, which is a function of the applied bias, and a transient component which is the emission current: I(t) = I0 + qNT A τ e−t/τ 8 (2) Figure 3: Panel (a) shows the gate pulse cycles. The "On" and "Off" segments are highlighted. On the top, the applied gate voltage during each segment. On the bottom Panel: The drain current during the capture (on red background) and emission (on white background) segments. (b) An emission segment, recorded at Vgs = 0V and Vds = 1V , averaged over four cycles. The black circles are the measured data and the red curve is the fit to a double-exponential rise equation. (c) Emission segments, recorded at Vgs = 0V and varying Vds values, from 0.2V to 1.0V in 0.2V intervals. The circles are the measured data and the solid curves are the double-exponential fits. (d) The pre-exponential coefficients for the short emission coefficients, A1 (Top Panel) and long emission coefficient A2. The red line represents the best linear fit. Inset: An equivalent circuit diagram of the transient threshold model proposed here. 9 I0 is the saturation current, q is the elementary charge and A is the surface area of the device contact. Within this model, the time dependence of the transient current is a function of the density of trapped charges (NT ) and the decay coefficient τ which is a function of the energetic position of the trap with respect to the valance band.(see supporting information). However, in atomically thin MoTe2 FET, the high sensitivity of the conducting channel to its surrounding media means that the charge carrier dynamics can lead to significant shifts in threshold voltage and charge neutrality point, effectively changing the resistance of the entire channel. Hence, an inclusive model in which the resistivity changes with time is needed. To this end, we use the well known expression that describes the linear regime of the transfer curve, where the current is determined by:33 (cid:0)Vth(t)− Vg (cid:1)Vd Id(t) = W µpCox L (3) Where W and L is the channel width and length, respectively, and µp is the hole mobility. Since in atomically-thin FETs, the only time-dependent component of the threshold voltage (Eq. 1) is the density of trapped charges we can write dVth(t)/dt = −(q/Cox) (d pT (t)/dt) where pT = QT /q is the density of occupied traps. To obtain a full description of the threshold voltage transient, Vth(t), we assume that the density of free carriers, p, directly correlates to the equilibrium density p0, by p = p0 − pT , i.e. that there is no net injection of charges through the contacts. We further use the well-known result of the Shockley-Reed-Hall derivation to write the time-dependent density of occupied traps as pT (t) = NT e−t/τ. The transient of the threshold voltage then becomes: Vth(t) = Vth,sat − qNT e−t/τ Cox (4) where all the time-independent quantities have been grouped in Vth,sat for convenience. With the expression for Vth(t) from Eq. 4, the expression for the transient current is readily obtained: Id(t) = Id,sat − qW µpNT Vd L e−t/τ . (5) The expression in Eq. 5 has one striking difference from the conventional expression for current transient (Eq. 2), it is linear with drain bias. Qualitatively, this is a simple manifestation of Ohm's 10 law: as the resistance of the conductive channel changes with time, the current responds linearly, proportional to the applied bias. In the emission segments of the gate-pulse experiment, we find that a significant increase in cur- rents occurs on a very short time scales, while a further, slower increase is easily discernible in longer time scales. This behaviour cannot be satisfied by a single exponential fit but is in excellent agree- ment with adouble exponential rise equation in the form I(t) = I0 + A1e−t/τ1 + A2e−t/τ2 (red line in Fig. 3(a)) suggesting that there are two types of traps25, a shallow trap and a deeper one, corre- sponding to emission coefficients τ1 ≈ 250s and τ2 ≈ 2,900s. Fig. 3(b) shows the recovery currents, measured by pulsing the gate between -10 V and 0 V at drain bias values ranging from 0.2 V to 1 V. The curves are then fitted with a double exponential rise curve, without any assumption on the form of the pre-exponential factors, A1 and A2, while maintaining the emission constants within reasonable boundaries. To distinguish between the different contributions to the transient current, the pre-exponential factors of the shallow and deep traps are plotted in Fig. 3(d) on the top and bottom panel, respectively. Within the measurement error, it is clear that the pre-exponential factor of the transient current that is due to emission from the shallow traps is constant, and independent of the drain bias. This suggests that the measured signal is, indeed, the emission current from the traps. For the deep traps, the pre- exponential factors are found to have a linear dependence on Vds. This is expected for deep traps that are uniformly distributed about the conductive channel and are not simply concentrated at the metal-semiconductor interface, and is consistent with the analysis of the hysteresis of the gate bias measurements. Comparing the two panels in Fig. 3(d) reveals two striking features in the transient mechanism. First, the two orders of magnitude difference in the pre-exponential coefficients show that the threshold transient is the significant factor, governing the transistor response over time. Second, the change in the trap population (∆NT = L(dA2/dVd)/(qW µp) ∼ 109 cm−2) is a small fraction of the overall estimated density of 1012 states per cm2,34 corresponding to the small dynamic window 11 of operation used here. This emphasizes the significant role that the threshold voltage transients play in the behavior of atomically thin MoTe2 transistors. The presented model of the threshold voltage transients is general, since it does not take into account features which are specific to MoTe2. For example, similar studies conducted on WS2 grown by chemical vapor deposition also show a bi-exponential decays of the transient current which is fully captured by our model (see Supporting Information).Most importantly, this model is independent of the spatial location of trapped charge states (e.g. semiconductor-substrate or -ambient interface) and it is universally valid for semiconductor channels thickness that are significantly smaller than the Debye screening length, a condition easily met in emerging atomically thin materials. Our proposed model of threshold voltage transients can be further expanded and included in well established methodology of charge trap spectroscopy, whether probed by temperature scans35 or by optical means.36 However, the added simplicity of our methodology means that it can be applied to a variety of materials and substrates, including those that are photo-active, or temperature sensitive. Finally, we calculate the overall resistance of the device and find that the transient resistance operates in parallel to the saturation resistance: (cid:18)dId,sat(t) dVd (cid:19) − qW µpNT L e−t/τ (6) (cid:18)dId(t) (cid:19) dVd = or R−1 = R−1 sat + R−1 trans which is a strong indication to the fact that both factors indeed stem from the channel itself. We note that the addition of series resistance to the circuit, such as contact resis- tance, does not affect the time-dependent characteristics of the model, as is discussed in details in the Supporting Information. In conclusion, we have demonstrated a new approach to the analysis of charge trapping and tran- sient response of TMDC-based FETs, which paves the way to a better understanding of the role of mid-gap state in the operation novel devices. Using a simple two terminal model system, we were able to distinguish between currents associated with the emission of trapped charges into the circuit 12 and currents that evolve in time due to the changes in effective threshold voltage across the channel. The mechanism of threshold voltage transients which we study and model is not limited to MoTe2 but it is valid to any device based on atomically thin materials. Indeed, as long as the channel depth is much smaller than the Debye screening length, the threshold voltage will be strongly modulated by the formation of space charge regions at both the semiconductor-dielectric and -ambient interfaces. Our model, which describes the basic physics that govern the hysteretic characteristics of atomically thin FETs, is instrumental for the design of defect-based devices, such as photodetectors and memory devices, as well as provides a new methodology to study the nature of these defects. Acknowledgments Iddo Amit acknowledges financial support from The European Commission Marie Curie In- dividual Fellowships (Grant number 701704). S. Russo and M.F. Craciun acknowledge financial support from EPSRC (Grant no. EP/J000396/1, EP/K017160/1, EP/K010050/1, EP/G036101/1, EP/M001024/1, EP/M002438/1), from Royal Society international Exchanges Scheme 2016/R1 and from The Leverhulme trust (grant title "Quantum Drums" and "Room temperature quantum electron- ics"). N.J. Townsend and S. Russo acknowledge DSTL grant scheme Sensing and Navigation using quantum 2.0 technologies. C.M. acknowledges the award of a Royal Society University Research Fellowship by the UK Royal Society, and the EPSRC-Royal Society Fellowship Engagement Grant EP/L003481/1. 13 References [1] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C.-Y. Chim, G. Galli, F. Wang, Nano letters 2010, 10, 1271. [2] Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman, M. S. Strano, Nature nanotechnology 2012, 7, 699. [3] H. Chen, H. Liu, Z. Zhang, K. Hu, X. Fang, Advanced Materials 2016, 28, 403. [4] F. Xia, H. Wang, D. Xiao, M. Dubey, A. Ramasubramaniam, Nature Photonics 2014, 8, 899. [5] Q. Tang, Z. Zhou, Progress in Materials Science 2013, 58, 1244. [6] G. Fiori, F. Bonaccorso, G. Iannaccone, T. Palacios, D. Neumaier, A. Seabaugh, S. K. Banerjee, L. Colombo, Nature nanotechnology 2014, 9, 768. [7] R. Lv, J. A. Robinson, R. E. Schaak, D. Sun, Y. Sun, T. E. Mallouk, M. Terrones, Accounts of chemical research 2014, 48, 56. [8] K. F. Mak, C. Lee, J. Hone, J. Shan, T. F. Heinz, Physical Review Letters 2010, 105, 136805. [9] S. Das, R. Gulotty, A. V. Sumant, A. Roelofs, Nano letters 2014, 14, 2861. [10] D. Jariwala, V. K. Sangwan, L. J. Lauhon, T. J. Marks, M. C. Hersam, ACS nano 2014, 8, 1102. [11] S. Das, M. Dubey, A. Roelofs, Applied Physics Letters 2014, 105, 083511. [12] D. Akinwande, N. Petrone, J. Hone, Nature communications 2014, 5. [13] H. Wang, H. Feng, J. Li, Small 2014, 10, 2165. [14] M. M. Furchi, D. K. Polyushkin, A. Pospischil, T. Mueller, Nano letters 2014, 14, 6165. [15] H. S. Lee, S.-W. Min, M. K. Park, Y. T. Lee, P. J. Jeon, J. H. Kim, S. Ryu, S. Im, Small 2012, 8, 3111. 14 [16] A. Allain, J. Kang, K. Banerjee, A. Kis, Nature Materials 2015, 14, 1195. [17] J. Pu, K. Funahashi, C.-H. Chen, M.-Y. Li, L.-J. Li, T. Takenobu, Advanced Materials 2016, 28, 4111. [18] I.-T. Cho, J. I. Kim, Y. Hong, J. Roh, H. Shin, G. W. Baek, C. Lee, B. H. Hong, S. H. Jin, J.-H. Lee, Applied Physics Letters 2015, 106, 023504. [19] I.-T. Cho, W.-M. Kang, J. Roh, C. Lee, S. H. Jin, J.-H. Lee, in 2015 IEEE 22nd International Symposium on the Physical and Failure Analysis of Integrated Circuits, IEEE, 2015 pp. 480 -- 483. [20] G.-H. Lee, Y.-J. Yu, X. Cui, N. Petrone, C.-H. Lee, M. S. Choi, D.-Y. Lee, C. Lee, W. J. Yoo, K. Watanabe, et al., ACS nano 2013, 7, 7931. [21] D. Kufer, G. Konstantatos, Nano letters 2015, 15, 7307. [22] D. Jariwala, V. K. Sangwan, D. J. Late, J. E. Johns, V. P. Dravid, T. J. Marks, L. J. Lauhon, M. C. Hersam, Applied Physics Letters 2013, 102, 173107. [23] V. K. Sangwan, H. N. Arnold, D. Jariwala, T. J. Marks, L. J. Lauhon, M. C. Hersam, Nano letters 2013, 13, 4351. [24] Y. Park, H. W. Baac, J. Heo, G. Yoo, Applied Physics Letters 2016, 108, 083102. [25] D. J. Late, B. Liu, H. R. Matte, V. P. Dravid, C. Rao, Acs Nano 2012, 6, 5635. [26] I. G. Lezama, A. Ubaldini, M. Longobardi, E. Giannini, C. Renner, A. B. Kuzmenko, A. F. Morpurgo, 2D Materials 2014, 1, 021002. [27] I. G. Lezama, A. Arora, A. Ubaldini, C. Barreteau, E. Giannini, M. Potemski, A. F. Morpurgo, Nano letters 2015, 15, 2336. 15 [28] H. Xu, S. Fathipour, E. W. Kinder, A. C. Seabaugh, S. K. Fullerton-Shirey, ACS nano 2015, 9, 4900. [29] N. R. Pradhan, D. Rhodes, S. Feng, Y. Xin, S. Memaran, B.-H. Moon, H. Terrones, M. Terrones, L. Balicas, ACS nano 2014, 8, 5911. [30] H. Tian, Z. Tan, C. Wu, X. Wang, M. A. Mohammad, D. Xie, Y. Yang, J. Wang, L.-J. Li, J. Xu, et al., Scientific reports 2014, 4. [31] D. Lang, Journal of Applied Physics 1974, 45, 3023. [32] J. Borsuk, R. Swanson, IEEE Transactions on Electron Devices 1980, 27, 2217. [33] S. M. Sze, K. K. Ng, Physics of semiconductor devices, John wiley & sons, 2006. [34] Y.-F. Lin, Y. Xu, C.-Y. Lin, Y.-W. Suen, M. Yamamoto, S. Nakaharai, K. Ueno, K. Tsukagoshi, Advanced Materials 2015, 27, 6612. [35] C. Chen, K. Abe, H. Kumomi, J. Kanicki, IEEE Transactions on Electron Devices 2009, 56, 1177. [36] K. Lee, M. S. Oh, S.-j. Mun, K. H. Lee, T. W. Ha, J. H. Kim, S.-H. K. Park, C.-S. Hwang, B. H. Lee, M. M. Sung, et al., Advanced Materials 2010, 22, 3260. 16 Supporting Information: Transient Threshold Voltage in Atomically Thin MoTe2 Transistors Iddo Amit1, Tobias J. Octon2, Nicola J. Townsend1, Francesco Reale3, C. David Wright2, Cecilia Mattevi3, Monica F. Craciun2 and Saverio Russo1† 1Centre for Graphene Science, Department of Physics, University of Exeter, Stocker Road, Exeter, 2Centre for Graphene Science, Department of Engineering, University of Exeter, North Park Road, United Kingdom, EX4 4QL Exeter, United Kingdom, EX4 4QF 3Department of Materials, Imperial College London, United Kingdom, SW7 2AZ †[email protected] Contents Additional Results 18 MoTe2 transfer and transient curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Threshold voltage transients in WS2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Response Time of the setup Detailed Derivation of the Threshold Transient Model 21 22 17 Additional Results MoTe2 transfer and transient curves The transfer curves, Ids-Vgs of two additional devices are shown in Fig. S1(a) and (b). The curves show that different devices, with different resistance and quantitative gate responses show a similar qualitative behavior. For the device shown in Fig. S1(a), the mobility is 0.04 cm2 V−1s−1 and the sub-threshold swing is ∼ 13 V dec−1, whereas the device in Fig. S1(b) has a mobility of 0.15 cm2 V−1s−1 and subthreshold swing value similar to the former device. Strikingly, given the large variation in quantitative properties, the trend in behavior, both in the transfer curves and the transient curves (Fig. S2) remains similar in all devices, which is a strong indication to the validity of the proposed model. Figure S1: Two additional transfer (Ids-Vgs) curves of MoTe2 devices. The black curves are shown on a linear scale and the red curves on a semi-logarithmic scale. Arrows indicate the sweep direction The transient response to gate pulses was measured on several devices, with different pulse heights. Here we show an analysis done on three additional devices. The results shown in Fig. S1(a) and (d) were collected by pulsing the gate between Vgs = −10 V ("open") and Vgs = −5 V 18 -40-200204002040Drain Current (nA)Gate Bias (V)10-1310-1210-1110-1010-910-8Drain Current (A)-40-2002040012345Drain Current (nA)Gate Bias (V)10-1210-1110-1010-910-8Drain Current (A)(a)(b)("close"). The results shown for the next two devices were collected using the same dynamic window of ∆Vgs = 10 V, as the device discussed in the main text, and the gate was pulsed between Vgs = −20 V ("open") and Vgs = −10 V ("close"). The transfer properties of all the MoTe2 transistors are qualitatively comparable. The transient curves shown in Fig. S1 (a) have a slower rise in the lower values of Vds. Indeed, the pre-exponential decay factor of the deep traps, A2 are constant up to Vds = 0.4 V and sub-linear up until Vds = 0.6 V. This result is attributed to the high conductivity of the channel at Vgs = −5 V and the low drift velocity across the channel at Vds ≤ 0.6 V. In this combination of parameters the magnitude of the emission current is comparable or larger than the drift current in the channel and, therefore, is the dominant current in the system. However, as the drift velocity increases with increased Vds, the threshold voltage transient becomes more dominant, as is evident from the linear dependence of the pre-exponential decay factor in Vds. In contrast, the transient curves shown in Fig. S1 (b) and (c) are of devices with higher resistivity than that of the device discussed in the main text, as is evident from the lower currents. While in these devices, the linear trend (Fig. S1 (e) and (f)) is visible from Vds = 0.2 V onwards, the shape of the transient curves in Fig S1 (b) is quite different from those of the other reported devices. From the analysis, we find that the emission currents from the shallow traps happen at time constants, τ1 ≈ 0.4 s, much smaller than the other reported devices, where τ1 is in the order of a few tens of seconds. We can, therefore, conclude that in this device, the contribution of the emission currents from the shallow traps were negligible, eliminating the initial (fast) recovery, and thus changing the shape of the transient curves. 19 Figure S2: (a) -- (c) Current recovery segments, for three different devices, recorded at Vgs = −5 V (panel (a)) and Vgs = −10 V (panels (b) and (c)) and varying Vds values, from 0.2V to 1.0V in 0.2V intervals. The circles are the measured data and the solid curves are the double-exponential fits. (d) -- (f) The pre-exponential coefficients for the short emission coefficients, A1 (Top Panel) and long emission coefficient A2. Fig S1(d) includes results from Vds values that are not shown in (a). The red line represents the best linear fit. Threshold voltage transients in WS2 The mechanism of threshold voltage transient changes to the conduction of the channel is not limited to devices with high density of trap states. To illustrate the generality of the model, we present some of the threshold voltage transients measured in a CVD grown bilayer WS2 transistor, which was in-situ annealed and measured in vacuum (< 2× 10−7 mbar) The transfer curve in Fig S3 (a) shows that in the WS2 transistors, the hysteresis is significantly lower than that found in the MoTe2. The transient curves and emission coefficients measurements for 20 Figure S3: (a) Transfer curves (Ids-Vgs) of a CVD-grown WS2 transistor in the linear (black) and semi-logarithmic (red) scales. (b) Current recovery segments recorded at Vgs = 10 V and varying Vds values, from 0.2V to 1.0V in 0.2V intervals. The circles are the measured data and the solid curves are the double-exponential fits. (c) The pre-exponential coefficients for the short emission coefficients, A1 (Top Panel) and long emission coefficient A2. The solid lines represent the best linear fit. WS2 clearly show that the same mechanism of time dependent resistance governs the behavior of the charge conduction in the channel, see Fig. S3 (b) and (c), respectively. Even though the charge carrier mobility in WS2 is more than 100 times larger than that in the studied MoTe2, we still find that both the short lived and the long lived traps contribute to the modulation of channel resistance. Response Time of the setup To ensure that the measured results are not an artefact of the measurement, the response time of the set-up was measured by pulsing the gate source unit and recording the current through a resistor of comparable resistance to that of the channel (∼ 1 GΩ). The rise time of the current was found to be less than 2 µs, corresponding to the rise time of the current preamplifier. The current follows an exponential rise equation with a rise coefficient τ = 0.22 µs, thus ensuring that the time response measured o the atomically thin MoTe2 is solely due to the carrier dynamics in the device. Fig. S3 21 Figure S4: The rise time of the set-up, recorded using a 1 GΩ resistor shows the measured rise time of the current. Detailed Derivation of the Threshold Transient Model Classical Current Transient Theory To allow for a thorough discussion of the difference between the classical model of transient currents and the newly presented model which accurately describes the carrier dynamics in atomically thin MoTe2-FETs, we must first present the main aspects of the classical current transients theory. We present an analysis for the process of capturing holes near the valance band maximum (EV ), as this is the relevant process for p-doped MoTe2. The rate of capturing holes from the valance band (Rhc) is proportional to the density of holes in the valance band (p) and the density of unoccupied traps (NT − pT ), where NT is the total density of trapping states and pT is the density of occupied states. It's important to note here that "unoccupied" 22 0.00.40.81.21.62.0-6.0-5.8-5.6-5.4-5.2-5.0Voltage (V)Time (µs)y=y0+A exp(-t/τ)τ = 0.22 µsRise Time < 2µsfrom holes means occupied by electrons and electrically neutral. = −cp(NT − pT )p (7) (cid:12)(cid:12)(cid:12)(cid:12)capture Rhc ≡ ∂ p ∂t Where cp is the capture coefficient for holes, and it equals the thermal velocity, vth, multiplied by the capture cross section, σp. The emission of holes from the traps is described using same considerations without taking into account the unoccupied states in the valance band, since it is assumed that for a non-degenerate semiconductor the emission rate is not limited by it. (cid:12)(cid:12)(cid:12)(cid:12)emission Rhe ≡ ∂ p ∂t = eppT (8) Where ep is the emission rate of holes from traps to the valance band. It is therefore clear, that the total change in trap occupation is given by: (cid:12)(cid:12)(cid:12)(cid:12)capture Rp = ∂ p ∂t (cid:12)(cid:12)(cid:12)(cid:12)emission = eppT − cp(NT − pT )p + ∂ p ∂t With the traps saturated we can write NT = pT and the capture rate will become zero. Rp = ∂ p ∂t = eppT (9) (10) In a simple process where every hole added to the valance band is removed from a trap (i.e. without any further charge injection), it is clear that: Combining Eq. 11 with Eq. 10 yields ∂ p ∂t + ∂ pT ∂t = 0 pT = pT (0)e−t/τ (11) (12) Where τ = 1/ep is the decay constant per trap, and pT (0) = NT is the trap occupation at the saturation point. 23 In the classical case, where the entire contribution to the current transient is from charges that are emitted from the traps back into the circuit, the current transient is given by I(t) = I0 + qRpA, Where I0 is the steady state current, and A is the area from which charges are emitted. Using Eq. 12, one can write an explicit expression for the current transient I(t) = I0 + qNT A τ e−t/τ (13) Derivation of the Threshold Voltage Transient For time-dependent currents that stem from evo- lution of the threshold voltage in the field-effect transistor (FET), there are a few parameters that determine the current transient. First, the current equation for an FET in the linear regime is (cid:0)Vth(t)− Vg (cid:1)Vd Id(t) = W µpCox L (14) Where Id is the drain current, W and L are the channel width and length, respectively, µp is the mo- bility of the holes, Cox is the capacitance of the gate dielectric, and Vth and Vg are the FET threshold voltage for conduction and the gate bias, respectively. The threshold voltage is given by Vth(t) = ΦMS − QT (t) Cox − ∆EF (15) Where ΦMS is the difference in workfunction between the gate electrode and the conduction channel and QT (t) = (Q0 + qpT (t)) accounts for both the stationary charges in the oxide (Q0) and the dynamic charges that are trapped and de-trapped on the channel. It is important to note here that in contrast to a conventional inversion-based FET, the MoTe2 is an accumulation-based transistor. Therefore, the "textbook" 2φF expression for strong inversion has been substituted here for a general ∆EF which represent the change in Fermi energy required to "open" the channel. From this equation, one can easily write an expression to describe the dynamics of the threshold voltage Using a simple model for the concentration of free charge carriers p(t) =(cid:0)Cox(Vth(t)− Vg)(cid:1) /q, it's dt (16) dVth(t) dt = − q Cox d pT (t) easy to see that charge is conserved in this model, p = p0 − pT , where p0 is the total density of 24 holes in the valance band in equilibrium conditions and without traps, and is constant. Using the previously found expression for the emission rate, we can now write an expression for the time- dependent threshold voltage Vth = ΦMS − Q0 Cox − qNT e−t/τ Cox − ∆EF = Vth,sat − qNT e−t/τ Cox (17) Where on the right hand side, all the terms that are time-independent were grouped together into Vth,sat The current then becomes (cid:32) Vth,sat − qNT e−t/τ Cox (cid:33) − Vg Vd = Id,sat − qW µpNT Vd L e−t/τ (18) Id(t) = W µpCox L Finally, to account for the case where the a resistance in series (e.g., contact resistance) plays an important role in the device performance, we add a constant resistance term, RS to Eq. 14: From this equation, we can easily isolate the current term: (cid:1) (Vd − IdRS) (cid:0)Vth(t)− Vg (cid:1) (cid:0)Vth(t)− Vg (cid:1)RS (cid:0)Vth(t)− Vg Vd Id(t) = W µpCox L Id(t) = 1 + W µpCox W µpCox L L (19) (20) S Which is still linear with Vd, in accordance with Ohm's law. The importance of this result is clear when we examine the limits where the contact resistance is much larger than the channel resistance, . In this limit, the current simply reduces to Id = VdR−1 i.e., when RS (cid:29)(cid:16)W µpCox which is time-independent. On the other limit, RS (cid:28)(cid:16)W µpCox (cid:0)Vth(t)− Vg (cid:0)Vth(t)− Vg , Eq. 20 simply (cid:1)(cid:17)−1 (cid:1)(cid:17)−1 L L reduces back to Eq. 14. The dominant time-dependent characteristics of the emission currents are therefore a strong indication that the major contribution to the transient profile stems from the time- dependent changes in the channel resistance. 25 

1612.07410

2

1612

Fractional spin and Josephson effect in time-reversal-invariant topological superconductors

Time reversal invariant topological superconducting (TRITOPS) wires are known to host a fractional spin hbar/4 at their ends. We investigate how this fractional spin affects the Josephson current in a TRITOPS-quantum dot-TRITOPS Josephson junction, describing the wire in a model which can be tuned between a topological and a nontopological phase. We compute the equilibrium Josephson current of the full model by continuous-time Monte Carlo simulations and interpret the results within an effective low-energy theory. We show that in the topological phase, the 0-to-pi transition is quenched via formation of a spin singlet from the quantum dot spin and the fractional spins associated with the two adjacent topological superconductors.

cond-mat.mes-hall

cond-mat

Fractional spin and Josephson effect in time-reversal-invariant topological superconductors Alberto Camjayi,1 Liliana Arrachea,2 Armando Aligia,3 and Felix von Oppen4 1Departamento de F´ısica, FCEyN, Universidad de Buenos Aires and IFIBA, Pabell´on I, Ciudad Universitaria, 1428 CABA Argentina 2International Center for Advanced Studies, ECyT-UNSAM, Campus Miguelete, 25 de Mayo y Francia, 1650 Buenos Aires, Argentina 3Centro At´omico Bariloche and Instituto Balseiro, CNEA, 8400 S. C. de Bariloche, Argentina 4Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, 14195 Berlin, Germany Time reversal invariant topological superconducting (TRITOPS) wires are known to host a frac- tional spin /4 at their ends. We investigate how this fractional spin affects the Josephson current in a TRITOPS-quantum dot-TRITOPS Josephson junction, describing the wire in a model which can be tuned between a topological and a nontopological phase. We compute the equilibrium Joseph- son current of the full model by continuous-time Monte Carlo simulations and interpret the results within an effective low-energy theory. We show that in the topological phase, the 0-to-π transition is quenched via formation of a spin singlet from the quantum dot spin and the fractional spins associated with the two adjacent topological superconductors. PACS numbers: 74.78.Na, 74.45.+c, 73.21.La Introduction.-The interplay of many-body interac- tions in quantum dots and superconductivity has been at the focus of interest for some time [1–6]. While elec- trons are paired in superconductors, the charging energy effectively suppresses pairing in quantum dots. A promi- nent consequence of this competition is the transition be- tween 0 and π junction behavior of the Josephson current in devices where a quantum dot (QD) connects between ordinary (nontopological) singlet-superconducting wires (S-QD-S junction) [7–10]. As a result of numerous stud- ies [11–19], this phenomenon is now well understood for conventional superconductors. Essentially, S-QD-S junc- tions exhibit π-junction behavior when the QD hosts an effective spin-1/2 degree of freedom. Here, we address the 0 to π transition for Joseph- son junctions in which a quantum dot connects be- tween time-reversal-invariant topological superconduc- tors (TRITOPS). Unlike their time-reversal-breaking cousins [2, 20–22], TRITOPS preserve time-reversal sym- metry and can coexist with an unpolarized quantum- dot spin. It is thus an interesting question whether π-junction behavior can be observed in TRITOPS-QD- TRITOPS junctions. Such junctions differ from con- ventional S-QD-S junctions in several ways. First, the Majorana-Kramers pairs present in the topological phase allow for the coherent transfer of single electrons, while the Josephson current of a conventional junction is car- ried by Cooper pairs. Even more intriguing, TRI- TOPS host a fractional /4 spin at their ends. Thus, a TRITOPS-QD-TRITOPS junction allows one to study the hybridization of fractional and ordinary spins. We show that the 0-π transition constitutes a signature which distinguishes between the topological and the nontopo- logical phase, and trace the quenching of the transition for TRITOPS to the formation of a spin singlet from the quantum-dot spin and the fractional spins of the adjacent TRITOPS. In the wake of proposals to engineer time-reversal- breaking topological phases and corresponding experi- ments, there has also been substantial interest in time reversal invariant topological superconductors (TRIP- TOPS) [1, 24–26, 28–37]. TRITOPS are characterized by Kramers pairs of Majorana end states and localized fractional spins [28]. Time reversal protects the pair of Majoranas from hybridizing which therefore generically remain at zero energy. Similarly, the fractional spin is topologically protected and cannot be determined from a local measurement without breaking time reversal. Sev- eral routes have been proposed to engineer TRITOPS al- though their experimental realization is more demanding than that of time-reversal-breaking topological supercon- ductors [36]. Conventional Josephson junctions assume their mini- mal energy at zero phase difference and their maximal energy at a phase difference of π (0-junction behavior). This behavior is reversed in π junctions which assume their minimal energy at a phase difference of π [1, 2]. In S-QD-S junctions, π-junction behavior occurs when the quantum dot forming the junction is singly occupied and acts effectively as a magnetic impurity. When the QD is weakly coupled to the superconductors, tunneling of Cooper pairs between the conventional superconductors relies on a forth-order cotunneling process [1, 10]. This process includes a π phase shift which originates from the Fermi statistics of electrons and becomes manifest in the π-junction behavior. As a consequence, the current- phase relation of the junction phase shifts by π when the occupation of the quantum dot is tuned from even to odd. When the quantum dot is strongly coupled to the super- conductors, the impurity spin can be screened, turning arXiv:1612.07410v2 [cond-mat.mes-hall] 31 Jul 2017 a doublet into a singlet ground state and resulting in 0- junction behavior. Depending on the parameter regime, this transition can be described as a result of Kondo cor- relations or a zero-energy crossing of a Yu-Shiba-Rusinov state [19]. Model.-Our considerations are based on a time- reversal-invariant superconductor with Hamiltonian [1] 2 Hα = N j=1Xσ (cid:16)−tc†α,j+1,σcα,j,σ + iλσc†α,j+1,σcα,j,σ X +∆σeiφα c†α,j+1,σc†α,j,σ + H.c. − µ nα,j,σ(cid:17) , (1) where λ↑,↓ = ±λ, ∆↑,↓ = ±∆ and ↑ =↓, ↓ =↑. More- over, t is the hopping parameter, µ the chemical poten- tial, and λ and ∆ are the strengths of Rashba spin-orbit coupling and extended s-wave pairing, respectively. The index α = L, R labels the left and right superconductors of the junction with order parameter phases φα. The phase difference φ = φL − φR = 2πΦ/Φ0 can be tuned by including the junction in a superconducting loop and threading the loop by a magnetic flux Φ. (Φ0 = h/2e denotes the superconducting flux quantum.) The entire TRITOPS-QD-TRITOPS Josephson junction is then de- scribed by the Hamiltonian Here, H = Xα=L,R Hd = εdXσ Hα + Hc + Hd. (2) nd,σ + U nd↑nd↓ (3) describes the quantum dot with gate-tunable level energy εd, spin-resolved level occupation nd,σ, and charging en- ergy U , and Hc = −t0Xσ h(cid:16)c†L,N,σ + c†R,1,σ(cid:17) dσ + H.c.i (4) accounts for the hybridization between quantum dot and superconductors. The Hamiltonian Hα supports topological and non- topological phases. The topological phase occurs when µ < 2λ and is characterized by Kramers pairs of Ma- jorana end states. For each lead, the corresponding Majorana operators can be combined into conventional fermionic operators ΓL/R =Z dxϕL/R(x)[ψ↑(x) ∓ iψ† ↓ (x)], (5) where ϕL/R(x) denotes the Majorana wavefunctions of the left (L) and right (R) superconducting lead and ψσ(x) denotes the electron field operator for spin σ (see Supple- mentary Material, Sec. 2 [38]). While the Majorana oper- ators mix the two spin components, the operators ΓL/R FIG. 1. (Color online) Top: Josephson current vs φ for a quantum dot at T = 0 with U = 0, t0 = t, λ = t/2, ∆ = t/5, and values of µ in the topological (µ < t) as well as the nontopological (µ > t) phase. The wires have N = 500 sites. Inset: Josephson current at finite temperature. Red, blue, and green lines correspond to β = 400, 200 and 100, respectively. The T = 0 case is plotted in black for reference. Bottom: Spectrum of HBdG for µ = εd = 0 (left), µ = 0, εd = t (middle), and µ = −εd = 1.8 (right). Other parameters as in top panel. Energies are measured in units of t = 1. remove a spin of /2 from one end of the wire. Thus, ΓL/R and Γ†L/R toggle the the system between ground states with fractional spins of ±/4 localized at the ends of the wire [28]. Numerical results.-The Josephson current can be computed from the Green function expression I = 2t02 β Xσ Xn Imhg(12) 1α,σ(iωn)G(21) (6) d,σ (iωn)i . The derivation is included in Ref. [38] (see Sec. 1). The Green functions correspond to the Matsubara compo- nents of frequency ωn = (2n + 1)π/β (β is the inverse temperature) of the imaginary-time Green functions g(12) d,σ (τ ) = 1α,σ(τ ) = −hTτhc†α,1,σ(τ )c†α,1,σ(0)ii0 and G(21) −hTτh dσ(τ ) dσ(0)ii, where h. . .i0 (h. . .i) denotes the en- semble average over the states of Hα (H). The first Green function can be obtained exactly. First consider a junction with a noninteracting quan- tum dot. For U = 0, the Green function Gd,σ(iωn) and thus the Josephson current can also be evaluated analyti- β = 400 β = 200 β = 100 0.1 I 0.05 0 0 0.25 0.75 1 0.5 φ/π 0.1 0.05 I 0 -0.05 -0.1 µ = 0.0 µ = 0.4 µ = 0.8 µ = 1.0 µ = 1.4 µ = 1.8 0 0.5 1 φ/π 1.5 2 0 0.5 1 f/p 1.5 2 0 0.5 1 f/p 1.5 2 0 0.5 1 f/p 1.5 2 0.2 0.1 -0.1 -0.2 0 E0(f) 3 Shiba transformation, mapping H to a particle-number conserving Hamiltonian with negative U [15]. The Green function of the transformed problem is then calculated by the algorithm introduced in Refs. [42, 43]. Inversion of the Shiba transformation leads to Gd,σ(iωn) which en- ters the Josephson current (8). Results for a half-filled configuration (i.e., hnd↑ + nd↓i = 1) are shown in Fig. 2. The nontopological case (bottom panel) shows the ex- pected 0 to π transition. When coupling the quantum dot to superconducting leads, the local moment per- sists when ∆ is larger than the Kondo temperature TK, but becomes Kondo screened by the quasiparticle states for ∆ (cid:28) TK. For a particle-hole symmetric configura- tion, the Kondo temperature of the junction is given by kBTK =pδU/2 exp (−πU/8δ) [3] with the hybridization parameter δ ∼ π(t0)2p1 − µ2/(2t)2/(2t). Consequently, there is a 0-π transition as U increases. For U = t, the dot is in the intermediate valence regime, while for U = 6t and U = 10t, it would be in the Kondo regime when attached to normal leads. In our case, kBTK ∼ ∆ for U ∼ 8.5t, consistent with the observed transition be- tween 0- and π-junction behavior between U = 6t and U = 10t. It is our central observation that there is no corre- sponding 0-π transition when the superconducting leads are in the topological regime. Instead, the current-phase relation remains similar to the noninteracting case for all interaction strengths U . In particular, the abrupt de- pendence at a phase difference of φ = π, while slightly smoothed by finite temperature, becomes more pro- nounced as the number of sites increases, as in the nonin- teracting case (cp. inset of Fig. 1). These results suggest that the impurity spin is efficiently screened in the topo- logical case, despite the presence of the superconducting gap. This robust screening of the spin of the quantum dot originates from its interaction with the subgap states emerging from the Kramers pairs of Majoranas of the adjacent left and right wires. Effective Hamiltonian.-To arrive at this conclusion, we interpret our numerical results in the context of an effective Hamiltonian. Consider a singly-occupied, interacting quantum dot coupled to two time-reversal- invariant topological superconductors. For simplicity, we assume that the superconducting gap is large compared to the Kondo temperature so that we can neglect hy- bridization with the quasiparticle continuum. Then, we only need to consider the hybridization with the sub- gap states originating from the Majorana bound states. We can project out the empty and doubly occupied dot states by employing a Schrieffer-Wolff transformation [4] (see also [37] for similar considerations). This yields an effective Hamiltonian in the eight-dimensional subspace spanned by the two eigenstates of the quantum dot spin Sd and the two states for each of the superconducting leads which are associated with the Kramers pair of Ma- jorana operators. Here, we sketch the derivation of this FIG. 2. (Color online) Josephson current for an interacting quantum dot at β = 400 with t0 = t, εd = −U/2, λ = t/2, ∆ = t/5. The upper (lower) panel corresponds to the topological (nontopological) phase. The values of U and µ are indicated in the Fig. Energies are expressed in units of t = 1 . cally. Moreover, our model can be written in Nambu rep- resentation with a Bogoliubov de-Gennes (BdG) Hamil- tonian HBdG = H0τz + ∆τx, where H0 results from the normal parts of the Hamiltonian H while ∆ originates from the pairing contributions. The Pauli matrices τi (with i = x, y, z) operate in particle-hole space. Diago- nalizing the BdG Hamiltonian, the Josephson current can be obtained from I = (2e/)∂E0(φ)/∂φ, where E0(φ) is the many-body ground state energy. Corresponding re- sults are presented in Fig. 4. The spectrum of HBdG is shown in the lower panels for the topological (left and middle) and the nontopo- logical (right) phase. In the topological phase, there are zero-energy bound states (light-blue curves) which emerge from the Majoranas states localized at the far ends of the finite-length chains. The solid red curves emerge from the hybridization of the dot states with the adjacent Majoranas. In the nontopological phase, the subgap states are gapped. As a consequence of Kramers theorem, the subgap states are twofold degenerate at φ equal to integer multiples of π. At other flux values, time reversal is broken by the phase bias and the subgap states are nondegenerate. The top panel of Fig. 4 shows the Josephson current for values of µ both in the topo- logical (µ < 2λ) and the nontopological (µ > 2λ) phases. In the topological phase, the Josephson current jumps at φ = π (up to finite-size effects), reflecting the level cross- ing of the subgap states. (Note that we assume complete equilibration over fermion parities.) The nontopological phase exhibits the usual smooth behavior. For a nonzero interaction U , the Josephson current can be calculated by evaluating G(21) d,σ (iωn) using quan- tum Monte Carlo simulations [39]. Previous works on S-QD-S junctions [12, 15, 16] as well as normal wires coupled to correlated dots and molecules [40, 41] proved this strategy to be accurate and reliable. We perform a µ = 0 µ = 1.8 0.1 I 0.05 0 0.02 0 I U = 0 U = 1 U = 6 U = 10 -0.02 0 0.25 0.5 φ/π 0.75 1 4 At all phase differences, the ground state is an equal- probability superposition of ↑, 0, 0i and ↓, 1, 1i. These states describe configurations with overall zero spin. In- deed, in both states the quantum dot spin of /2 is point- ing opposite to the fractional spins of /4 of the two ad- jacent superconductors. Thus, these configurations can be interpreted as an effective singlet configuration of the quantum dot spin and the fractional spins of the topo- logical superconductors. Similar to the singlet formation via hybridization with the quasiparticle continuum of nontopological supercon- ductors [19], this singlet formation with the fractional spins quenches the π-junction behavior. Indeed, the low-energy spectrum emerging from the Schrieffer-Wolff treatment predicts a Josephson energy which is mini- mal at phase differences equal to integer multiples of 2π. Moreover, we also see that there is a cusp in the ground state energy at a phase difference of π. Both of these results are consistent with our numerical results which incorporate the hybridization with the quasiparticle con- tinuum above the superconducting gap. quantum dot ρ(ω) = −2Pσ Im[GR In Fig. 3, we benchmark our low-energy Hamiltonian with results for the local density of states (LDOS) at the d,σ(ω)]. The latter was calculated by analytically continuing the Monte Carlo data to the real frequency axis. Results are shown in the color plot. The low-energy spectrum obtained from Heff is shown as solid lines for comparison. The peaks in the LDOS reflect the energy necessary to add or re- move one particle. Thus, the peak positions can be es- timated from Heff by the energy difference between the odd-parity eigenstates and the ground state, which yields ± [J/2 + J cos(φ/2) ± J/2 sin(φ/2)]. We find that our numerics is qualitatively consistent with the predictions of Heff , although the numerics is performed in a regime where the addition spectrum already hybridizes with the quasiparticle continuum. Apart from shifts in energy, the hybridization lifts the degeneracies at φ = 0 and 2π. Besides the low energy features which are qualitatively described by Heff , the numerical results also exhibit high- energy features at ±U/2, which are associated with the charge-transfer peaks of the impurity Anderson model. TRITOPS-QD-TRITOPS Josephson junctions com- bine topological superconductivity with time reversal symmetry and electron-electron interactions. While this is superficially similar to quantum spin Hall Josephson junctions including interactions either within the edge states [47, 48] or through coupling to an interacting quan- tum dot [49, 50], these two types of Josephson junctions are governed by remarkably different physics. Quan- tum spin Hall Josephson junctions exhibit an 8π-perodic Josephson effect which can be interpreted as resulting from the tunneling of e/2 charges enabled by the forma- tion of Z4 parafermions or from a spin transmutation as a consequence of the fermion parity anomaly [49]. In con- trast, the present system has a Josephson effect which is FIG. 3. (Color online) LDOS at the quantum dot obtained by QMC. The black dashed lines are the predictions for the peaks in the density of states on the basis of Heff with J = 0.2. The amplitude of the superconducting gap ∆ = 0.2 is indicated in thin lines. Other parameters are U = 4t, J = 0.2, λ = 0.5t, t0 = t, µ = 0 and β = 400. Hamiltonian (for details, see [38], Sec. 3). In a first step, we project the tunneling Hamiltonian Hc to the subgap states of the wires, giving Hc = teff eiφ/4Pσ(cid:16)Γ†L,σdσ + d†σΓR,σ(cid:17) + H.c, where teff . t0 and the Bogoliubov operators for the zero-energy modes satisfy Γ†L = Γ†L,↑ = −iΓR,↓ (see [46] and [38], Sec. 2). Focusing on the particle- hole symmetric point εd = −U/2, and eliminating the empty and doubly occupied states of the quantum dot by a Schrieffer-Wolff transformation, we obtain (see [38], Sec. 3 for details, including more general configurations) = iΓL,↓, and Γ†R = Γ†R,↑ (cid:16)Γ†LΓR − Γ†RΓL(cid:17)(cid:21) (7) φ 2 d(cid:20)(nL + nR − 1) + i sin d ΓRΓL(cid:17)(cid:27) , (cid:16)S−d Γ†LΓ†R − S+ Heff = J(cid:26)Sz φ 2 +i cos The Hamiltonian Heff is easily diagonalized. where J = 4t2 eff /U and we defined the occupations nα = Γ†αΓα. A convenient basis for this Hamiltonian is σ, nL, nRi with nα = 0, 1 and σ =↑,↓. Note that nα also labels the polarization of the fractional spins. It con- serves the number parity of n = nL + nR. For n = 1, the terms involving S±d do not contribute and we find doubly-degenerate eigenstates which are linear superpo- sitions of σ, 1, 0i and σ, 0, 1i with energy ±J/2 sin(φ/2). For even occupations n, we have two phase-independent states with degenerate eigenergies J/2 corresponding to ↑, 1, 1i and ↓, 0, 0i as well as a pair of nondegenerate states with energies −J/2 ± J cos(φ/2), which are linear combinations of the states ↑, 0, 0i and ↓, 1, 1i. -1.8 0.0 0.5 1.0 φ/π 1.5 2.0 1.8 1.2 0.6 ∆ 0.0 −∆ -0.6 -1.2 E(φ) 4π periodic and results from an effective singlet forma- tion with two fractional spins. Acknowledgements. We acknowledge support from CONICET, and UBACyT, Argentina as well as Deutsche Forschungsgemeinschaft and Alexander von Humboldt Foundation, Germany. LA thanks the ICTP-Trieste for hospitality through a Simons associateship. This work was sponsored by PIP 112-201101-00832 of CONICET and PICT 2013-1045, PICT 2012- of the ANPCyT. [1] S. De Franceschi, L. Kouwenhoven, C. Schonenberger and W. Wersdorfer, Nature Nanotechology, 5, 703 (2010). [2] A. Mart´ın-Rodero and A. Levy Yeyati, Adv. Phys. 60, 899 (2011). [3] M. R. Buitelaar, T. Nussbaumer, and C. Schonenberger, Phys. Rev. Lett. 89, 256801 (2002). [4] T. Sand-Jespersen, J. Paaske, B. M. Andersen, K. Grove- Rasmussen, H. I. Jorgensen, M. Aagesen, C. B. Sorensen, P. E. Lindelof, K. Flensberg, and J. Nygard, Phys. Rev. Lett. 99, 126603 (2007). [5] C. Buizert, A. Oiwa, K. Shibata, K. Hirakawa, and S. Tarucha, Phys. Rev. Lett. 99, 136806 (2007). [6] J.-D. Pillet, P. Joyez, R. Zitko, and M. F. Goffman, Phys. Rev. B 88, 045101 (2013). [7] I.O. Kulik, Zh. Eksp. Teor. Fiz. 49, 585 (1965) [Sov. Phys. JETP 22, 841 (1966)]. [8] H. Shiba and T. Soda, Prog. Theor. Phys. 41, 25 (1969). [9] L. I. Glazman and K. A. Matveev, Pisma Zh. Eksp. Teor. Fiz. 49, 570 (1989) [JETP Lett. 49, 659 (1989)]. [10] B. I. Spivak and S. A. Kivelson, Phys. Rev. B 43, 3740 (1991). [11] E. Vecino, A. Martin-Rodero, and A. Levy Yeyati, Phys. Rev. B 68, 035105 (2003). [12] F. Siano and R. Egger , Phys. Rev. Lett. 93, 047002 (2004). [13] M.-S. Chi, M. Lee, K. Kang, and W. Belzig, Phys. Rev. B ( R) 70, 020502 (2004). [14] T. Meng, P. Simon, S. Florens, Phys. Rev. B 79, 224521 (2009). [15] D. J. Luitz and F. F. Assaad, Phys. Rev. B 81, 024509 (2010). [16] D. J. Luitz, F. F. Assaad, T. Novotny, C. Karrasch, and V. Meden, Phys. Rev. Lett. 108, 227001 (2012). [17] A. Oguri, Y. Tanaka, and J. Bauer, Phys. Rev. B 87, 075432 (2013). [18] R. Allub and C. R. Proetto, Phys. Rev. B 91, 045442 (2015). [19] G. Kirsanskas, M. Goldstein, K. Flensberg, L. I. Glaz- man, and J. Paaske, Phys. Rev. B 92, 235422 (2015). [20] J. Alicea, Rep. Prog. Phys. 75, 076501, (2012). [21] A. Y. Kitaev, Physics-Uspekhi 44, 131 (2001). [22] R. M. Lutchyn, J. D. Sau,and S. Das Sarma, Phys. Rev. 5 Lett. 105 077001 (2010). [23] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105 177002 (2010). [24] C. L. M. Wong and K. T. Law, Phys. Rev. B 86, 184516 (2012). [25] S. Nakosai, Y. Tanaka, and N. Nagaosa, Phys. Rev. Lett. 108, 147003 (2012). [26] S. Deng, L. Viola and G. Ortiz, Phys. Rev. Lett. 108, 036803 (2012). [27] F. Zhang, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 111, 056402 (2013). [28] A. Keselman, L. Fu, A. Stern, and E. Berg, Phys. Rev. Lett. 111, 116402 (2013). [29] E. Dumitrescu and S. Tewari, Phys. Rev. B 88, 220505 (2013). [30] S. B. Chung, J. Horowitz, and X-L. Qi, Phys. Rev. B 88, 214514 (2013). [31] S. Nakosai, J. C. Budich, Y. Tanaka, B. Trauzettel, and N. Nagaosa, Phys. Rev. Lett. 110, 117002 (2013). [32] A. Haim, A. Keselman, E. Berg, and Y. Oreg, Phys. Rev. B 89, 220504(R) (2014). [33] E. Gaidamauskas, J. Paaske, and K. Flensberg, Phys. Rev. Lett. 112, 126402 (2014). [34] J. Klinovaja, A. Yacoby, and D. Loss, Phys. Rev. B 90, 155447 (2014). [35] C. Schrade, A. A. Zyuzin, J. Klinovaja, and D. Loss, Phys. Rev. Lett. 115, 237001 (2015). [36] A. Haim, E. Berg, K. Flensberg, and Y. Oreg, Phys. Rev. B 94, 161110(R) (2016). [37] Y. Kim, D. E. Liu, E. Gaidamauskas, J. Paaske, K. Flens- berg, and R. M. Lutchyn, Phys. Rev. B 94, 075439 (2016). [38] Supplementary Material. [39] J.E. Hirsch and R. M. Fye, Phys. Rev. Lett. 56, 2521 (1986). [40] L. Arrachea and M. J. Rozenberg, Phys. Rev. B 72, 041301 (2005). [41] A. Camjayi and L. Arrachea, Phys. Rev. B 86, 235143 (2012) and J. Phys.: Condens. Matter 26 035602 (2014). [42] P. Werner, A. Comanac, L. De Medici, M. Troyer, and A.J. Millis, Phys. Rev. Lett. 97, 076405 (2006). [43] K. Haule, Phys. Rev. B 75, 155113 (2007). [44] A. Hewson, The Kondo Problem to Heavy Fermions, Cambridge University Press (Cambridge, 1993). [45] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). [46] We introduce the notation ΓL/R,σ to make the tunneling Hamiltonian more compact. It also turns out that this no- tation is helpful in performing the Schrieffer-Wolff trans- formation (see [38], Section 3). [47] F. Zhang and C.L. Kane, Phys. Rev. Lett. 113, 036401 (2014). [48] C. P. Orth, R. P. Tiwari, T. Meng, and T. L. Schmidt, Phys. Rev. B 91, 081406(R) (2015). [49] Y. Peng, Y. Vinkler-Aviv, P.W.Brouwer, L.I. Glazman, and F. von Oppen, Phys. Rev. Lett. 117, 267001 (2016). [50] H.-Y. Hui and J.D. Sau, Phys. Rev. B 95, 014505 (2017). 1. JOSEPHSON CURRENT AND GREEN FUNCTIONS Starting from the definition, the Josephson current it may be written as I = −2t0Xσ Imhhc†α,1,σdσii , I = − 2t0 β Xσ Xn ImhG(11) 1α,d,σ(iωn)i , 6 (8) (9) where G(11) Green function 1α,d,σ(iωn) is the Matsubara component of the matrix element (1, 1) of the Nambu-Gorkov imaginary-time G1α,d,σ(τ ) = −hTτhcα,1,σ(τ ) d†σ(0)ii −hTτhc†α,1,σ(τ ) d†σ(0)ii −hTτhcα,1,σ(τ ) dσ(0)ii −hTτhc†α,1,σ(τ ) dσ(0)ii The latter satisfies the following Dyson equation   . G1α,d,σ(τ ) = g1α,σ(τ )tc Gd,σ(τ ), with tc = −t0 τ3, being τ3 the third Pauli matrix and g1α,σ(τ ) = −hTτhcα,1,σ(τ )c†α,1,σ(0)ii0 −hTτhc†α,1,σ(τ )c†α,1,σ(0)ii0 −hTτ [cα,1,σ(τ )cα,1,σ(0)]i0 −hTτhc†α,1,σ(τ )cα,1,σ(0)ii0 . (10) (11) (12) (15) Here hi0 denotes the mean value with respect to the density matrix of the disconnected wire, while hi is the mean value with respect to the density matrix of the full Hamiltonian. Substituting (11) in (9) leads to I = 2t02 β Xσ Xn Imhg(12) 1α,σ(iωn)G(21) d,σ (iωn)i . (13) 2. CONTINUUM MODEL FOR THE TRITOPS WIRE The Hamiltonian proposed in Ref. [1], written in Eq. (1) of the main text, expressed in k-space reads H =Xk,σ {[−2t cos(ka) + 2λσ sin(ka) − µ] c†k,σck,σ + 2∆ cos(ka)c†k,σc† −k,−σ + H.c}, (14) where a is the lattice constant. Without the pairing term, this Hamiltonian defines dispersion relations for ↑ and ↓ fermions, which are shifted one another along the k-axis as a consequence of the spin-orbit term. Hence, there are in general four Fermi vectors, kα,σ, where α = l, r denotes left and right movers, respectively. They satisfy kr,σ = −kl,σ, with ↑ =↓ and ↓ =↑. In addition, the pairing term opens a gap ∆ cos(k), which vanishes at k = ±π/2 and has different signs for k < π/2 and for k > π/2. The low-energy model for this Hamiltonian is derived by writing cj,σ ∼ √a(cid:2)eikr,σjaψr,σ + e−ikl,σjaψl,σ(cid:3) , where ψα,σ(x), with α = L, R are fermionic fields for the left and right movers, respectively, with spin σ. Let us focus on the case where kr,↑ = −kl,↓ = π/2, which corresponds to the critical case where the superconducting Instead, the other fermions at the Fermi level with kr,↓ = −kl,↑ are under the effect of a finite gap vanishes. superconducting pairing, and a gap opens in the spectrum at these points. For the derivation of the low-energy model in this particular case we project on the gapless states and (15) reads cj,↑ ∼ √aeiπja/2ψr,↑, cj,↓ ∼ √ae−iπja/2ψl,↓, (16) 7 where the indices l, r are now redundant and will be omitted. Substituting in (14) and, keeping the leading orders for each of the terms of the Hamiltonian, we get the following low-energy continuum model H ∼ (2λ − µ)Xσ Z dxψ†σ(x)ψσ(x) + 2taXσ Z dxψ†σ(x)i∂xsσψσ(x) + 2∆aZ dxψ† ↑ (x)i∂xψ† ↓ (x) + H.c., (17) with s↑,↓ =↑,↓. For simplicity, and without affecting the key features of the original model, we will drop the second term ∝ t. The same arguments above can be repeated for kr,↓ = −kl,↑ = π/2, in which case the pairing between states with kr,↑ = −kl,↓ have different sign with respect to the ones in previous case. Hence, we get the Hamiltonian (17) but with a − sign in front of the last term. The corresponding Bogoliubov-de Gennes matrix for the full low-energy Hamiltonian reads HT RIT OP S = (2λ − µ)τz − 2a∆pτxσz, (18) where τx,y,z are Pauli matrices acting on the spinors ψ†p↑ The topological phase takes place for 2λ − µ > 0, while 2λ − µ < 0, corresponds to a trivial superconductor. If we define a domain wall where 2λ − µ changes sign, we get zero-energy modes localized at the domain wall. To get the structure of the zero-energy modes, we consider a linear domain wall centered at x = 0, i.e., 2λ(x) − µ(x) = αx. For α > 0, we have a nontopological (topological) superconducting phase for x < 0 (x > 0), while the opposite situation takes place for α < 0. Following Ref. [2] we can calculate the zero-energy modes localized at the wall by calculating H 2 and identifying it with the Hamiltonian for an Harmonic oscillator. The solutions for the zero-modes read c−p↓(cid:17), while σz acts on the spin degrees of freedom. =(cid:16)c†p↑ Γ↑ =Z dxφ0(x)hψ↑(x) ± iψ† ↓ (x)i , Γ↓ =Z dxφ0(x)hψ↓(x) ± iψ† ↑ (x)i , (19) where the ± corresponds to α <, > 0, respectively and φ0(x) is the zero-mode wave function of the effective harmonic- oscillator Hamiltonian. If we now represent the junction between the TRITOPS wires with a domain wall at the left with α > 0 followed by another at the right with α < 0. The corresponding zero-modes will have the structure of the Bogoliubov operators Γ†L = Γ†L,↑ = iΓL,↓, Γ†R = Γ†R,↑ = −iΓR,↓. (20) 3. SCHRIEFFER-WOLFF TRANSFORMATION AND DERIVATION OF THE EFFECTIVE LOW-ENERGY HAMILTONIAN In this Section, we derive the effective Hamiltonian Heff in the limit of occupation 1 at the dot for arbitrary coupling to the left and right leads tL, tR (cid:28) −εd, U + εd. This is analogous to the derivation of a Kondo Hamiltonian from the Anderson model [3, 4] Projecting the operators of the left and right superconductors attached to the quantum dot onto the low-energy subgap modes leads to the following Hamiltonian to describe the low-energy physics Hlow = −eiφ/4Xσ (cid:16)tLΓ†L,σdσ + tRd†σΓR,σ(cid:17) + H.c. + Hd, (21) where tL, tR . t0 represent the coupling to the TRITOPS leads and Γα,σ are the Bogoliubov operators corresponding to the zero-modes of the wires localized at the left and right sides of the quantum dot which satisfy the properties (20). In the limit of tL, tR (cid:28) −εd, U + εd, we can eliminate the high-energy states of the quantum dot by recourse to a Schrieffer-Wolff transformation [4]. Since Γα,↑ and Γα,↓ are related by Eqs. (20), we first write the Hamiltonian Hlow in terms of the independent operators ΓL,↑ = Γ↑ and ΓR,↓ = Γ↓. This choice leads to a form of Heff with spin-spin interactions similar to the usual Kondo model. Following the conventional procedure [3] we obtain Heff = −(JLLnd↑ + JRRnd↓)/2 + Sz +JLR cos(φ/2)(cid:0)S+ d S−Γ + S−d S+ d [JLLn↑ − JRRn↓ + JLR sin(φ/2)(cid:16)Γ† Γ(cid:1) − iW cos(φ/2)(cid:16)Γ† ↓ − Γ↓Γ↑(cid:17) . Γ† ↑ Γ† ↓ + Γ↓Γ↑(cid:17)] (22) ↑ 8 (23) sin2(φ/2) + W 2 cos2(φ/2). (27) This implies the lifting of ↑,↓ degeneracies of the eigenstates and the opening of a gap at φ = π. The consequence of the gap opening is a change in the periodicity of the Josephson current to 2π, as in junctions of non-topological superconductors. The ground state remains in the even space and the behavior of the Josephson current corresponds to the 0-phase. In the case of a magnetic field perpendicular to the direction of the spin-orbit interaction B = Bex the ground state is also in the even subspace. Some degeneracies are lifted but the four-fold degeneracy at φ = π remains and the Josephson current preserves the 4π-degeneracy and the behavior of the Josephson current corresponds to the 0-phase. Results for B = Bez and B = Bex are shown in Fig. 4 for the half-filled configuration and in Fig. 5 away from half-filling. The non-symmetric coupling to the leads tLL 6= tRR does not change qualitatively these pictures. [2] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. [1] F. Zhang, C. L. Kane and E. J. Mele, Phys. Rev. Lett. 111, 056402 (2013). (26) LR cos2(φ/2), 1 2 ± JLL + JRR 2 E±e = − Γ = Γ† ↑ . This Hamiltonian conserves the parity number and total spin projection Sz = Sz Γ = (n↑ − n↓) /2, with nσ = Γ†σΓσ, S+ The spin operators are Sz and S+,−d main text corresponds to (22) for tLL = tRR = t and εd = −U/2. Restricting for the moment to the subspace with total even number of particles [for which the terms of Eq. (22) containing Γ† and Γ↓Γ↑ become irrelevant], in the case of symmetric coupling where JLL = JRR = JLR = J, the ↑ Hamiltonian reduces to the Heisenberg Hamiltonian with antiferromagnetic coupling 2J along z and J cos(φ/2) on the x− y plane. Interestingly, the exchange interaction takes place between the 1/2-spin localized at the quantum dot and an effective 1/2-spin formed by the 1/4-spin excitations localized at the ends of the chains close to the quantum dot. In the general case, the lowest-energy eigenstates and can be written as ψ±e i = ( ↑↓i ± ↓↑i) /√2, where the first entry corresponds to the spin of the dot and the second one to the effective spin of the side-chains. The corresponding eigenenergies are: Γ↑ and similar expressions for Sz d Γ. Eq. (7) in the d + Sz Γ† ↓ E±e = −(JLL + JRR)/2 ± JLR cos(φ/2). (24) The two additional eigenstates are σσi, with σ =↑,↓ and have zero energy. In the subspace with odd number of particles, the eigenstates are a mixture of σ, 0i and σ,↑↓i, with eigenenergies E±o = − JLL + JRR 4 ±s(cid:20) JLL − JRR 4 (cid:21)2 + J 2 LR 4 sin2(φ/2) + W 2 cos2(φ/2) (25) and they are doubly degenerate due to the two possible orientations of σ at the dot. JLL = JRR = J and W = 0. The eigenenergies given in the main text correspond (up to an energy shift of −J/2) to Eqs. (24) and (25) for For arbitrary φ, the GS of the full Hamiltonian is in the even subspace. For φ = (2m + 1)π, with m integer, the states ψ±e i cross and they become also degenerate with the doublet of the odd subspace, hence, defining a 4-fold degenerate level crossing. In all the cases, this main features do not depend on the value of εd and tL, tR. Due to this level crossing, the Josephson current has a peridicity of 4π. The diagonalization of the effective Hamiltonian Heff leads to the 8 lowest-energy states of the spectrum of Hlow. The latter Hamiltonian has 8 additional higher energy states corresponding to the dot empty or doubly occupied. The effect of a magnetic field B locally applied at the quantum dot can be analyzed in the limit of B (cid:28) U by adding to the effective Hamiltonian of Eq. (22) a term B · Sd. In the case of a magnetic field in the direction of the spin-orbit interaction, B = Bez, the eigenenergies of Eq. (24) are modified to with Jαβ = 2U tαtβ (εd + U )(−εd) , W = . (2εd + U )tLtR (εd + U )(−εd) Γ↓, S−Γ = Γ† ↓ qB2 + 4J 2 (cid:19)2 J 2 LR 4 + ±s(cid:18) JLL − JRR 4 B 2 ± JLL + JRR 4 Eo = − while the eigenenergies (25) change to 9 , , FIG. 4. (Color online) Spectrum of Heff for the quantum dot at half filling εd = −U/2 and symmetric coupling to the leads tLL = tRR, corresponding to J = 1, W = 0. Left: B = 0, center B = Bez and right B = Bex. Black and red corresponds, respectively to even and odd subspaces. , , FIG. 5. (Color online) Spectrum of Heff for the quantum dot away from half-filling and symmetric coupling to the leads tLL = tRR, corresponding to J = W = 1. Left: B = 0, center B = Bez and right B = Bex. Black and red corresponds, respectively to even and odd subspaces. 105 177002 (2010). [4] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 [3] A. Hewson, The Kondo Problem to Heavy Fermions, Cam- (1966). bridge University Press (Cambridge, 1993). [5] We consider a general situation where the coupling to left and right may be different. 0.5 φ/π 1 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.5 φ/π 1 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.2 0.4 0.6 0.8 1 φ/π 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.5 φ/π 1 0.5 0 -0.5 -1 -1.5 -2 0 E(φ)/J 0.5 φ/π 1 0 -1 E(φ)/J -2 0 0.5 φ/π 1 0.5 0 -1 -0.5 E(φ)/J -1.5 -2 0 Supplemental Material: Fractional spin and Josephson effect in time-reversal-invariant topological superconductors Alberto Camjayi,1 Liliana Arrachea,2 Armando Aligia,3 and Felix von Oppen4 1Departamento de F´ısica, FCEyN, Universidad de Buenos Aires and IFIBA, Pabell´on I, Ciudad Universitaria, 1428 CABA Argentina 2International Center for Advanced Studies, UNSAM, Campus Miguelete, 3Centro At´omico Bariloche and Instituto Balseiro, CNEA, 8400 S. C. de Bariloche, Argentina 4Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, 14195 Berlin, Germany 25 de Mayo y Francia, 1650 Buenos Aires, Argentina 1. JOSEPHSON CURRENT AND GREEN FUNCTIONS Starting from the definition, the Josephson current it may be written as I = −2t0Xσ β Xσ Xn 2t0 Imhhc†α,1,σdσii , 1α,d,σ(iωn)i , ImhG(11) I = − (1) (2) where G(11) Green function 1α,d,σ(iωn) is the Matsubara component of the matrix element (1, 1) of the Nambu-Gorkov imaginary-time G1α,d,σ(τ ) = −hTτhcα,1,σ(τ ) d†σ(0)ii −hTτhc†α,1,σ(τ ) d†σ(0)ii −hTτhcα,1,σ(τ ) dσ(0)ii −hTτhc†α,1,σ(τ ) dσ(0)ii   . The latter satisfies the following Dyson equation G1α,d,σ(τ ) = g1α,σ(τ )tc Gd,σ(τ ), with tc = −t0τ 3, being τ3 the third Pauli matrix and arXiv:1612.07410v2 [cond-mat.mes-hall] 31 Jul 2017 Here hi0 denotes the mean value with respect to the density matrix of the disconnected wire, while hi is the mean value with respect to the density matrix of the full Hamiltonian. Substituting (4) in (2) leads to g1α,σ(τ ) = −hTτhcα,1,σ(τ )c†α,1,σ(0)ii0 −hTτhc†α,1,σ(τ )c†α,1,σ(0)ii0 −hTτ [cα,1,σ(τ )cα,1,σ(0)]i0 −hTτhc†α,1,σ(τ )cα,1,σ(0)ii0 . β Xσ Xn d,σ (iωn)i . Imhg(12) 1α,σ(iωn)G(21) 2t02 I = 2. CONTINUUM MODEL FOR THE TRITOPS WIRE The Hamiltonian proposed in Ref. [1], written in Eq. (1) of the main text, expressed in k-space reads H =Xk,σ {[−2t cos(ka) + 2λσ sin(ka) − µ] c†k,σck,σ + 2∆ cos(ka)c†k,σc† −k,−σ + H.c}, where a is the lattice constant. Without the pairing term, this Hamiltonian defines dispersion relations for ↑ and ↓ fermions, which are shifted one another along the k-axis as a consequence of the spin-orbit term. Hence, there are in (3) (4) (5) (6) (7) cj,σ ∼ √a(cid:2)eikr,σjaψr,σ + e−ikl,σjaψl,σ(cid:3) , where ψα,σ(x), with α = L, R are fermionic fields for the left and right movers, respectively, with spin σ. Let us focus on the case where kr,↑ = −kl,↓ = π/2, which corresponds to the critical case where the superconducting Instead, the other fermions at the Fermi level with kr,↓ = −kl,↑ are under the effect of a finite gap vanishes. superconducting pairing, and a gap opens in the spectrum at these points. For the derivation of the low-energy model in this particular case we project on the gapless states and (8) reads cj,↑ ∼ √aeiπja/2ψr,↑, cj,↓ ∼ √ae−iπja/2ψl,↓, (9) where the indices l, r are now redundant and will be omitted. Substituting in (7) and, keeping the leading orders for each of the terms of the Hamiltonian, we get the following low-energy continuum model H ∼ (2λ − µ)Xσ Z dxψ†σ(x)ψσ(x) + 2taXσ Z dxψ†σ(x)i∂xsσψσ(x) + 2∆aZ dxψ† ↑ (x)i∂xψ† ↓ (x) + H.c., (10) with s↑,↓ =↑,↓. For simplicity, and without affecting the key features of the original model, we will drop the second term ∝ t. The same arguments above can be repeated for kr,↓ = −kl,↑ = π/2, in which case the pairing between states with kr,↑ = −kl,↓ have different sign with respect to the ones in previous case. Hence, we get the Hamiltonian (10) but with a − sign in front of the last term. The corresponding Bogoliubov-de Gennes matrix for the full low-energy Hamiltonian reads HT RIT OP S = (2λ − µ)τz − 2a∆pτxσz, where τx,y,z are Pauli matrices acting on the spinors ψ†p↑ The topological phase takes place for 2λ − µ > 0, while 2λ − µ < 0, corresponds to a trivial superconductor. If we define a domain wall where 2λ − µ changes sign, we get zero-energy modes localized at the domain wall. To get the structure of the zero-energy modes, we consider a linear domain wall centered at x = 0, i.e., 2λ(x) − µ(x) = αx. For α > 0, we have a nontopological (topological) superconducting phase for x < 0 (x > 0), while the opposite situation takes place for α < 0. Following Ref. [2] we can calculate the zero-energy modes localized at the wall by calculating H 2 and identifying it with the Hamiltonian for an Harmonic oscillator. The solutions for the zero-modes read c−p↓(cid:17), while σz acts on the spin degrees of freedom. =(cid:16)c†p↑ 2 (8) (11) (12) general four Fermi vectors, kα,σ, where α = l, r denotes left and right movers, respectively. They satisfy kr,σ = −kl,σ, with ↑ =↓ and ↓ =↑. In addition, the pairing term opens a gap ∆ cos(k), which vanishes at k = ±π/2 and has different signs for k < π/2 and for k > π/2. The low-energy model for this Hamiltonian is derived by writing Γ↑ =Z dxφ0(x)hψ↑(x) ± iψ† ↓ (x)i , Γ↓ =Z dxφ0(x)hψ↓(x) ± iψ† ↑ (x)i , where the ± corresponds to α <, > 0, respectively and φ0(x) is the zero-mode wave function of the effective harmonic- oscillator Hamiltonian. If we now represent the junction between the TRITOPS wires with a domain wall at the left with α > 0 followed by another at the right with α < 0. The corresponding zero-modes will have the structure of the Bogoliubov operators Γ†L = Γ†L,↑ = iΓL,↓, Γ†R = Γ†R,↑ = −iΓR,↓. (13) 3. SCHRIEFFER-WOLFF TRANSFORMATION AND DERIVATION OF THE EFFECTIVE LOW-ENERGY HAMILTONIAN In this Section, we derive the effective Hamiltonian Heff in the limit of occupation 1 at the dot for arbitrary coupling to the left and right leads tL, tR (cid:28) −εd, U + εd. This is analogous to the derivation of a Kondo Hamiltonian from the Anderson model [3, 4] Projecting the operators of the left and right superconductors attached to the quantum dot onto the low-energy subgap modes leads to the following Hamiltonian to describe the low-energy physics Hlow = −eiφ/4Xσ (cid:16)tLΓ†L,σdσ + tRd†σΓR,σ(cid:17) + H.c. + Hd, (14) where tL, tR . t0 represent the coupling to the TRITOPS leads and Γα,σ are the Bogoliubov operators corresponding to the zero-modes of the wires localized at the left and right sides of the quantum dot which satisfy the properties (13). In the limit of tL, tR (cid:28) −εd, U + εd, we can eliminate the high-energy states of the quantum dot by recourse to a Schrieffer-Wolff transformation [4]. Since Γα,↑ and Γα,↓ are related by Eqs. (13), we first write the Hamiltonian Hlow in terms of the independent operators ΓL,↑ = Γ↑ and ΓR,↓ = Γ↓. This choice leads to a form of Heff with spin-spin interactions similar to the usual Kondo model. Following the conventional procedure [3] we obtain 3 Heff = −(JLLnd↑ + JRRnd↓)/2 + Sz +JLR cos(φ/2)(cid:0)S+ d S−Γ + S−d S+ d [JLLn↑ − JRRn↓ + JLR sin(φ/2)(cid:16)Γ† ↓ − Γ↓Γ↑(cid:17) . Γ(cid:1) − iW cos(φ/2)(cid:16)Γ† Γ† ↑ ↑ with Γ† ↓ + Γ↓Γ↑(cid:17)] (15) (16) Jαβ = 2U tαtβ (εd + U )(−εd) , W = . (2εd + U )tLtR (εd + U )(−εd) Γ↓, S−Γ = Γ† ↓ Γ = Γ† ↑ . This Hamiltonian conserves the parity number and total spin projection Sz = Sz Γ = (n↑ − n↓) /2, with nσ = Γ†σΓσ, S+ The spin operators are Sz and S+,−d main text corresponds to (15) for tLL = tRR = t and εd = −U/2. Restricting for the moment to the subspace with total even number of particles [for which the terms of Eq. (15) containing Γ† and Γ↓Γ↑ become irrelevant], in the case of symmetric coupling where JLL = JRR = JLR = J, the ↑ Hamiltonian reduces to the Heisenberg Hamiltonian with antiferromagnetic coupling 2J along z and J cos(φ/2) on the x− y plane. Interestingly, the exchange interaction takes place between the 1/2-spin localized at the quantum dot and an effective 1/2-spin formed by the 1/4-spin excitations localized at the ends of the chains close to the quantum dot. In the general case, the lowest-energy eigenstates and can be written as ψ±e i = ( ↑↓i ± ↓↑i) /√2, where the first entry corresponds to the spin of the dot and the second one to the effective spin of the side-chains. The corresponding eigenenergies are: Γ↑ and similar expressions for Sz d Γ. Eq. (7) in the d + Sz Γ† ↓ E±e = −(JLL + JRR)/2 ± JLR cos(φ/2). (17) The two additional eigenstates are σσi, with σ =↑,↓ and have zero energy. In the subspace with odd number of particles, the eigenstates are a mixture of σ, 0i and σ,↑↓i, with eigenenergies E±o = − JLL + JRR 4 ±s(cid:20) JLL − JRR 4 (cid:21)2 + J 2 LR 4 sin2(φ/2) + W 2 cos2(φ/2) (18) and they are doubly degenerate due to the two possible orientations of σ at the dot. JLL = JRR = J and W = 0. The eigenenergies given in the main text correspond (up to an energy shift of −J/2) to Eqs. (17) and (18) for For arbitrary φ, the GS of the full Hamiltonian is in the even subspace. For φ = (2m + 1)π, with m integer, the states ψ±e i cross and they become also degenerate with the doublet of the odd subspace, hence, defining a 4-fold degenerate level crossing. In all the cases, this main features do not depend on the value of εd and tL, tR. Due to this level crossing, the Josephson current has a peridicity of 4π. The diagonalization of the effective Hamiltonian Heff leads to the 8 lowest-energy states of the spectrum of Hlow. The latter Hamiltonian has 8 additional higher energy states corresponding to the dot empty or doubly occupied. The effect of a magnetic field B locally applied at the quantum dot can be analyzed in the limit of B (cid:28) U by adding to the effective Hamiltonian of Eq. (15) a term B · Sd. In the case of a magnetic field in the direction of the spin-orbit interaction, B = Bez, the eigenenergies of Eq. (17) are modified to (19) LR cos2(φ/2), 1 2 ± JLL + JRR 2 E±e = − sin2(φ/2) + W 2 cos2(φ/2). (20) qB2 + 4J 2 (cid:19)2 J 2 LR 4 + ±s(cid:18) JLL − JRR 4 B 2 ± JLL + JRR 4 Eo = − while the eigenenergies (18) change to 4 , , FIG. 1. (Color online) Spectrum of Heff for the quantum dot at half filling εd = −U/2 and symmetric coupling to the leads tLL = tRR, corresponding to J = 1, W = 0. Left: B = 0, center B = Bez and right B = Bex. Black and red corresponds, respectively to even and odd subspaces. , , FIG. 2. (Color online) Spectrum of Heff for the quantum dot away from half-filling and symmetric coupling to the leads tLL = tRR, corresponding to J = W = 1. Left: B = 0, center B = Bez and right B = Bex. Black and red corresponds, respectively to even and odd subspaces. This implies the lifting of ↑,↓ degeneracies of the eigenstates and the opening of a gap at φ = π. The consequence of the gap opening is a change in the periodicity of the Josephson current to 2π, as in junctions of non-topological superconductors. The ground state remains in the even space and the behavior of the Josephson current corresponds to the 0-phase. In the case of a magnetic field perpendicular to the direction of the spin-orbit interaction B = Bex the ground state is also in the even subspace. Some degeneracies are lifted but the four-fold degeneracy at φ = π remains and the Josephson current preserves the 4π-degeneracy and the behavior of the Josephson current corresponds to the 0-phase. Results for B = Bez and B = Bex are shown in Fig. 1 for the half-filled configuration and in Fig. 2 away from half-filling. The non-symmetric coupling to the leads tLL 6= tRR does not change qualitatively these pictures. [1] F. Zhang, C. L. Kane and E. J. Mele, Phys. Rev. Lett. 111, 056402 (2013). [2] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105 177002 (2010). [3] A. Hewson, The Kondo Problem to Heavy Fermions, Cam- bridge University Press (Cambridge, 1993). [4] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966). [5] We consider a general situation where the coupling to left and right may be different. 0.5 φ/π 1 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.5 φ/π 1 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.2 0.4 0.6 0.8 1 φ/π 0 -0.5 -1 E(φ)/J -1.5 -2 0 0.5 φ/π 1 0.5 0 -0.5 -1 -1.5 -2 0 E(φ)/J 0.5 φ/π 1 0 -1 E(φ)/J -2 0 0.5 φ/π 1 0.5 0 -1 -0.5 E(φ)/J -1.5 -2 0 

1403.4464

2

1403

Effects of tilting the magnetic field in 1D Majorana nanowires

We investigate the effects that a tilting of the magnetic field from the parallel direction has on the states of a 1D Majorana nanowire. Particularly, we focus on the conditions for the existence of Majorana zero modes, uncovering an analytical relation (the sine rule) between the field orientation relative to the wire, its magnitude and the superconducting parameter of the material. The study is then extended to junctions of nanowires, treated as magnetically inhomogeneous straight nanowires composed of two homogeneous arms. It is shown that their spectrum can be explained in terms of the spectra of two independent arms. Finally, we investigate how the localization of the Majorana mode is transferred from the magnetic interface at the corner of the junction to the end of the nanowire when increasing the arm length.

cond-mat.mes-hall

cond-mat

Effects of tilting the magnetic field in 1D Majorana nanowires Javier Osca,1, ∗ Daniel Ruiz,1 and Lloren¸c Serra1, 2 1Institut de F´ısica Interdisciplin`aria i de Sistemes Complexos IFISC (CSIC-UIB), E-07122 Palma de Mallorca, Spain 2Departament de F´ısica, Universitat de les Illes Balears, E-07122 Palma de Mallorca, Spain (Dated: February 25, 2014) We investigate the effects that a tilting of the magnetic field from the parallel direction has on the states of a 1D Majorana nanowire. Particularly, we focus on the conditions for the existence of Majorana zero modes, uncovering an analytical relation (the sine rule) between the field orientation relative to the wire, its magnitude and the superconducting parameter of the material. The study is then extended to junctions of nanowires, treated as magnetically inhomogeneous straight nanowires composed of two homogeneous arms. It is shown that their spectrum can be explained in terms of the spectra of two independent arms. Finally, we investigate how the localization of the Majorana mode is transferred from the magnetic interface at the corner of the junction to the end of the nanowire when increasing the arm length. PACS numbers: 73.63.Nm,74.45.+c I. INTRODUCTION In 2003 Kitaev pointed out the usefulness of topolog- ical states for quantum computing operations.1 Essen- tially, topological states are quantum states with a hid- den internal symmetry.2 They are usually localized close to the system edges or interfaces and their nonlocal na- ture gives them a certain degree of immunity against local sources of noise. A subset of this kind of states called Majorana edge states is attracting much interest in condensed matter physics.3–13 Majorana states are ef- fectively chargeless zero-energy states that behave as lo- calized non abelian anyons. It is theorized that nontriv- ial phases arise from their mutual interchange, caused by their nonlocal properties.14,15 Furthermore, these states have the property of being their own anti-states, giving rise to statistical behavior that is neither fermionic nor bosonic. Instead, the creation of two Majorana quasi- particle excitations in the same state returns the system to its equilibrium state. This kind of quasiparticles in- herits its name from Ettore Majorana who theorized the existence of fundamental particles with similar statistical properties.16 Majorana states have been theoretically predicted in many different systems and some of them have been re- alized experimentally. In particular, evidences of their formation at the ends of semiconductor quantum wires inside a magnetic field with strong spin-orbit interaction and in close proximity to a superconductor have been seen in Refs. 17–21. Superconductivity breaks the charge symmetry creating quasiparticle states without a defined charge that are a mixture of electron and hole excita- tions. On the other hand, the spin-orbit Rashba effect is caused by an electric field perpendicular to the prop- agation direction that breaks the inversion symmetry of the system while the external magnetic field breaks the spin rotation symmetry of the nanowire. The combined action of both effects makes the resulting state effectively spinless and, including superconductivity, also effectively chargeless and energyless.22–38 This work addresses the physics of 1D nanowires with varying relative orientations between the external mag- netic field and the nanowire (see Fig. 1). This physics is of relevance, e.g., for the exchange of Majoranas on net- works of 1D wires, where it has been suggested that Ma- joranas can be braided by manipulating the wire shapes and orientations.39–41 The Hamiltonian of the system is expressed in the continuum and the analysis is per- formed using two complementary approaches: the com- plex band structure of the homogeneous wire and the numerical diagonalization for finite systems. The com- plex band structure allows a precise characterization of the parameter regions of the semi-infinite wire where Ma- joranas, if present, are not distorted by finite size effects. On the contrary numerical diagonalizations of finite sys- tems, even though reflecting the same underlying physics, yield smoothened transitions between different physical regions of parameter space. For the semi-infinite system we uncover an analytical law limiting the existence of Majorana modes below crit- ical values of the angles between the magnetic field and the nanowire. This law, referred to in this article as the sine rule, is shown to be approximately valid in finite sys- tems too. We find a correspondence of the finite system spectrum with its infinite wire counterpart, explaining this way its distinctive features and regimes in simplest terms. The results for the homogeneous nanowire are subse- quently used to explain the spectrum of a junction of two nanowires with arbitrary angle. The junction is mod- eled as a non homogeneous straight nanowire with two regions characterized by different magnetic field orienta- tions (see Fig. 1b). While the magnetic field remains par- allel to the nanowire in one arm, we study the spectrum variation when changing the magnetic field angles in the other. Similarities between the homogeneous and inho- mogeneous nanowire spectra allow us to explain many of the features of the latter in terms of those of the former. Finally, we investigate the dependence with the distance of the magnetic interface (the corner of the junction) to the end of the nanowire, finding a transfer phenomenon where the Majoranas change localization from the inter- face for a short arm to the nanowire end as the arm length is increased. This work is organized as follows. In Sec. II the phys- ical model is introduced and Sec. III presents the above mentioned sine rule. In Sec. IV we discuss the spectrum of excited states of a homogeneous nanowire while in Sec. V we address an inhomogeneous system representing a nanowire junction. We study changes in the spectrum due to the tilting (V A) and stretching (V B) of one of the junction arms. Finally, the conclusions of the work can be found in Sec. VI. II. PHYSICAL MODEL We assume a one dimensional model of a semiconduc- tor nanowire as a low energy representation of a higher dimensional wire with lateral extension, when only the first transverse mode is active. The system is described by a Hamiltonian of the Bogoliubov-deGennes kind, HBdG = (cid:18) p2 x 2m + V (x) − µ(cid:19) τz +∆B (sin θ cos φ σx + sin θ sin φ σy + cos θ σz) + ∆s τx + α h pxσyτz , (1) where the different terms are, in left to right order, ki- netic, electric and chemical potential, Zeeman, supercon- ducting and Rashba spin-orbit terms. The Pauli oper- ators for spin are represented by σx,y,z while those for isospin are given by τx,y,z. Superconductivity is modeled as an s-wave superconductive term that couples different states of charge. The superconductor term in Eq. (1) is an effective mean field approximation to a more complicated phonon- assisted attractive interaction between electrons. This interaction leads to the formation of Cooper pairs with break-up energy ∆s. Experimentally, superconductivity can be achieved by close proximity between the semi- conductor nanowire and a metal superconductor. The semiconductor wire becomes superconducting when its width is smaller than the coherence length of the Cooper pairs. On the other hand, the Rashba spin orbit term arises from the self-interaction between an electron (or hole) spin with its own motion. This self interaction is due to the presence of a transverse electric field that is perceived as an effective magnetic field in the rest frame of the quasiparticle. This electric field can be induced ex- ternally but, usually, is a by-product of an internal asym- metry of the nanostructure. In the Hamiltonian Eq. (1) we have taken x as the orientation of the 1D nanowire while an effective spin orbit magnetic field ~Bso pointing along y may be defined due to the coupling of the Rashba term with the y component of the spin. We consider the nanowire in an external magnetic field, giving spin splittings through the Zeeman term in Eq. (1). 2 a) b) FIG. 1. Sketches of the physical systems considered in this work. a) Straight nanowire on the x axis in a homogeneous magnetic field characterized by spherical angles θ and φ. b) Junction of two nanowires in a uniform magnetic field (left) represented as a straight nanowire with a magnetic inhomo- geneity (right). In this paper we assume the magnetic field in arbitrary di- rection, including the possibility of being inhomogeneous in space for some setups. The direction of the magnetic field is parametrized by the spherical polar and azimuthal angles θ and φ. These two angles are constant for a ho- mogeneous wire (Fig. 1a) and they change smoothly from one to the other arm in a nanowire junction (Fig. 1b). Summarizing, superconductor, Rashba spin-orbit and Zeeman effects are parametrized in Eq. (1) by ∆s, α and ∆B, respectively. These parameters are taken constant because the nanowire is considered to be made of an ho- mogeneous material. The only inhomogeneity allowed in certain cases is a change in the magnetic field direction at a single magnetic interface between two homogeneous regions. Along this work the Hamiltonian of Eq. (1) is solved for homogeneous parameters in the infinite, semi-infinite and finite wires, as well as for the inhomogeneous finite case, using different approaches. When a direct diago- nalization of the Hamiltonian for a finite system is per- formed, soft potential edges and magnetic interface are used. The shape of the potential edges is modeled as Fermi-like functions centered on those edges. High po- tential is imposed outside the nanowire while low poten- tial (usually zero) is assumed inside. When a magnetic interface is present a smooth variation in the field angles is modeled in the same way. Specifically, those smooth functions read V (x) = V0 [1 + F (x; xL, sv) − F (x; xR, sv)] , θ(x) = θL + (θR − θL) [1 − F (x; xm, sm)] , φ(x) = φL + (φR − φL) [1 − F (x; xm, sm)] , (2) (3) (4) for the potential and the field polar and azimuthal angles, 3 in the semi-infinite system and use those results to un- derstand the physics of Majoranas in a finite system. In this approach we eliminate from the analysis the finite size effects caused by the overlapping of the Majorana wave functions at both ends of a finite nanowire. Al- though it is obvious that for long enough wires the size effect becomes negligible, disentangling finite size behav- ior from intrinsic Majorana physics using calculations of only finite systems is much less obvious. Majorana mode creation has been understood as a phase transition of the lowest excited state, signaled by the closing and reopening of a gap in the infinite nanowire band spectrum,23 as shown in Figs. 3a and 3b. The phase transition follows in this case a well known law, requiring high-enough fields for Majoranas to exist, ∆B ≥ p∆2 s + µ2 . (8) Notice that, as mentioned, for the equality in Eq. (8) a gap closes for k = 0 in Fig. 3b. In Ref. 37 Eq. (8) was derived, in an alternative way, from the analysis of the complex-k solutions compatible with the boundary condition of a semi-infinite nanowire in a parallel field. This approach relies on the property that the complex band structure (allowing an imaginary part in k) of the homogeneous wire contains all the in- formation about all possible eigenstates of any piecewise homogeneous wire. In general, an eigenstate of the infi- nite homogeneous wire with a given arbitrary k can be expressed as sσsτ eikx χsσ (ησ)χsτ (ητ ) , (9) Ψ(k) Ψ(k)(x, ησ, ητ ) = Xsσ,sτ where Ψ(k) bers are sσ = ± and sτ = ±. sσsτ are state amplitudes and the quantum num- The sharp semi-infinite wire with x > 0 is obviously piecewise homogeneous, implying that the Majorana so- lution allowed by the existence of an edge at x = 0 must be a linear superposition of the homogeneous nanowire eigenstates of complex wave number with Im(k) > 0, otherwise it could not be a localized state. The resulting restriction is Xk, Im(k)>0 CkΨ(k) sσsτ = 0 , (10) where the Ck's are complex numbers characterizing the superposition of state amplitudes. The allowed wave numbers are calculated solving the determinant det(cid:8)Hsσ sτ ,s′ σs′ τ (k) − E 11(cid:9) = 0 , (11) for E = 0. In fact, the allowed k's can be calculated for any energy but we are interested in particular in those at zero energy corresponding to Majorana solutions. The wave number dependence on magnetic field is de- picted in Fig. 4a for a selected case. For a fixed energy E there are always eight possible wave numbers, but only FIG. 2. Spectrum of a finite length nanowire with L = 50Lso as a function of the external magnetic field magnitude ∆B. Other nanowire parameters are ∆s = 0.25Eso and µ = 0. The magnetic field angles are θ = 90◦ and φ = 15◦. Only the eight states lying closer to zero energy are displayed. Note that a zero energy Majorana mode is created at around ∆B = 0.3Eso and destroyed for values of ∆B near one unit. The vertical lines (dots) indicate the onset and destruction of the Majo- rana mode as predicted by Eqs. (8) and (13), respectively. respectively. The Fermi function F is defined as F (x; x0, s) = 1 1 + e(x−x0)/s . (5) In Eq. (2) V0 is the value of the potential outside the nanowire while θL/R and φL/R are the field angles at left and right of the magnetic interface. The potential left and right edges are centered on xL and xR and the magnetic interface is centered on xm. Their softness is controlled by the parameters sv and sm, where zero soft- ness means a steep interface and a high value implies a smooth one. The numerical results of this work are presented in special units obtained by taking , m and the Rashba spin-orbit interaction α as reference values. That is, our length and energy units are Lso = Eso = 2 αm α2m 2 , . (6) (7) III. A SINE RULE Let us consider a nanowire in a uniform magnetic field with θ = 90◦ and an arbitrary φ (see Fig. 1a). A direct diagonalization of Eq. (1) for a finite length of the wire and φ = 15◦ yields the spectrum depicted in Fig. 2 as a function of the magnetic field intensity. A main feature of this figure is the existence of a Majorana mode, lying very near zero energy, but only for a particular range of values of the magnetic field. For the parameters of the figure the Majorana mode is created around ∆B = 0.3Eso and destroyed in a rather abrupt way around ∆B = Eso. It is well known that Majorana wave functions decay to zero towards the nanowire interior. We can there- fore analyze the creation and destruction of Majoranas 4 FIG. 4. Imaginary parts of the wave numbers (only positive ones) in an infinite homogeneous nanowire with ∆s = 0.25Eso and µ = 0 as function of a) the value of the longitudinally oriented magnetic field, b) the azimuthal angle φ of a magnetic field with ∆B = 0.4Eso and polar angle θ = 90◦. Gray color is used for non degenerate modes while black is indicating degeneracy with two or more modes actually having the same Im(k). critical value of the magnetic field ∆(c) not under this quantity, thus leading to Eq. (8). Further details on the methodology can be found in Ref. 37. Here we want to use this approach to determine whether a similar condition on the field orientation, with critical values of the angles, exists or not. B = p∆2 + µ2, but Figure 4b shows the evolution of the wave numbers when increasing φ while maintaining θ = 90◦, i.e., main- taining the magnetic field in the plane formed by the nanowire direction and the effective spin orbit magnetic field direction ~Bso. This means that for φ = 0◦ the mag- netic field is aligned with the nanowire, while for φ = 90◦ it is completely perpendicular to it and parallel to ~Bso. In Fig. 4b care has been taken to choose a value of ∆B that fulfills the Majorana condition for the parallel φ = 0 orientation Eq. (8). We can see that for φ = 40.1◦ two of the complex wave numbers become real, thus destroy- ing the Majorana mode for azimuthal angles above this value. The physical behavior implied by Fig. 4b is a sud- den loss of the Majorana mode as the tilting angle φ exceeds a critical value, due to the system no longer hav- ing the required four evanescent modes with Im(k) > 0. The evanescent modes are lost because of the closing of the gap between states of opposite wave numbers FIG. 3. Band structure of the infinite homogeneous nanowire with ∆s = 0.25Eso and µ = 0 for a) parallel field (θ, φ) = (90◦, 0◦) with ∆B = 0.4Eso, b) the same as a) but on the phase transition point ∆B = 0.25Eso, c) tilted field (θ, φ) = (90◦, 38.68◦) with ∆B = 0.4Eso. those with Im(k) > 0 are displayed in Fig. 4a. In this representation the closing of the k = 0 gap in Fig. 3b cor- responds to a node of Im(k) in Fig. 4a. In order to be able to hold a Majorana a semi-infinite nanowire has to fulfil two simultaneous requirements. First, the nanowire must have four complex wave numbers with Im(k) > 0 allowed at zero energy; and second, a solution different from zero (nontrivial) must be possible for the Ck's in Eq. (10). That is, interpreting the state amplitudes Ψ(k) sσsτ as a 4×4 matrix where the four k's correspond for instance to rows and the four spin-isospin values {++, +−, −+, −−} to columns, the condition for a nontrivial solution is det{Ψ(k) sσsτ } = 0 . (12) In a parallel field this condition is fulfilled only above a (k ≈ ±2L−1 so for the particular case shown in Fig. 3c). We characterize next the dependence of the critical an- gle on ∆B and ∆s. In Fig. 5a we can see a contour plot of Im(k) as a function of φ and the ratio ∆s/∆B for an external branch wave number,33 corresponding to the lower black line of Fig. 4b. The values where Im(k) vanish separate the plot into two regions, the lower one where the Majorana is allowed and the upper (white) where no Majorana can exist. Although Eq. (11) can be solved analytically, the angles where Im(cid:0)k(φ, ∆s/∆B)(cid:1) vanishes can be obtained only numerically because φ ap- pears as argument of sine and cosine functions and no isolation is possible. As a consequence, the values of φ where the wave number first reaches zero have been found numerically and are plotted in Fig. 5c against the test function arcsin(∆s/∆B). The perfect coincidence be- tween the two results within computer precision demon- strates that a Majorana can not exist for angles such that sin φ > ∆s/∆B, provided θ = 90◦. Figure 5b shows a contour plot of Im(k) for an in- ternal branch wave number,33 corresponding to the up- per mode in Fig. 4b. In this plot the φ roots of Im(k) lie inside an upper and lower bounded region around 0.95Eso < ∆B < Eso. In fact, two of the wave num- bers become real in the white region of the contour plot. Note that this region lies in the non Majorana sector, above the transition discussed in panel a) which is now signaled by the dotted line. Theoretically the existence of this region determines two different fermionic regimes. One where a fermion mode at zero energy is constructed of plane waves with two complex and two real wave num- bers and another one made of a full set of real wave num- bers. Since we assume bound states in order to extrap- olate the results to finite systems, these cases have no relevance to us. Nevertheless the underlying causes for the existence of this region will be relevant in the study of the excited states of the finite nanowire. This will be further developed in Sec. IV. Repeating the analysis for different polar angles θ, as shown in Fig. 5c, we conclude that the angular restriction for the existence of Majoranas is ∆B sin θ sin φ < ∆s . (13) In other words, the projection of the magnetic field en- ergy parameter into the spin orbit effective magnetic field ~Bso needs to be smaller than the superconductor gap en- ergy in order to have Majoranas in a semi-infinite wire. We refer to this condition as the sine rule. Notice that Eq. (13) is not a generalization of Eq. (8), but an addi- tional law. Both Eq. (8) and Eq. (13) have to be simul- taneously met for the existence of a Majorana mode in a semi-infinite wire. In general, the sine rule Eq. (13) yields an extra bound to be considered when identifying regions of Majoranas in parameter space. For instance, assuming fixed angles (θ, φ) and varying ∆B there is a lower bound on ∆B from Eq. (8) and an upper bound from the sine rule. Analo- gously, if for a fixed ∆B the Majorana is allowed by Eq. 5 FIG. 5. a) Contour plot of Im(k) for the external branch of the nanowire propagating bands. The horizontal axis contains the ratio ∆s/∆B and the vertical one the azimuthal angle φ. The polar angle is fixed to θ = 90◦ and µ = 0. b) Contour plot of Im(k) for the internal branch of the nanowire propagating bands. The horizontal axis shows ∆B and the vertical one the azimuthal angle φ. The polar angle is fixed to θ = 90◦, ∆s = 0.8Eso and µ = 0. c) Plot of the azimuthal critical angle where Im(k) vanishes in the upper left panel as function of ∆s/∆B (points) checked against the sine rule prediction Eq. (13). Besides the θ = 90◦ case of the upper left panel, the figure also contains the comparison for other values of θ. The value of the chemical potential can be taken arbitrarily since it is irrelevant for this comparison. (8) at φ = 0◦ and and we increase φ the sine rule yields an upper bound on φ. Therefore, as explained, both equations must be met simultaneously to obtain a Majo- rana mode. Furthermore, after some parameter testing we have determined that the sine rule is not affected by the value of the chemical potential µ. This means that the overall dependence on µ for the existence of Majo- rana modes in the semi-infinite nanowire is completely covered by Eq. (8). The disappearance of the Majorana when increasing φ is not a phase transition in the sense that no imaginary part of a mode wave number crosses zero in between two regions with non null values. As shown in Fig. 4b for the polar angle θ = 90◦, above the critical φ the value of Im(k) remains stuck at zero value. The main differ- ence between the phase transition law in Eq. (8) and the sine rule Eq. (13) lies in the different type of gap closing for both cases. As shown in Fig. 3b the phase transi- 6 dently of the value of the other angle (provided Eq. (8) is fulfilled). That is, below the critical angle a projection into ~Bso is never high enough to break the Majorana. In practice, if the Majorana is allowed by Eq. (8), it will survive for any φ provided θ < θc or, alternatively, for any θ provided φ < φc. These critical angles are θc = φc = arcsin(cid:18) ∆s ∆B(cid:19) . (14) IV. EXCITED STATES While in the preceding section we focussed on the physics of the Majoranas at zero energy, comparing semi- infinite and finite nanowires, in this section we address the spectrum of excited states. The main effect of the boundary conditions is to allow only a discrete set of wave numbers instead of a continuous one. What we have done is sketch the finite nanowire spectrum by se- lecting wave numbers at regular intervals and tracking the evolution of their energy levels with an increasing angle φ. For these examples we maintain the polar angle θ = 90◦ because this is the most physically interesting configuration due to the possibility of aligning external and spin-orbit magnetic fields; nevertheless, analogous plots can be done for different values of θ. The resulting spectrum, shown in Fig. 7, explains the main features of the numerical diagonalization results of Fig. 6 for the same parameters. In principle, we could also set the boundary conditions exactly as we did in Eq. (10), but we have found this approach impossible to follow on a practical level. The resulting set of equations reads C(L) k Ψ(k) sσ sτ e−ikL/2 = 0 , Xk Xk and C(R) k C(R) k Ψ(k) sσsτ eikL/2 = 0 , (15) k where C(L) are the coefficients at the left and right nanowire ends, respectively. Basically, the resulting matrix from Eq. (15) is ill defined since it contains very large and very small matrix elements. The spectra of both panels of Fig. 7 can be divided into three different regions depending on the angle φ. First, for low values of φ there is a region where a Majo- rana mode exists and is topologically protected. In Fig. 7 the Majorana is not seen since only excited states of real wave number are shown, but we can see the corre- sponding gap. For values of φ above those determined by the sine rule the Majorana mode is destroyed and we can see a region of many level crossings. This behavior of the spectrum is explained by the gap closing of the external branches of the conduction band noticing that in the finite model only some discrete values are allowed, as sketched in Fig. 8. Finally, for higher angular values the region of zero crossings finishes and a third region arises with two possible behaviors. FIG. 6. a) Spectrum of a nanowire of length L = 50Lso with ∆s = 0.8Eso and µ = 0 as a function of φ with a fixed θ = 90◦ and ∆B = 0.9Eso. Note the spectrum change at the angle predicted by the sine rule (dotted line) as well as the spectrum collapse for values of φ close to 90◦. For values of φ above 90◦ the spectrum is given by the mirror image of the shown values. b) The same as a) but for a fixed magnetic field value ∆B = 1.1Eso. Note that for values of φ close to 90◦ now the two modes closer to zero are fermionic modes separated from each other by an energy gap. tion delimited by Eq. (8) is caused by a gap closing and reopening on a single wave number k = 0 (labeled as interior branches of the spectrum). In the language of semiconductor band structure physics we may call this the closing of a direct gap. Oppositely, the sine rule is caused by the closing of an indirect gap for k ≈ ±kf (la- beled as exterior branches of the spectrum), as shown in Fig. 3c for a selected case. We have checked these laws against the direct numer- ical diagonalization for a finite nanowire, finding a rea- sonable agreement as shown in Figs. 2 and 6. In Fig. 2 the magnetic field orientation is kept fixed to a tilted orientation while the field magnitude is changed and in Fig. 6 the magnitude is fixed while the orientation is changed. The main difference between the precise laws for the semi-infinite model and the finite system results is in the smoothness of the spectrum evolution around the transition points. While in the semi-infinite model the transition between fermionic modes to Majorana modes and vice versa happens at a single point in the parame- ter space, in the finite system we can see these transitions smoothed. This occurs due to the finite size effects, i.e., the little overlap of Majoranas on opposite ends of the nanowire. Furthermore, while Majoranas lie at exactly zero energy in the semi-infinite model, this small interac- tion makes the finite system Majoranas to have a finite small energy ǫ. A close inspection of the sine rule Eq. (13) reveals that there exist critical values for θ and φ such that if they are not surpassed a Majorana is always allowed, indepen- 7 FIG. 7. Spectrum obtained from the homogeneous nanowire band energies at selected wave numbers, displayed as tracers as a function of φ. Panel a) corresponds to the same param- eters of Fig. 6a. Note that qualitatively similar regions occur for increasing angles in both figures. In particular the spec- trum nearly collapses for values of φ close to 90◦. Panel b) shows the same as a) but for the parameters of Fig. 6b. Note that for values of φ close to 90◦ the two lower states (closer to zero energy) are fermionic modes separated by a gap as in Fig. 6b. As shown in Figs. 6 and 7 for high φ angles (near 90◦), depending on the parameters the spectrum either opens a gap or collapses near zero energy. The behavior depends on the way the internal branches of the band cross the zero energy value for those angles. The internal branches of the band can cross the zero energy level for high angles in one k > 0 point, like in Fig. 8a, thus leading (jointly with the external branch crossing point) to four real and four complex wave numbers; or, alternatively, the interior brach can cross zero energy in more than one k > 0 point, like in Fig. 8b, leading to wavefunctions characterized by eight real wave numbers. In the latter case there is a wave number range where the band spectrum lies very close to zero energy, yielding this way a collapse of the finite wire spectrum. The particular set of parameters where one or the other situation happens depends on the behavior of the internal branches of the band structure and it is not as easily predictable as the behavior of the external branches that led to the sine rule. The region of values where this collapse arises coincides with the region where the allowed solutions at zero energy are made of real wave numbers only and it was already presented in Fig. 5a for the µ = 0 case. FIG. 8. Approximate spectrum of a finite nanowire in a particular configuration of the external magnetic field. a) With only two positive real wave numbers. b) With four positive real wave numbers. V. MAGNETIC INHOMOGENEITY MODELS In this section we explore the physics of a junction of two straight nanowires with a certain angle in pres- ence of a homogeneous magnetic field parallel to one of the arms, as sketched in Fig. 1b. We assume a repre- sentation of the system as a single straight 1D nanowire containing a magnetic interface. The inhomogeneity sep- arates two homogeneous regions with different directions (but the same magnitude) of the external magnetic field. The system is solved by numerical diagonalization, as- suming a soft magnetic interface, interpreting the results by comparing with the homogeneous nanowire discussed in the preceding section. We focus on two specific effects, tilting and stretching of one of the two junction arms. A. Arm tilting The magnetic field is aligned with the left arm and the spectrum of the nanowire is computed for varying orientations of the field in the the right arm (see Fig. 9). As mentioned, this model represents under certain approximations a bent nanowire in a homogeneous mag- netic field. It was shown in Ref. 42 that bent nanowires can be approximated by 1D models with a potential well simulating the effect of the bending. Here we have only considered the magnetic field change of direction as the main inhomogeneity source, disregarding the electrical 8 FIG. 10. Majorana density function of an inhomogeneous nanowire similar to the one described in Fig. 9a. The mag- netic field azimuthal angles in the two arms are φL = 0◦ and φR = 90◦, while all along the nanowire it is θ = 90◦. Other parameters are ∆B = 0.4, ∆s = 0.25. FIG. 9. a) Nanowire spectrum for ∆s = 0.8Eso and µ = 0 with a magnetic inhomogeneity at its center as a function of the tilting angle φ. On the left side of the nanowire the magnetic field is parallel, while on the right side its angles are (θ = 90◦, φ). The magnetic field strength is constant in both sides and equal to ∆B = Eso. b) Spectrum of a nanowire in a homogeneous magnetic field with angles (θ = 90◦, φ) and with the rest of parameters as in a). potential effects of the bending. The spectrum of the inhomogeneous nanowire can be explained in terms of the homogeneous one for a tilted magnetic field. Figure 9 compares the inhomogeneous (upper) with the homogeneous (lower) nanowire spec- trum for the same set of parameters, showing that both results share the same essential features. More precisely, three φ regions can be found in both cases, but with two main differences. First, while for the homogeneous nanowire increasing φ leads to the destruction of the Ma- joranas on both ends, for the inhomogeneous nanowire only the right side Majorana is destroyed. The density of the Majorana for the inhomogeneous nanowire is shown in Figs. 10 and 11 for selected values of the parameters. As a consequence, the bent junction holds a Majorana mode (the one localized in the left side of the inhomo- geneity) independently of the magnetic angle at the right side. A second difference between upper and lower panels of Fig. 9 is that the spectra for the inhomogeneous nanowire is not symmetric with respect to φ = 90◦, in contrast with the homogeneous nanowire. A zero energy crossing local- ized in the inhomogeneity interface arises at φ = 145◦ for the selected parameters in Figs. 9a and 11. The corre- sponding bound state originates in the second excited state of the system and it is not Majorana in nature. Furthermore, this localized state is caused completely by the magnetic inhomogeneity and has no relationship with FIG. 11. Density function of the first excited state of the inhomogeneous nanowire of Fig. 9a. This state becomes lo- calized at the magnetic inhomogeneity for an azimuthal angle φ = 145◦ (and polar angle θ = 90◦). the localized states found in the bending region in Ref. 42 because we have disregarded those effects. Although we know these states are related with the magnetic inho- mogeneity, a deep understanding of their causes and the particular set of parameters leading to their enhancement or quenching is yet to be understood. B. Arm stretching We study now the behavior of the Majorana modes in the nanowire as a function of the inhomogeneity distance to the nanowire end. The magnetic field directions are fixed at (θ = 90◦, φ = 0◦) on the left end and (θ = 90◦, φ = 180◦) on the right end of the nanowire. This is a particularly interesting configuration as it is the only setup where both ends lie inside a longitudinal magnetic field, apart from the homogeneous case. This way, all the observed effects must be caused by the inhomogeneity and its distance with respect to the left nanowire end. Figure 12 shows the probability densities of the zero energy state at different positions of the magnetic inter- face with respect the left side of the nanowire. From upper to lower panels of Fig. 12 we may follow the evo- lution as the distance of the magnetic interface to the 9 FIG. 13. Spectrum of the nanowire described in the caption of Fig. 12 as a function the magnetic interface position. dle point of the nanowire both Majoranas are located at their corresponding ends. It is also worth noticing that this Majorana transfer does not imply a departure of the mode from zero energy because the Majorana on the other end of the nanowire is not affected (see Fig. 13). Additionally, the transfer phenomenon is not caused by finite size effects since we have checked that it happens for the same characteristic distance when the right end is further displaced to the right. VI. CONCLUSIONS In this work we have studied the spectra of 1D nanowires for arbitrary orientation of the magnetic field, focussing in particular on the conditions leading to a Ma- jorana mode. This study has been realized from different perspectives and methods in an effort to explain the vari- ety of observed phenomena. We have combined the com- plex band structure techniques of infinite homogeneous nanowires with numerical diagonalizations of finite sys- tems. We have demonstrated an additional condition, besides the well known topological transition law, that needs to be taken into account in order to predict the regimes of existence of Majorana modes with tilted fields. We have named this additional condition the sine rule. The sine rule predicts an upper bound on the magnetic field at which Majoranas are to be found in a 1D wire with tilted field. When the topological law is fulfilled, the sine rule leads to critical values of the field angles θc and φc, such that a Majorana mode is always found for any φ provided θ < θc or, alternatively, for any θ provided φ < φc. We have extended our analysis to nanowire junctions with an arbitrary angle, modeled as magnetically inho- mogeneous nanowires, explaining most of their proper- ties in terms of the behavior of its homogeneous parts. We have focussed, particularly, on the role of tilting and stretching of one of the junction arms. We also reported the existence of a bound non Majorana state located on the magnetic inhomogeneity. Finally, we have studied the Majorana transfer phenomenon as the distance of the magnetic inhomogeneity to the nanowire end is in- FIG. 12. Density distributions of the Majorana mode in a finite nanowire with a magnetic inhomogeneity. We used ∆s = 0.25Eso, µ = 0 and a magnetic field of magnitude ∆B = 0.4Eso oriented parallel in the left side of the magnetic interface and antiparallel in the right side. In each panel the potential well and the position of the magnetic interface are shown. The latter corresponds to a Fermy-type function whose position shifts to the right following the sequence from upper to lower panels. left end of the system is increased. Most remarkably, for short distance the left Majorana is not peaked on the left end, but remains stuck on the magnetic interface (upper panels). If the distance is increased, however, the Ma- jorana is eventually transferred to the left nanowire end after some critical distance (lower panels). This trans- fer is seen as a smooth decrease of the density maximum at the magnetic interface accompanied by an increase at the left end. Finally, when the interface is on the mid- creased. Testing these predictions would require exper- iments of nanowires in inhomogeneous magnetic fields. Alternatively, it has been suggested in this work that a bent nanowire in a homogeneous field should display sim- ilar phenomena, while being more feasible in practice. As an interesting continuation of this work we are presently analyzing the validity of the sine rule in higher dimen- sional nanowires, where the transverse degrees of freedom require a multimode description of the electronic states. ACKNOWLEDGMENTS 10 This work was funded by MINECO-Spain (grant FIS2011-23526), CAIB-Spain (Conselleria d'Educaci´o, Cultura i Universitats) and FEDER. We hereby acknowl- edge the PhD grant provided by the University of the Balearic Islands. ∗ [email protected] 1 A. Y. Kitaev, Annals Phys. 303, 2 (2003). 2 I. Affleck, T. Kennedy, E. H. Lieb, and H. Tasaki, Phys. Rev. Lett. 59, 799 (1987). 3 F. Wilceck, Nature Phys. 5, 614 (2009). 4 X. L. Qi and S. C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). 5 J. Alicea, Rep. Prog. Phys. 75, 076501 (2012). 6 M. Leijnse and K. Flensberg, Semicond. Sci. Technol. 27, 124003 (2012). Lett. 105, 077001 (2010). 23 Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). 24 T. D. Stanescu, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. B 84, 144522 (2011). 25 K. Flensberg, Phys. Rev. B 82, 180516 (2010). 26 A. C. Potter and P. A. Lee, Phys. Rev. Lett. 105, 227003 (2010). 27 A. C. Potter and P. A. Lee, Phys. Rev. B 83, 094525 7 C. W. J. Beenakker, Annu. Rev. Condens. Matter Phys. (2011). 4, 113 (2013). 28 S. Gangadharaiah, B. Braunecker, P. Simon, and D. Loss, 8 T. D. Stanescu and S. Tewari, J. Phys. Condens. Matter Phys. Rev. Lett. 107, 036801 (2011). 25, 233201 (2013). 9 M. Franz, Nature Nanotechnology 8, 149 (2013). 10 L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). 11 A. R. Akhmerov, J. Nilsson, and C. W. J. Beenakker, 29 R. Egger, A. Zazunov, and A. L. Yeyati, Phys. Rev. Lett. 105, 136403 (2010). 30 A. Zazunov, A. L. Yeyati, and R. Egger, Phys. Rev. B 84, 165440 (2011). Phys. Rev. Lett. 102, 216404 (2009). 31 E. Prada, P. San-Jos´e, and R. Aguado, Phys. Rev. B 86, 12 Y. Tanaka, T. Yokoyama, and N. Nagaosa, Phys. Rev. 180503(R) (2012). Lett. 103, 107002 (2009). 32 J. Klinovaja, S. Gangadharaiah, and D. Loss, Phys. Rev. 13 K. T. Law, P. A. Lee, and T. K. Ng, Phys. Rev. Lett. 103, Lett. 108, 196804 (2012). 237001 (2009). 14 C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das 33 J. Klinovaja and D. Loss, Phys. Rev. B 86, 085408 (2012). 34 J. S. Lim, L. Serra, R. Lopez, and R. Aguado, Phys. Rev. Sarma, Rev. Mod. Phys. 80, 1083 (2008). B 86, 121103 (2012). 15 J. K. Pachos, Introduction to topological Quantum Compu- 35 J. S. Lim, R. Lopez, and L. Serra, New J. Phys. 14, 083020 tation (Cambridge University Press, 2012). 16 E. Majorana, Nuovo Cimento 14, 171 (1937). 17 V. Mourik, K. Zuo, S. Frolov, S. Plissard, E. Bakkers, and L. Kouwenhoven, Science 336, 1003 (2012). 18 M. T. Deng, C. L. Yu, G. Y. Huan, M. Larsson, and P. Caroff, Nano Lett. 12, 6414 (2012). (2012). 36 J. S. Lim, R. Lopez, and L. Serra, Europhys. Lett. 103, 37004 (2013). 37 L. Serra, Phys. Rev. B 87, 075440 (2013). 38 J. Osca and L. Serra, Phys. Rev. B 88, 144512 (2013). 39 J. Alicea, G. Refael, F. von Oppen, and M. P. A. Fisher, 19 L. P. Rokhinson, X. Liu, and J. K. Furdyna, Nature Nature Phys. 7, 412 (2011). Physics 8, 795 (2012). 20 A. Das, Y. Ronen, Y. Most, Y. Oreg, M. Heiblum, and H. Shtrikman, Nature Physics 8, 887 (2012). 21 A. D. K. Finck, D. J. Van Harlingen, P. K. Mohseni, K. Jung, and X. Li, Phys. Rev. Lett. 110, 126406 (2013). 22 R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. 40 J. D. J. Clarke, Phys. Rev. B 84, 094505 (2011). Sau, D. and S. Tewari, 41 B. I. Halperin, Y. Oreg, A. Stern, G. Refael, J. Alicea, and F. von Oppen, Phys. Rev. B 85, 144501 (2012). 42 H. Wu, D. W. L. Sprung, and J. Martorell, Phys. Rev. B 45, 11960 (1992). 

1803.00376

1

1803

Orbital, spin and valley contributions to Zeeman splitting of excitonic resonances in MoSe$_2$, WSe$_2$ and WS$_2$ monolayers

We present a comprehensive optical study of the excitonic Zeeman effects in transition metal dichalcogenide monolayers, which are discussed comparatively for selected materials: MoSe$_2$, WSe$_2$ and WS$_2$. We introduce a simple semi-phenomenological description of the magnetic field evolution of individual electronic states in fundamental sub-bands by considering three additive components: valley, spin and orbital terms. We corroborate the validity of the proposed description by inspecting the Zeeman-like splitting of neutral and charged excitonic resonances in absorption-type spectra. The values of all three terms are estimated based on the experimental data, demonstrating the significance of the valley term for a consistent description of magnetic field evolution of optical resonances, particularly those corresponding to charged states. The established model is further exploited for discussion of magneto-luminescence data. We propose an interpretation of the observed large g-factor values of low energy emission lines, due to so-called bound/localized excitons in tungsten based compounds, based on the brightening mechanisms of dark excitonic states.

cond-mat.mes-hall

cond-mat

Orbital, spin and valley contributions to Zeeman splitting of excitonic resonances in MoSe2, WSe2 and WS2 monolayers M. Koperski1,2,3*, M. R. Molas1,2, A. Arora1, K. Nogajewski1, M. Bartos1, J. Wyzula1,4, D. Vaclavkova1,4, P. Kossacki2, M. Potemski1,2 1 Laboratoire National des Champs Magnétiques Intenses, CNRS-UJF-UPS-INSA, Grenoble, 2 Institute of Experimental Physics, Faculty of Physics, University of Warsaw, Warsaw, Poland 3School of Physics and Astronomy, The University of Manchester, Manchester M13 9PL, UK 4Department of Experimental Physics, Faculty of Science, Palacký University, Olomouc, Czech France Republic *[email protected] ABSTRACT We present a comprehensive optical study of the excitonic Zeeman effects in transition metal dichalcogenide monolayers, which are discussed comparatively for selected materials: MoSe2, WSe2 and WS2. We introduce a simple semi-phenomenological description of the magnetic field evolution of individual electronic states in fundamental sub-bands by considering three additive components: valley, spin and orbital terms. We corroborate the validity of the proposed description by inspecting the Zeeman-like splitting of neutral and charged excitonic resonances in absorption-type spectra. The values of all three terms are estimated based on the experimental data, demonstrating the significance of the valley term for a consistent description of magnetic field evolution of optical resonances, particularly those corresponding to charged states. The established model is further exploited for discussion of magneto-luminescence data. We propose an interpretation of the observed large g-factor values of low energy emission lines, due to so- called bound/localized excitons in tungsten based compounds, based on the brightening mechanisms of dark excitonic states. Subject Areas: Condensed Matter Physics, Nanophysics, Semiconductor Physics Magneto-optical response of semiconducting transition metal dichalcogenide (sc-TMD) monolayers is commonly investigated to probe the coupling strength of effective out-of-plane angular momentum of carriers in K-valleys with external magnetic field in the limit of 2D confinement. The direct band gap character of sc-TMD monolayers [1-5] allows for a comparative study of absorption-type (reflectance and transmission) and photoluminescence processes, both related to the K-valley states. The most obvious manifestation of magnetic coupling is revealed by Zeeman-like splitting of excitonic resonances observed when the magnetic field is applied perpendicularly to the surface of the structure [6-13]. The magnitude of the splitting is a quantitative measure of magnetic-field-induced energy shifts of conduction and valence states involved in the optical transitions. As such, the Zeeman-like effect for optical resonances may act as a validity test for theoretical predictions and basic understanding of magnetic field impact on the energy of electronic states in fundamental, K-point, sub-bands [14-23]. As for now, the accuracy of developed models is rather poor, so that the explanation of even the simplest observations remains disputable. Especially elusive is the role of so-called valley term, which is responsible for band-type effects sensitive to magnetic field. Here, we aim to deepen the understanding of magneto-optical properties of sc-TMD monolayers by introducing a phenomenological description of linear-with-magnetic-field evolution of conduction and valence states involved in experimentally identified optical transitions. This simple model includes valley, spin and orbital terms treated as additive components. Three parameters related to these contributions can be estimated by comparatively analyzing polarization- resolved magneto-reflectance spectra of different representatives of sc-TMD family (MoSe2, WSe2 and WS2 are considered in this work). Particularly, the examination of the resonances due to negatively charged excitons provides information on magnetic field splitting of an individual state in the conduction band, which unambiguously demonstrates the significance of the valley term in the interpretation of experimental data and yields an estimation of its value. Further consequences arising from the introduced model are revealed by the analysis of magneto-luminescence spectra, which unveil large g-factor values of lines corresponding to bound/localized excitons combined sometimes with peculiar polarization properties. The band-edge reflectivity spectra of sc-TMD monolayers are dominated by two robust resonances, known as A and B, which arise due to large spin-orbit splitting [24] strongly pronounced in the valence band at the K-point. Both resonances are related to free neutral exciton states and are classified according to the valence sub-band involved in the optical transition [25-32]. The lower/higher energy excitons A/B are related to transitions involving a hole from the upper/lower valence band state, respectively. Representative reflectance spectra of MoSe2, WSe2 and WS2 monolayers, measured with circular polarization resolution as a function of the external out-of-plane is used to probe optical transition in K+/K- valleys, providing a remarkably simple sensitivity to valley degree of freedom [33-35]. Upon application of an out-of-plane magnetic field, the splitting between magnetic field, are presented in Fig. 1. In sc-TMD monolayers, the (cid:1)(cid:2)/(cid:1)(cid:4) circular polarization of light (cid:1)(cid:2)/(cid:1)(cid:4) components (defined as (cid:5)(cid:6)(cid:7)−(cid:5)(cid:6)(cid:9) =(cid:11)(cid:12)(cid:13)(cid:14)) appears to be equal for A and B excitons for all close to (cid:11)≈−4. Assuming that Zeeman splitting of an excitonic transition reflects the relative Zeeman shifts of corresponding single-particle subbands, the experimental observation, (cid:11)(cid:17)=(cid:11)(cid:13)(≈ −4), represents one perceptible condition to be fulfilled by Zeeman shifts of individual electronic reversal symmetry, the (cid:11)(cid:17)=(cid:11)(cid:13) condition leads to a 3-term parametrization of linear-with-field having a simple physical meaning. Consequently, we introduce the valley ((cid:5)(cid:21)), spin ((cid:5)(cid:22)) and orbital ((cid:5)(cid:23)(cid:24)) terms which contribute to the energies of spin-orbit-split conduction and valence subbands in contributions to the energy of fundamental conduction and valence subbands. Although different possible sets of these 3 parameters can be formally introduced, we follow here a common wisdom and assume that Zeeman shifts of all electronic subbands are expressed with 3 additive terms, each three materials within the experimental accuracy and the resulting g-factor values are found to be states. When combined with symmetric evolution of K+/K- states imposed by preservation of time- the following way: (cid:5)(cid:29),↑ ±=±(cid:5)(cid:21)+(cid:5)(cid:22) =(±(cid:11)(cid:21)+(cid:11)(cid:22))(cid:12)(cid:13)(cid:14) (cid:5)(cid:29),↓ ±=±(cid:5)(cid:21)−(cid:5)(cid:22) =(±(cid:11)(cid:21)−(cid:11)(cid:22))(cid:12)(cid:13)(cid:14) (cid:5)$,↑ ±=±(cid:5)(cid:21)+(cid:5)(cid:22)±(cid:5)(cid:23)(cid:24) =%±(cid:11)(cid:21)+(cid:11)(cid:22)±(cid:11)(cid:23)(cid:24)&(cid:12)(cid:13)(cid:14) (cid:5)$,↓ ±=±(cid:5)(cid:21)−(cid:5)(cid:22) ± (cid:5)(cid:23)(cid:24) =%±(cid:11)(cid:21)−(cid:11)(cid:22)±(cid:11)(cid:23)(cid:24)&(cid:12)(cid:13)(cid:14) ' (cid:26)(cid:28) (cid:25)(cid:26)(cid:27) where (cid:5)(cid:29)($),↑(↓) ± is the energy of conduction (c) or valence (v), spin-up (› ) or spin-down (fl ) subband in K+ or K- valley, m B is Bohr magneton, B stands for the strength of the magnetic field, and gV, gS, gd2 are g-factors corresponding to EV, ES and Ed2 terms, respectively. A pictorial representation of how each term independently affects the energy of states in the conduction band (CB) and valence band (VB) in K+/K- valleys is illustrated in Fig. 2. It is important to note, that there exist two possible alignments of spin in the conduction spin-orbit-split subbands. The energy distance between A and B excitons is mostly defined by a large separation of spin-orbit- split subbands in the valence band whereas a considerable smaller spin-orbit effects in the conduction band cannot be directly inferred from simple absorption/reflectance spectra. Information whether the optically bright A exciton is associated with lower or upper conduction band subband have been however supplied by extensive studies of the photoluminescence response of different sc-TMD monolayers and under various conditions (such as analysis of the spectral response to the in-plane magnetic field and including studies involving unusual configuration of the emitted light with respect to the monolayer plane). It is now rather clear that MoSe2 monolayer is a material with bright ground exciton state [36,37], contrary to WSe2 and WS2 for which the ground exciton state is dark [37-40]. The relevant consequence of the existence of two types of materials, tentatively named as 'bright' and 'darkish', is a significant difference in g-factor value for the lower energy conduction band state. The spin and valley terms add up in 'bright' materials and compensate each other in 'darkish' materials, leading to larger magnetic-field-induced splitting of lower conduction state for 'bright' than 'darkish' materials. In order to investigate the validity of this prediction, we will analyze the magneto- reflectance spectra focusing on charged exciton resonance (CX) (assuming it is a negatively charged state1), which appears as a weaker feature below the energy of a free neutral exciton A [42]. Typical reflectivity spectra showing CX resonance for MoSe2, WSe2 and WS2 monolayers are presented in Fig. 3. We will focus on the interpretation of data for MoSe2 and WS2 materials, considering them as representatives of 'bright' and 'darkish' family. The CX resonance in WSe2 monolayer is very weak and broad in our samples, making it difficult to analyze and interpret the data unambiguously. In the absorption process leading to creation of CX, there is a single electron in the initial state, which at low temperature occupies the lower energy conduction band state. In such case, thermal redistribution of the occupation of this individual electronic state leads to polarization of the CX absorption resonance at higher magnetic fields. The rate of the polarization degree increase is directly indicative of the g-factor value corresponding to this individual state. In Boltzmann approximation to Fermi-Dirac statistics, which is valid if the thermal energy is larger than Fermi energy (()≫(cid:5)+), i.e., when the electron concentration is sufficiently small and effective mass is large, what we expect in our remotely doped monolayers, the polarization degree in external magnetic field is given by a formula: -(cid:6)(cid:7)+-(cid:6)(cid:9),=./0ℎ2(cid:11)(cid:12)(cid:13)(cid:14)() 3= ./0ℎ4((cid:11)(cid:22)±(cid:11)(cid:21))(cid:12)(cid:13)(cid:14) ,-(cid:6)(cid:7)−-(cid:6)(cid:9) 5 where -(cid:6)(cid:7)/-(cid:6)(cid:9) is the absorption strength of CX resonance detected in s +/s polarization, (cid:11)=(cid:11)(cid:22)± (cid:11)(cid:21) is the g-factor of lower energy conduction band state and might be expressed by a sum or a difference of spin and valley g-factors for bright or dark materials respectively, m B is Bohr magneton, B is the value of magnetic field, k is Boltzmann constant and T is the effective temperature of the electron gas. In case of MoSe2 monolayer the application of this formula is straightforward, as the CX resonance simply splits with a g-factor (cid:11)67=−4.2±0.2 (same value as for neutral exciton resonance (cid:11)7 =(cid:11)67) and the polarization degree clearly increases with raising magnetic field. Consequently, one can directly estimate the value (cid:11)/). We will assume that the temperature of the state of MoSe2 (cid:11)=1.84. The evolution of the CX resonance in WS2 appears to be much more electron gas is equal to 10 K, taking into account weak heating effects of white light illumination of the sample kept in ~ 4.2 K exchange gas. Then we obtain a g-factor value for lower energy conduction complicated. The energy dependence on magnetic field for s +/s components is apparently non- linear (see Fig. 3(f)). This peculiar evolution originates, as discussed previously in literature [43-48] , from contribution of two different CX states to the feature observed in absorption-type spectra. The existence of two CX states in 'darkish' materials is a consequence of spin alignment in conduction sub-bands, which allows an excess electron from either K+ or K- valley to accompany the photo- 1 This assumption is justified by rather large energy distances of charged to neutral exciton observed in our samples. Experiments on gated monolayers [41] show binding energies of negatively charged excitons to be considerably larger those of positively charged excitons. () - - created electron-hole pair and form a bound CX state. As a result, the final configuration of two conduction electrons may constitute a singlet (inter-valley) or triplet (intra-valley) state. In 'bright' materials only a singlet state may be expected to be bound due to significant spin-spin (exchange) interaction strength, when two parallel spin states occupy the same valley (hence are characterized by the same pseudo-spin quantum number) for triplet state in bright materials. Therefore, the absence of triplet CX state can be seen as a consequence of Pauli blocking mechanism. The singlet and triplet configurations of conduction electrons in 'bright' and 'darkish' materials are presented in Fig. 4. The magnetic field evolution of the CX state in 'darkish' WS2 may be accounted for if two and zero-field splitting of 5 meV are considered [47], in combination with polarization effects imposed by the thermal redistribution of population of the excess electron state. For the singlet state transitions exhibiting the same g-factor value ((cid:11)67=−3.8; same as for the neutral exciton) value of the charged exciton, the higher energy (cid:1)(cid:4) component accumulates the majority of oscillator strength at higher fields, contrary to the triplet state, for which the lower energy (cid:1)(cid:2) component gets enhanced. Overall, due to small zero field splitting with respect to the linewidths of transitions (~ 20 meV) the CX features in 'darkish' WS2 gives impression of a single resonance, but upon closer inspection it becomes clear that both singlet and triplet CX states are involved and careful analysis allows estimation of the oscillator strength of (cid:1)(cid:2)/(cid:1)(cid:4) components (see Fig. 3(g)) and eventually factor (cid:11)=1.08. establish g-factor of the lower energy conduction electron by describing the increase of polarization degree with aforementioned Boltzmann-type formula (see Fig. 3(h)). As a result, continuing the assumption of 10 K temperature of electron gas, we obtain the value of the lower conduction state g- In order to translate the parameters extracted from experiments into the spin and valley terms appearing in our model, we will need to make certain assumptions. One can assume, for instance, that the spin and valley g-factors are equal for MoSe2 and WS2 monolayers, which will lead to a simple linear equation: (cid:11)(cid:22)−(cid:11)(cid:21) =1.08'→>(cid:11)(cid:22)=0.38 >(cid:11)(cid:22)+(cid:11)(cid:21) =1.84 (cid:11)(cid:21) =1.46' An alternative method is to assume that the value of the spin term is the same as for a free electron in vacuum ((cid:11)(cid:22)=1, in our convention) and then obtain the values of the valley terms for 'bright' MoSe2 and 'darkish' WS2 monolayers, >(cid:11)(cid:21) =2.08, for 'bright' MoSe2 monolayers (cid:11)(cid:21) =0.84, for 'darkish' WS2 monolayers' The results of both treatments of experimental data are summarized in Tab. 1. Although these estimations are very rough, they quite certainly indicate that the valley term is a substantial part of a cohesive description of the magneto-optical properties of TMDC monolayers and cannot be neglected. Our description of the magnetic field evolution of the electronic states may be further exploited to shed more light onto the magneto-luminescence spectra. As demonstrated in Fig. 5, the PL spectra of MoSe2 monolayers are rather simple, showing neutral and charged exciton lines (X and CX), which split with g-factors values mimicking those observed in absorption-type spectra. Both resonances are polarized in higher fields so that the lower energy ((cid:1)(cid:2)) component gains in intensity, most likely due to thermal distributions of excitons between field-split states. The 'darkish' WS2 and WSe2 monolayers exhibit much more complicated PL response. The neutral and charged exciton lines (X and CX) are accompanied by a lower energy multi-peak PL band. It has been argued that the lines forming this band originate, perhaps partially, from recombination of dark exciton states via different brightening mechanisms [37,49,50]. A rather outstanding fingerprint of these additional lines is that they exhibit large g-factor values. Tab. 2 presents the g-factor values for all distinguishable lines in PL spectra of MoSe2, WSe2 and WS2 monolayers. Notably, the lines forming low energy band in WSe2 and WS2 exhibit g-factors from 4 up to 13.5 in absolute value, in most cases significantly larger than g-factors of free excitonic resonances. The origin of enhanced magnetic field splitting of lower energy PL lines in 'darkish' materials remains unclear. However, our description of magnetic effects indicates that it is plausible that larger g-factors are related to recombination of dark exciton states. Conceptually, dark excitons may be brightened due to mechanisms mixing states from K+/K- valleys hence allowing intra-valley recombination or one could consider inter-valley recombination mediated e. g. by phonons to preserve carrier momentum or defect states which are known to be particularly pronounced in 'darkish' TMDc mono- and multilayers [51-54]. Both types of processes are illustrated in Fig. 6. Employing our estimation of parameters, we can estimate the g-factor values of such optical transitions in a following way: V(cid:11)WXYZ[ =2%(cid:11)(cid:23)(cid:24)+2(cid:11)(cid:21)&≈10 (cid:11)WXY[\ =2%(cid:11)(cid:23)(cid:24)+2(cid:11)(cid:22)&≈8' assuming the following values of our parameters: (cid:11)(cid:23)(cid:24) ≈2,(cid:11)(cid:21) ≈1.5,(cid:11)(cid:22) ≈1. Notably large g-factors (about -8) have been recently reported for intra valley dark excitons in WSe2 monolayer encapsulated in hBN [55]. Our simple consideration takes into account recombination of neutral complexes. More complicated transitions could be realised with charged excitons, involving for instance shake-up process with elevation of an excess electron from lower to upper conduction band state. That could potentially lead to peculiar polarisation properties, when higher energy line gains in intensity at higher fields, as is observed, e. g. for lines A1 in WSe2 monolayer and A2 in WS2 monolayer (see Fig. 5). In summary, we have presented a collection of magneto-optical spectra, based on reflectance and PL measurements, of three sc-TMD compounds: MoSe2, WSe2 and WS2. This data helped us to develop a simple model of three additive terms (valley, spin and orbital contributions), which we used to describe the linear-with-magnetic-field (Zeeman) energy shifts of electronic states in fundamental conduction and valence sub-bands. We have demonstrated the significance of the valley term by analysing the magnetic-field-induced increase of polarisation degree of CX state in absorption-type spectra. Particularly, we have observed a clear signature of singlet and triplet CX states in 'darkish' WS2 monolayers. The values of the g-factors of lower energy conduction band state obtained for MoSe2 and WS2 materials allowed us to estimate the value of the valley term and therefore have a tentative description of the evolution of all the electronic states involved in fundamental optical transitions. As a result of this analysis, we have proposed a reasoning, which sheds light onto the expatiation of large g-factor values for lower energy PL lines in 'darkish' monolayers by attribution them to dark states, which become partially allowed through various brightening mechanisms. Acknowledgements We kindly thank J. Marcus for providing a piece of WSe2 crystal used and acknowledge helpful discussions with A. Slobodeniuk and D. Basko. The authors acknowledge the support from the European Research Council (MOMB project No. 320590), the EC Graphene Flagship project (No. 604391) and the ATOMOPTO project (TEAM programme of the Foundation for Polish Science co- financed by the EU within the ERDFund). The support of LNCMI-CNRS, a member of the European Magnetic Field Laboratory is also acknowledged. The authors of this manuscript declare no competing financial interest.  APPENDIX: MATERIALS AND METHODS 1. Sample preparation by mechanical exfoliation Monolayer MoSe2, WSe2 and WS2 flakes were obtained by mechanical exfoliation (polydimethylsiloxane-based technique [56]) of bulk crystals in 2H phase. Crystals from different sources were used including commercial suppliers (HQ Graphene). Exfoliated flakes were deposited on Si/SiO2 substrates and initially inspected under an optical microscope to determine the thickness of flakes based on their optical contrast. Eventually, the characteristic PL response from measured flakes unambiguously confirms their monolayer thickness. 2. Experimental setup and methods The magneto-optical experiments were done in high magnetic field facility in Grenoble. Specially designed probes were used with resistive magnets (supplying magnetic field up to 30 T with 50 mm Bohr radius) at low temperatures (via helium exchange gas) to measure reflected or emitted light from the samples. The fiber based set-up allowed focalization of the incoming light to a spot of a about ten micrometers in size. Piezo-positioners were used for x-y-z movement of the sample. A halogen lamp provided illumination for reflectance measurements and 514.5 nm Ar+ laser line for photoluminescence investigations. A quarter-waveplate followed by a polarizer was mounted in a fixed position before the detection fiber entrance and (cid:1)(cid:2)/(cid:1)(cid:4) helicity was determined by the magnetic field polarity. The reflectivity contrast was obtained by measuring the spectra on two locations: on the flake and on the nearby Si/SiO2 substrate region, then calculated as: ^((cid:5))=^_'\aZ((cid:5))−^bcdbY[\YZ((cid:5)) ^_'\aZ((cid:5))+^bcdbY[\YZ((cid:5)) The reflectivity spectra were reproduced by using a transfer matrix method including Lorentzian contributions to account for resonances observed in the experimental spectra. The data were fitted with such phenomenological curves to establish the energy and oscillator strength of transitions separately in spectra measured in (cid:1)(cid:2)/(cid:1)(cid:4) polarizations. REFERENCES [1] K. F. Mak, C. Lee, J. Hone, J. Shan, T. F. Heinz, Atomically Thin MoS2: A New Direct-Gap Semiconductor, Phys. Rev. Lett. 105, 136805 (2010). [2] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C. Y. Chim, G. Galli, F. Wang, Emerging photoluminescence in monolayer MoS2, Nano. Lett. 10 (4), 1271–1275 (2010). [3] W. Zhao, Z. Ghorannevis, L. Chu, M. Toh, C. Kloc, P. H. Tan, G. Eda, Evolution of electronic structure in atomically thin sheets of WS2 and WSe2, ACS Nano 7 (1), 791–797 (2013). [4] Y. Zhang, T.-R. Chang, B. Zhou, Y.-T. Cui, H. Yan, Z. Liu, F. Schmitt, J. Lee, R. Moore, Y. Chen, H. Lin, H. T. Jeng, S.-K. Mo, Z. Hussain, A. Bansil, Z.-X. Shen, Direct observation of the transition from indirect to direct bandgap in atomically thin epitaxial MoSe2, Nature Nanotechnology 9, 111–115 (2014). [5] I. G. Lezama, A. Arora, A . Ubaldini, C. Barreteau, E. Giannini, M. Potemski, A. F. Morpurgo, Indirect-to-Direct Band Gap Crossover in Few-Layer MoTe2, Nano. Lett. 15 (4), 2336–2342 (2015). [6] Y. Li, J. Ludwig, T. Low, A. Chernikov, X. Cui, G. Arefe, Y. D. Kim, A. M. van der Zande, A. Rigosi, H. M. Hill, S. H. Kim, J. Hone, Z. Li, D. Smirnov, T. F. Heinz, Valley Splitting and Polarization by the Zeeman Effect in Monolayer MoSe2, Phys. Rev. Lett. 113, 266804 (2014). [7] A. Srivastava, M. Sidler, A. V. Allain, D. S. Lembke, A. Kis, A. Imamoglu, Valley Zeeman effect in elementary optical excitations of monolayer WSe2, Nature Physics 11, 141–147 (2015). [8] G. Wang, L. Bouet, M. M. Glazov, T. Amand, E. L. Ivchenko, E. Palleau, X. Marie, B. Urbaszek, Magneto-optics in transition metal diselenide monolayers, 2D Materials 2, 034002 (2015). [9] G. Aivazian, Z. Gong, A. M. Jones, R.-L. Chu, J. Yan, D. G. Mandrus, C. Zhang, D. Cobden, W. Yao, X. Xu, Magnetic control of valley pseudospin in monolayer WSe2, Nature Physics 11, 148 (2015). [10] D. MacNeill, C. Heikes, K. F. Mak, Z. Anderson, A. Kormányos, V. Zólyomi, J. Park, D. C. Ralph, Breaking of Valley Degeneracy by Magnetic Field in Monolayer MoSe2, Phys. Rev. Lett. 114, 037401 (2015). [11] A. A. Mitioglu, P. Plochocka, Á. Granados del Aguila, P. C. M. Christianen, G. Deligeorgis, S. Anghel, L. Kulyuk, and D. K. Maude, Optical Investigation of Monolayer and Bulk Tungsten Diselenide (WSe2) in High Magnetic Fields, Nano Lett. 15 (7), 4387–4392 (2015). [12] A. Arora, R. Schmidt, R. Schneider, M. R. Molas, I. Breslavetz, M. Potemski, R. Bratschitsch, Valley Zeeman Splitting and Valley Polarization of Neutral and Charged Excitons in Monolayer MoTe2 at High Magnetic Fields, Nano. Lett. 16 (6), 3624–3629 (2016). [13] A. V. Stier, K. M. McCreary, B. T. Jonker, J. Kono, S. A. Crooker, Exciton diamagnetic shifts and valley Zeeman effects in monolayer WS2 and MoS2 to 65 Tesla, Nature Communications 7, 10643 (2016). [14] G.-B. Liu, D. Xiao, Y. Yao, X. Xu, W. Yao, Electronic structures and theoretical modelling of two- dimensional group-VIB transition metal dichalcogenides, Chem. Soc. Rev. 44, 2643-2663 (2015). [15] Z. Y. Zhu, Y. C. Cheng, U. Schwingenschlögl, Giant spin-orbit-induced spin splitting in two- dimensional transition-metal dichalcogenide semiconductors, Phys. Rev. B 84, 153402 (2011). [16] A. Ramasubramaniam, Large excitonic effects in monolayers of molybdenum and tungsten dichalcogenides, Phys Rev B 86, 115409 (2012). [17] D. Xiao, G.-B. Liu, W. Feng, X. Xu, W. Yao, Coupled Spin and Valley Physics in Monolayers of MoS2 and Other Group-VI Dichalcogenides, Phys. Rev. Lett. 108, 196802 (2012). [18] A. Kumar, P. K. Ahluwalia, Electronic structure of transition metal dichalcogenides monolayers 1H-MX2 (M = Mo, W; X = S, Se, Te) from ab-initio theory: new direct band gap semiconductors, Eur. Phys. J. B 85, 186 (2012). [19] T. Cheiwchanchamnangij, W. R. L. Lambrecht, Quasiparticle band structure calculation of monolayer, bilayer, and bulk MoS2, Phys Rev B 85, 205302 (2012). [20] G.-B. Liu, W.-Y. Shan, Y. Yao, W. Yao, D. Xiao, Three-band tight-binding model for monolayers of group-VIB transition metal dichalcogenides, Phys Rev B 88, 85433 (2013).  [21] K. Kośmider, J. W. González, J. Fernández-Rossier, Large spin splitting in the conduction band of transition metal dichalcogenide monolayers, Phys Rev B 88, 245436 (2013). [22] A. Molina-Sánchez, D. Sangalli, K. Hummer, A. Marini, L. Wirtz, Effect of spin-orbit interaction on the optical spectra of single-layer, double-layer, and bulk MoS2, Phys Rev B 88, 45412 (2013). [23] A. Kormányos, G. Burkard, M. Gmitra, J. Fabian, V. Zólyomi, N. D. Drummond, V. Fal'ko, k · p theory for two-dimensional transition metal dichalcogenide semiconductors, 2D Mater. 2, 22001 (2015). [24] J. M. Riley, F. Mazzola, M. Dendzik, M. Michiardi, T. Takayama, L. Bawden, C. Granerød, M. Leandersson, T. Balasubramanian, M. Hoesch, T. K. Kim, H. Takagi, W. Meevasana, Ph. Hofmann, M. S. Bahramy, J. W. Wells, P. D. C. King, Direct observation of spin-polarized bulk bands in an inversion-symmetric semiconductor, Nat. Phys. 10, 835-839 (2014). [25] J. A. Wilson, A. D. Yoffe, The transition metal dichalcogenides discussion and interpretation of the observed optical, electrical and structural properties, Adv. Phys. 18, 193–335 (1969). [26] M. Koperski, M. R. Molas, A. Arora, K. Nogajewski, A. O. Slobodeniuk, C. Faugeras, M. Potemski, Optical properties of atomically thin transition metal dichalcogenides: observations and puzzles, Nanophotonics 6, 1289 (2017). [27] M. M. Ugeda, A. J. Bradley, S.-F. Shi, F. H. da Jornada, Y. Zhang, D. Y. Qiu, W. Ruan, S.-K. Mo, Z. Hussain, Z.-X. Shen, F. Wang, S. G. Louie, M. F. Crommie, Giant bandgap renormalization and excitonic effects in a monolayer transition metal dichalcogenide semiconductor, Nature Materials 13, 1091–1095 (2014). [28] A. Chernikov, T. C. Berkelbach, H. M. Hill, A. Rigosi, Y. Li, O. B. Aslan, D. R. Reichman, M. S. Hybertsen, T. F. Heinz, Exciton Binding Energy and Nonhydrogenic Rydberg Series in Monolayer WS2, Phys. Rev. Lett. 113, 076802 (2014). [29] K. He, N. Kumar, L. Zhao, Z. Wang, K. F. Mak, H. Zhao, J. Shan, Tightly Bound Excitons in Monolayer WSe2, Phys. Rev. Lett. 113, 026803 (2014). [30] B. Zhu, X. Chen, X. Cui, Exciton Binding Energy of Monolayer WS2, Sci. Rep. 5, 9218 (2015). [31] H. M. Hill, A. F. Rigosi, C. Roquelet, A. Chernikov, T. C. Berkelbach, D. R. Reichman, M. S. Hybertsen, L. E. Brus, T. F. Heinz, Observation of Excitonic Rydberg States in Monolayer MoS2 and WS2 by Photoluminescence Excitation Spectroscopy, Nano Lett 15 (5), 2992–2997 (2015). [32] A. Chernikov, A. M. van der Zande, H. M. Hill, A. F. Rigosi, A. Velauthapillai, J. Hone, T. F. Heinz, Electrical Tuning of Exciton Binding Energies in Monolayer WS2, Phys Rev Lett 115, 126802 (2015). [33] T. Cao, G. Wang, W. Han, H. Ye, C. Zhu, J. Shi, Q. Niu, P. Tan, E. Wang, B. Liu, J. Feng, Valley- selective circular dichroism of monolayer molybdenum disulphide, Nature Commun. 3, 887 (2012). [34] K. F. Mak, K. He, J. Shan, T. F. Heinz, Control of valley polarization in monolayer MoS2 by optical helicity, Nature Nanotechnology 7, 494–498 (2012). [35] T. Smoleński, M. Goryca, M. Koperski, C. Faugeras, T. Kazimierczuk, A. Bogucki, K. Nogajewski, P. Kossacki, M. Potemski, Tuning Valley Polarization in a WSe2 Monolayer with a Tiny Magnetic Field, Phys. Rev. X 6, 021024 (2016). [36] A. Arora, K. Nogajewski, M. Molas, M. Koperski, M. Potemski, Exciton band structure in layered MoSe2: from a monolayer to the bulk limit, Nanoscale 7, 20769-20775 (2015). [37] M. R. Molas, C. Faugeras, A. O. Slobodeniuk, K. Nogajewski, M. Bartos, D. M. Basko, M. Potemski, Brightening of dark excitons in monolayers of semiconducting transition metal dichalcogenides, 2D Mater. 4, 021003 (2017). [38] Z. Ye, T. Cao, K. O'Brien, H. Zhu, X. Yin, Y. Wang, S. G. Louie, X. Zhang, Probing excitonic dark states in single-layer tungsten disulphide, Nature 513 214–218 (2014). [39] X.-X. Zhang, Y. You, S. Y. F. Zhao, T. F. Heinz, Experimental Evidence for Dark Excitons in Monolayer WSe2, Phys. Rev. Lett. 115, 257403 (2015). [40] A. Arora, M. Koperski, K. Nogajewski, J. Marcus, C. Faugeras, M. Potemski, Excitonic resonances in thin films of WSe2: from monolayer to bulk material, Nanoscale 7, 10421-10429 (2015). [41] A. M. Jones, H. Yu, N. J. Ghimire, S. Wu, G. Aivazian, J. S. Ross, B. Zhao, J. Yan, D. G. Mandrus, D. Xiao, W. Yao, X. Xu, Optical generation of excitonic valley coherence in monolayer WSe2, Nature Nanotechnology 8, 634–638 (2013). [42] J. S. Ross, S. Wu, H. Yu, N. J. Ghimire, A. M. Jones, G. Aivazian, J. Yan, D. G. Mandrus, D. Xiao, W. Yao, X. Xu, Electrical control of neutral and charged excitons in a monolayer semiconductor, Nature Commun. 4, 1474 (2013). [43] A. Boulesbaa, B. Huang, K. Wang, M. –W. Lin, M. Mahjouri-Samani, C. Rouleau, K. Xiao, M. Yoon, B. Sumpter, A. Puretzky, D. Geohegan, Observation of two distinct negative trions in tungsten disulfide monolayers, Phys. Rev. B 92, 115443 (2015). [44] A. Singh, K. Tran, M. Kolarczik, J. Seifert, Y. Wang, K. Hao, D. Pleskot, N. M. Gabor, S. Helmrich, N. Owschimikow, U. Woggon, X. Li, Long-Lived Valley Polarization of Intravalley Trions in Monolayer WSe2, Phys. Rev. Lett. 117, 257402 (2016). [45] G. Plechinger, P. Nagler, A. Arora, R. Schmidt, A. Chernikov, A. Granados del Águila, P. C. M. Christianen, R. Bratschitsch, C. Schüller, T. Korn, Trion fine structure and coupled spin–valley dynamics in monolayer tungsten disulfide, Nature Communications 7, 12715 (2016). [46] E. Courtade, M. Semina, M. Manca, M. M. Glazov, C. Robert, F. Cadiz, G. Wang, T. Taniguchi, K. Watanabe, M. Pierre, W. Escoffier, E. L. Ivchenko, P. Renucci, X. Marie, T. Amand, B. Urbaszek, Charged excitons in monolayer WSe2 : Experiment and theory, Phys. Rev. B 96, 085302 (2017). [47] D. Vaclavkova, J. Wyzula, K. Nogajewski, M. Bartos, A. O. Slobodeniuk, C. Faugeras, M. Potemski, M. R. Molas, Singlet and triplet trions in WS2 monolayer encapsulated in hexagonal boron nitride, arXiv:1802.05538 (2018). [48] M. R. Molas, K. Nogajewski, A. O. Slobodeniuk, J. Binder, M. Bartos, M. Potemski, The optical response of monolayer, few-layer and bulk tungsten disulfide, Nanoscale 9, 13128-13141 (2017). [49] X.-X. Zhang, T. Cao, Z. Lu, Y.-C. Lin, F. Zhang, Y. Wang, Z. Li, J. C. Hone, J. A. Robinson, D. Smirnov, S. G. Louie, T. F. Heinz, Magnetic brightening and control of dark excitons in monolayer WSe2, Nature Nanotechnology 12, 883–888 (2017). [50] A. O. Slobodeniuk, D. M. Basko, Spin–flip processes and radiative decay of dark intravalley excitons in transition metal dichalcogenide monolayers, 2D Mater. 3, 035009 (2016). [51] M. Koperski, K. Nogajewski, A. Arora, V. Cherkez, P. Mallet, J.-Y. Veuillen, J. Marcus, P. Kossacki, M. Potemski, Single photon emitters in exfoliated WSe2 structures, Nature Nanotechnology 10, 503- 506 (2015). [52] A. Srivastava, M. Sidler, A. V. Allain, D. S. Lembke, A. Kis, A. Imamoğlu, Optically active quantum dots in monolayer WSe2, Nature Nanotechnology 10, 491–496 (2015). [53] Y.-M. He, G. Clark, J. R. Schaibley, Y. He, M.-C. Chen, Y.-Jia Wei, X. Ding, Q. Zhang, W. Yao, X. Xu, C.-Y. Lu, J.-W. Pan, Single quantum emitters in monolayer semiconductors, Nature Nanotechnology 10, 497–502 (2015). [54] C. Chakraborty, L. Kinnischtzke, K. M. Goodfellow, R. Beams, A. Nick Vamivakas, Voltage- controlled quantum light from an atomically thin semiconductor, Nature Nanotechnology 10, 507-511 (2015). [55] C. Robert, T. Amand, F. Cadiz, D. Lagarde, E. Courtade, M. Manca, T. Taniguchi, K. Watanabe, B. Urbaszek, X. Marie, Fine structure and lifetime of dark excitons in transition metal dichalcogenide monolayers, Phys. Rev. B 96, 155423 (2017). [56] A. Castellanos-Gomez, M. Buscema, R. Molenaar, V. Singh, L. Janssen, H. S. J. van der Zant, G. A. Steele, Deterministic transfer of two-dimensional materials by all-dry viscoelastic stamping, 2D Materials 1, 011002 (2014). FIGURES FIG. 1. Magneto-reflectivity spectra, measured with circular polarization resolution (red/blue curves corresponding to e(cid:2)/e(cid:4) polarization, respectively) under excitation with linearly polarised laser light (514 nm), are shown in the energy range of free neutral exciton resonances A and B for (a-b) MoSe2, (e-f) WSe2 and (i-j) WS2 monolayers. The energies of resonances are obtained by fitting the spectra with curves derived by transfer matrix method including Lorentzian contributions to account for excitonic resonances. The magnetic field evolution of the energy of s + (blue dots) components are presented for (c-d) MoSe2, (g-h) WSe2 and (k-l) WS2 monolayers. The open circles represent a mean value of the energy of both components (fe(cid:7)+fe(cid:9))/g, to demonstrate that no detectable diamagnetic shift (term ∝ig) is seen. The values of the g- factors for each transition, obtained by fitting the field-dependent energy fe(cid:7)−fe(cid:9) with linear function jkii, are presented in tables. (red dots) and s -  FIG. 2. Scheme of the valley, spin and orbital contributions to the energies of electronic levels is presented for fundamental sub-bands relevant for the optical transitions at K point of the Brillouin zone, which are observed in absorption-type (and emission) spectra. The upper panel shows the evolution of states in the conduction band, the middle panel shows the evolution of states in the valence band and the bottom panel shows the total energy variation (when all contributions are taken into account) for all states involved in optical transition, which are marked with red/blue arrows for s +/s active optical excitations, respectively. -  FIG. 3. Magneto-reflectivity spectra of (a) MoSe2, (d) WSe2 and (e) WS2 monolayers measured with circular polarization resolution (red curves for e(cid:2) polarization and blue curves for e(cid:4) polarization) are presented in the energy region corresponding to CX state (which appears on the lower energy side of the neutral exciton A). The oscillator strength for (b) MoSe2 and (g) WS2 materials is obtained by fitting the curves using transfer matrix method with Lorentzian contributions. In case of MoSe2, a single transition is used which corresponds to a singlet state of CX. For WS2, it was necessary to include two transitions, which we attribute to singlet (CXs) and triplet (CXt) states of CX, with a zero-field splitting (5meV) between them (see main text for further details). For both materials (c and h) the evolution of the polarization degree with magnetic field is presented. The data are fitted with a function (black solid curves) based on Boltzmann occupation of the electronic states, as described in the main text. FIG. 4. A scheme presenting possible configurations of electrons and a hole forming CX complexes at low temperatures for 'bright' and 'darkish' sc-TMD materials. The thermalisation effects are demonstrated in the presence of the magnetic field (following the simple additive model described in the main text), leading to an enhancement in higher fields of e(cid:2)/e(cid:4) red/blue transitions involving carriers marked with red/blue dots. The triplet state for 'bright' materials is most likely not bound due the parallel configurations of electrons occupying the lower energy conduction state in the same valley.  TAB. 1. Experimentally estimated values of the three terms used to describe the magnetic field impact on the energy of individual states in fundamental sub-bands of sc-TMD monolayers are presented. Two approaches are used, based on different assumptions. In the first case, it is assumed that valley and spin terms are the same for MoSe2 and WS2 materials. In the second case, the spin term is fixed to the value of the free electron in vacuum (gS = 1). Independently of the approach, the valley term is found to provide a significant contribution to the energy of individual states and its value can be estimated to be gV = 1.5 ± 0.5. FIG. 5. A collection of magneto-PL spectra detected in circular polarization resolution (red lines and dots for monolayers. The energy of distinguishable lines is presented as a function of magnetic field ((b) for MoSe2, (e) for WSe2 and (h) for WS2 monolayers) and the g-factors are established (see Tab. 2). The open circles e(cid:2)polarization/blue lines and dots for e(cid:2)polarization) is presented for (a) MoSe2, (d) WSe2 and (g) WS2 represented an arithmetic mean value of the energy of e(cid:2) and e(cid:4) polarized transitions (flmno = pg(fe(cid:7)+fe(cid:9))). No impact of terms quadratic with magnetic field (∝ig) is observed. The polarization degree increase with magnetic field for free excitonic states is also presented (for (c) MoSe2, (f) WSe2 and (i) WS2 monolayers), which is indicative of the electron gas effective temperature (about 10 K) under laser light illumination.  TAB. 2. The g-factor values are presented for PL lines of MoSe2, WSe2 and WS2 monolayers. The values for free exciton states correspond well to those obtained from absorption-type experiments. Notably, the g-factors of lower energy PL lines in tungsten based compounds (representatives of 'darkish' sub-class of materials) are significantly larger in absolute value that g-factors of free exciton states (around 4) and reach values as high as 13. FIG 6. A schematic illustration of forbidden transition in 'darkish' sc-TMD monolayers (such as WSe2 and WS2). (a) The momentum forbidden transitions involve processes with inter-valley recombination, which could be partially allowed considering, e. g. transitions with phonons providing matching momentum. Another type of forbidden transition constitutes (b) inter-valley process, which do not conserve spin. These can be brightened either by mixing of states from different valleys or by considering charged states with recombination realised through spin-flip processes. 

1606.08452

2

1606

Probing Majorana and Andreev Bound States with Waiting Times

We consider a biased Normal-Superconducting junction with various types of superconductivity. Depending on the class of superconductivity, a Majorana bound state may appear at the interface. We show that this has important consequences on the distribution of waiting times of electrons flowing out of such an interface. Therefore, the waiting time distribution is shown to be a clear fingerprint of Majorana bound state physics and may be considered as an experimental signature of its presence.

cond-mat.mes-hall

cond-mat

epl draft 6 1 0 2 c e D 2 1 ] l l a h - s e m . t a m - d n o c [ 2 v 2 5 4 8 0 . 6 0 6 1 : v i X r a Probing Majorana and Andreev Bound States with Waiting Times D. Chevallier1, M. Albert2 and P. Devillard3 1 Department of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland 2 Universit´e Cote d'Azur, CNRS, INLN, France 3 Aix Marseille Univ, Univ Toulon, CNRS, CPT, Marseille, France PACS 02.50.Ey -- Stochastic processes PACS 72.70.+m -- Noise processes and phenomena PACS 74.50.+r -- Superconductivity-Tunneling phenomena Abstract -- We consider a biased Normal-Superconducting junction with various types of super- conductivity. Depending on the class of superconductivity, a Majorana bound state may appear at the interface. We show that this has important consequences on the statistical distribution of time delays between detection of consecutive electrons flowing out of such an interface, namely the waiting time distribution. Therefore, this quantity is shown to be a clear fingerprint of Majorana bound state physics and may be considered as an experimental signature of its presence. Introduction. -- During the last two decades, Majo- rana fermionic states in condensed matter physics, have received a lot of interest because of their exotic properties such as non-Abelian statistics, that open the perspective of using them for quantum computation. These exotic states have been studied extensively in various systems [1 -- 18] with among them, a conceptually simple one made with a semiconducting nanowire of InAs or InSb, with strong spin-orbit coupling, subjected to an external Zee- man field and in the proximity of an s-wave supercon- ductor (SC) [3, 4, 19, 20]. In this situation, a Majorana Bound State (MBS) may appear at the interface of a nor- mal/superconducting junction, under proper conditions, and strongly affects the electronic conduction properties (see Fig. 1). Several experiments have reported the ob- servation of a zero-bias conductance peak in such physical setups, which are in good qualitative agreement with all theoretical predictions based on Majorana physics so far but still not fully consistent with the predicted conduc- tance and magnetic field value needed for the existence of a MBS [21 -- 24]. Therefore, several works have been con- ducted in order to understand these inconsistencies based on alternative interpretations by including other physi- cal processes [25 -- 29]. However, a clear consensus is still lacking mostly because of the absence of an experimental smoking gun for Majoranas. Generally these MBS appear in hybrid junction by tuning one of the parameter of the system (i.e. phase difference or Zeeman field) in a topo- logical phase. Along this transition, these states mutate from Andreev Bound State (ABS) in the non-topological phase to MBS in the topological one and understanding their differences is thus of fundamental importance in or- der to distinguish them. So far, many efforts have focused on the relation and the evolution of ABS onto MBS by tuning the system parameters [8, 30, 31] but less on their own properties [20, 32 -- 34] and the consequences on phys- ical observables which is the purpose of this contribution. Recently, an intriguing feature due to MBS was identi- fied in Ref. [35] and named selective equal-spin Andreev re- flection (SESAR). The presence of a MBS drastically mod- ifies Andreev reflection and leads to a spin polarization of the current as well as to interesting correlations between different spin components which are visible in the zero fre- quency noise [36] for instance. However, such fingerprints are based on the possibility to observe fine quantitative dif- ferences between spin resolved current-current cross cor- relations which seems to be complicated experimentally in the present situation. In this letter, we show that a very clear qualitative difference is visible in the Waiting Time Distribution (WTD) of electrons flowing out of the in- terface making it an interesting and alternative signature of MBS. The WTD is the statistical distribution of time delay between the detection of two consecutive electrons and has been shown to be a very informative and power- ful quantity for understanding correlations in mesoscopic quantum conductors [37 -- 50]. Model. -- We consider two types of hybrid junctions as depicted on Fig. 1 at zero temperature. The first p-1 Andreev reflection takes place, meaning that a hole with a given spin is reflected as an electron with an opposite spin leading to the presence of ABS in such a junction (See Fig. 1 a)). Replacing the s-wave superconductor by a topological one strongly changes scattering properties and especially the Andreev reflection. If the Zeeman field is strong enough to enter the topological phase (Vz ≥ ∆), the p-wave pairing dominates and the Andreev reflection is spin selective [35] meaning that a hole with spin up is reflected as an electron with the same spin and a hole with spin down is normally reflected as a hole with spin down (See Fig. 1 b)). More precisely, the presence of a MBS in the latter case, leads to a spin scattering sym- metry breaking. There is a special spin orientation n, called the Majorana's polarization, along which electrons or holes are totally Andreev reflected as a hole or electron respectively with spin conservation whereas particles with opposite spin are normally reflected. This is the essence of SESAR effect [35] that leads to spin polarized current in this kind of hybrid junction. However, this precise di- rection cannot be determined from first principles and, in general, incoming particles are not spin oriented along this direction which leads to formally more complicated scattering although everything can be understood by de- composing the state onto this spin basis. In order to evaluate the WTD and use it as a tool to probe the scattering properties of ABS and MBS in hy- brid junctions we now need the expressions of the different outgoing scattering states. In such a junction, the inter- face plays an important role on the transmission which gives a finite width to the states [51 -- 53]. In Appendix A, we discuss this effect. However, for the sake of simplic- ity we focus on the zero temperature and perfect Andreev reflection limit and following Ref. [35, 48] write down the different out-going quantum states. Outgoing states for N/S junction. -- In this case, the interface acts as a perfect Andreev mirror where all the holes with a given spin are Andreev reflected as electrons with opposite spin [48] kV(cid:89) k=0 D. Chevallier et al. Fig. 1: (color online) Schematic picture of the Andreev re- flection processes in the two different junctions. a) Normal- (trivial)Superconducting junction: a hole with spin ↑n (↓n) is converted into an electron with opposite spin ↓n (↑n) and b) Normal-(topological)Superconducting junction: a hole with spin ↑n is converted into an electron with same spin ↑n and a hole with spin ↓n is reflected as a hole also with the same spin ↓n. In the first case n denotes any possible direction whereas in the second case a Majorana bound state appears at the inter- face and sets a special spin direction n for scattering (see text). In both cases a bias voltage eV is imposed and brings the su- perconducting chemical potential µS above the Fermi energy EF of the normal metal with the restriction that eV (cid:28) ∆ the superconducting gap. one is a normal metal(N)/s-wave superconductor(S) car- rying an ABS at the interface and the second one is a N/topological superconductor(TS) where the topological junction is made of a Rashba nanowire in proximity with an s-wave superconductor and in presence of Zeeman field Vz carrying a MBS at each boundary. However, we as- sume the nanowire to be long enough to decouple the two MBS. In both situations, the Fermi energy of the normal metal is EF , the superconducting gap is ∆ and the su- perconducting chemical potential µS is biased in such a way that EF = µS − eV and eV (cid:28) ∆ as shown on Fig. 1. As a consequence, a stream of non-interacting holes is approaching the interface from the Normal part where it is scattered as a coherent superposition of electron and holes. This incoming scattering state reads kV(cid:89) k=0 ψin(cid:105) = ck,nck,−n0(cid:105), (1) ψABS out (cid:105) = † −k,−nc c † −k,n0(cid:105). (2) where ck,n is the destruction operator of electron with mo- mentum k, energy E = ¯hvFk and spin orientation n (or creation of holes with opposite properties), 0(cid:105) stands for the Fermi sea filled with states of energies up to µs and kV = eV /¯hvF with vF the Fermi velocity. So far, n de- notes any unitary vector and not necessary z or x for in- stance. Below we will connect it to the polarization axes of the MBS but up to now this is just a choice of basis. It is important to note that spin components ↑z / ↓z or more generally ↑n / ↓n are equally distributed namely the in- coming quantum state is isotropic in spin space. The key difference between these two junctions is how Andreev re- flection occurs. In the case of a NS junction, the usual Again, this quantum state is isotropic in spin space and simply corresponds to a stream of one-dimensional free electrons with energies between µs and µs + eV and two possible spin states (two channels of free fermions). Outgoing states for N/TS junction. -- The pres- ence of the MBS deeply affects the outgoing state. As mentioned before, the key point is that a hole with a spin n is totally reflected as an electron with the same spin whereas a spin −n hole is subjected to perfect specular re- flection with the same spin as well. It is therefore simpler to write the outgoing state in the Majorana polarization frame which reads p-2 a)SNb)NTSMFProbing Majorana and Andreev Bound States with Waiting Times kV(cid:89) k=0 ψMBS out (cid:105) = c † −k,nc−k,−n0(cid:105), (3) where it is now obvious that the outgoing stream of elec- trons is totally spin-polarized in the n direction. There- fore, if one were able to measure the electronic current in this spin direction, one would get the result of a perfect single quantum channel. On the contrary, if one mea- sures it in a random direction, for instance z, one would measure partition noise just because the spin operators Sz does not commute with Sn. In that sense, such an experiment is very similar to the so called Stern and Ger- lach historical experiment as we will discuss later. Apart from this remark the many body state can be simply ob- † n with the proper lin- tained from (3), by replacing the c † ↓ of the right basis. This can ear combination of c be done using the scattering matrix obtained in [35] de- pending on the value of the Zeeman field Vz. However, in the simple case where the Zeeman field is large com- pared to the superconducting gap, the Majorana is fully polarized in the z-direction which means that n and z are the same. Another simple case is right at the topolog- ical transition when the Zeeman field is just above the ). Finally, it is im- gap where c portant to note that in the non-topological case, namely when Vz (cid:28) ∆, the outgoing state behaves like in the s- wave junction (usual Andreev reflection on an ABS) [35] and along the transition, the scattering matrix is not con- tinuous. † −k,↑z 2 (c † −k,↓z + c † ↑ and c † −k,↑ n = 1√ Waiting time distribution. -- We now turn to the calculation of the WTD. To do so, we need to specify the detection process. A time-resolved single electron detector is placed far away from the interface and is assumed to be sensitive to electrons only with energy above the super- conducting chemical potential µS. In addition the detec- tor can be spin selective or not. Following Ref. [46,48] the WTD is obtained from the Idle Time Probability (ITP), namely the probability of not detecting any electron dur- ing a time slot τ . The precise definition of it depends on the detector capabilities. Without spin filtering it reads (cid:82) x0+vF τ Π(τ ) = (cid:104)ψout : e −Q↑,E>µs :: e −Q↓,E>µs : ψout(cid:105), (4) where : ··· : stands for the normal ordering and Qσ,E>µs = c† σ(x)cσ(x)Θ(E − µS) dx is nothing else than the x0 probability of presence of a charge Q during a time slot τ with E > µS and spin projection σ =↑ / ↓ along a given direction (eg x, z...). In Appendix B, we discuss in more details the derivation of Q and Π depending on the applied filtering, energy or/and spin. In the case of spin filtered detection (for instance spin up with respect to a given direction), this quantity is Π(τ ) = (cid:104)ψout : e −Q↑,E>µs : ψout(cid:105). dτ 2 In both cases, the WTD is obtained from the second derivative of the ITP with respect to τ , W(τ ) = (cid:104)τ(cid:105) d2Π(τ ) , where (cid:104)τ(cid:105) is the mean waiting time given by 1/(cid:104)τ(cid:105) = dΠ dτ (τ = 0). Eq. (4) and (5) are evaluated numerically for both many-body scattering states (2) and (3) with the same method as Ref. [46, 48]. However, before discussing our results it is useful to recall several established results on WTD in quantum coherent conductors. In Ref. [39,46], it was shown that for a single quantum channel with a volt- age bias eV (spinless electrons), the scattering quantum state is a train of non interacting fermions whose WTD is approximately the Wigner Surmise WW S(τ ) = 32 π2 τ 2 τ 3 exp 4 π − , (6) (cid:20) (cid:17)2(cid:21) (cid:16) τ τ with τ = h/eV is the average waiting time, which means that due to Pauli's exclusion principle, electrons are sepa- rated in time by τ on average. An important feature of this WTD is the fact that it vanishes for τ (cid:28) τ which is the hallmark of fermionic statistics. If now this stream of elec- tron is partitioned by a scatterer with energy independent transmission coefficient T , this WTD is continuously mod- ified until it reaches an exponential form T exp(−T τ /τ )/τ when T (cid:28) 1. This exponential shape is the signature of uncorrelated events since detected electrons are well sep- arated in time and therefore uncorrelated. In this case, the mean waiting time is (cid:104)τ(cid:105) = τ /T and therefore the average current e/(cid:104)τ(cid:105) = e2 h V T in agreement with the so called Landauer's formula [54]. Finally, when spin 1/2 are considered there are two conducting channel at disposal and the WTD no longer vanishes for small waiting times. At perfect transmission, it is described by the generalized Wigner-Dyson statistics [46]. WTD without spin filtering. -- We start with the simplest situation where the single electron detector is spin insensitive. In the absence of MBS, it was shown [48] that the situation reduces to a stream of one dimensional free electrons with two spin components. Indeed, at perfect Andreev reflection, all the incoming holes are converted into electrons (Andreev mirror) with spin flip and energy between µS and µS + eV . The WTD is therefore the one of two perfect and independent quantum channels and is described by the generalized Wigner-Dyson distribution [46] depicted on Fig. 2a. The average waiting time is (cid:104)τ(cid:105) = h/2eV or in other words the average current is 2e2 h V . On the other hand, in the topological case, the SESAR effect selects only one spin species reducing the possibilities to a single perfect quantum channel (with spin orientation +n). As a consequence, the WTD boils down to the so called Wigner surmise [39] also depicted on Fig. 2a. The average waiting time is h/eV and the average current e2 h V therefore twice smaller than for the topological case. This is in agreement with the common interpretation that a Majorana behaves as "half an electron" [55]. (5) We therefore conclude that not only the WTD repro- p-3 D. Chevallier et al. its WTD is therefore Wigner surmise (see Fig. 2b). This WTD is characterized by a single peak centered around τ = τ with broad fluctuations and exactly zero value at zero which is the hallmark of Pauli exclusion principle as already mentioned. When a MBS is present, the precise shape of the WTD crucially depends on the detector spin orientation. Although quite academical, we can start by setting it to n. In that case, the detector collects every electron coming from the interface and the WTD is also the one of a single quantum channel. In this situation both ABS and MBS yield the same spin resolved WTD. If we choose now d = −n the detector collects nothing and if d slightly deviates from −n only a few electrons are kept and the WTD is expected to be exponential with rate eV h P d where P d is the overlap (cid:104)↑ d ↑n(cid:105)2. For arbitrary d, the detector will partition the single quantum channel according to spin. The situation is al- most formally equivalent to the one of a spinless single quantum channel flowing across a Quantum Point Con- tact (QPC) with energy independent transmission proba- bility [39]. Here this transmission probability will simply be given by the overlap P d between ↑ d(cid:105) and ↑n(cid:105). In the special case where d ⊥ n, the quantum state (3) is a bal- anced mixture of ↑ d(cid:105) and ↓ d(cid:105) and then will be filtered exactly like a single quantum channel across a QPC with transmission probability 1/2. The situation can be imple- mented experimentally either by tuning Vz just above ∆ and setting d = z or in the limit Vz (cid:29) ∆ where n = z and filtering spin along x or y. This is shown on Fig. 2b where we have evaluated Eq. 5 along z right above the topological transition by brute force numerics (limited to a quite small number of basis state (thirteen here) which explains the small discrepancy) and compared it to the expected result with very good agreement. At this point, it is important to give some explanations on the experimental feasibility. The substrate, namely the heterojunction, has already been fabricated during the quest for Majorana quasi-particle [21 -- 23] and consists of a Rashba nanowire partially in contact with an s-wave superconductor and in presence of a Zeeman field. The crucial point is to detect reflected electrons one by one in the normal part. Although still quite challenging, single electron detection technology is progressing very fast and might become a routine in the near future as reviewed in [45, 49]. Otherwise, partial information on waiting times can be extracted from the average current, shot noise or second order coherence function obtained from Hong-Ou- Mandel experiment [45, 56, 57]. Conclusion. -- We have studied the consequences of the presence or not of a MBS at the interface between a normal and a superconducting conductor on the electronic WTD. When a single electron detector is placed far away from the interface and detects electrons above the super- conducting chemical potential µS without spin filtering we observe a clear qualitative distinction between the topological and non topological situations. In addition, Fig. 2: (color online) WTDs versus τ /τ of electrons flowing out of the interface without spin filtering a) and with spin filtering orthogonal to the majorana polarization b). The solid gray line (resp. red line) corresponds to an NS junction without (resp. with a) MBS (see text) at perfect Andreev reflection. The dashed black line represents the WTD of a single channel normal conductor with transmission probability one (a)) and one-half (b)) for comparison [39]. duces well known differences about the average current but also exhibits a qualitative mismatch between the two situations. With MBS, the WTD is exactly zero at τ = 0 because of Pauli's exclusion principle whereas it is not in the ABS case since two channels are available [46]. How- ever, this discrepancy must be visible in any statistical measure of the electronic current like noise, third cumu- lant and Full Counting Statistics (FCS) in the same way that it is between one and two channel standard meso- scopic conductors. WTD with spin filtering. -- We now turn to a richer situation where we assume the single electron detector to be spin sensitive along a direction d and therefore collects only electrons with spin projection ↑ d. The following dis- cussion is basically equivalent to the interpretation of the famous Stern and Gerlach experiment. The key point is that quantum state (2) is spin isotropic whereas (3) is spin polarized in the n direction. In particular, all possible ob- servables are totally independent of the detector spin ori- entation in the ABS case. This is in strong contrast with the MBS where, for instance, filtering spin along ±n leads obviously to orthogonal results. We illustrate this state- ment on the WTD but it is very important to note that it applies to any other observables such as the spin resolved average current [35] or noise [4] and FCS in general. For ABS, the single particle detector, whatever its spin orientation d, filters one spin species, namely ↑ d. Since they are equally populated and independent, the outgoing state (2) reduces to a single perfect quantum channel and p-4 0 0.5 1 1.5 0 1 2 3 4 0 0.5 1 0 1 2 3 4 τ/τW(τ)ABSMBSWignerABSQPCT=1/2MBSa)b)WithoutspinfilteringWithspinfilteringd⊥nProbing Majorana and Andreev Bound States with Waiting Times Appendix B: Derivation of the Idle Time Prob- ability. -- In this appendix we explain how to calculate the idle time probability, following the notations of Ref. [46]. It is important to note that the energy range of de- tected particles is assumed to be small enough that the dis- persion relation of electrons or holes is linear E = ¯hvF k. In that case, charge measurements over a time window ∆t are equivalent to charge measurements over a space window ∆x/vF thanks to Galilean invariance. The key point to calculate Qσ,E is to specify the detection proce- dure. This includes the possibility to only detect posi- tive/negative energies, spin projection up/down with re- spect to a given quantization axis. In an actual experi- ment, the detection can be done by connecting the super- conductor or the topological superconductor to two quan- tum dots instead of a normal metal. By doing so one can filter energy by applying an external gate on the two dots or select the spin by using interacting quantum dots with strong repulsion in order to get rid of the spin degeneracy [61,62] . Analytically these properties can be implemented easily in the definition of Q. This operator can be repre- sented in the basis of the scattering states as QE>µs = ∗ t(k)t (cid:48) ) (k ei(k−k(cid:48))vF τ − 1 i(k − k(cid:48)) dk(cid:48) 2π dk 2π (8) where the t(k) are the energy dependent transmission am- plitudes of a scattering state and may be chosen in that case as t(k) = 1 if E(k) > µs and t(k) = 0 if E(k) < µs. In order to compute the ITP, the transport window has to be discretized into N energy compartments of size eV /N with corresponding momentum intervals of size κ = eV N ¯hvF with vF = ¯hkF m is the Fermi velocity defined with m the electron mass. Using this discretization leads to the fol- lowing matrix elements for QE>µs in the large N limit (cid:90) (cid:104) (κn−κm)vF τ (cid:105) 2 (κn − κm) (9) [Q]m,n = κt∗ κmtκn π − i 2 κ(n−m)vF τ e sin with m, n = 1, ...., N . From this definition of the ITP and when the average is taken over a Slater determinant of free fermions, the WTD can be cast as a determinant of the form [41, 46] Π(τ ) = det(1 − Qτ ), (10) which can be evaluated with a computer. Then, it is straigthforward to extend this detection procedure to a spin selective one by setting spin dependent transmission coefficients. REFERENCES [1] Fu L. and Kane C. L., Phys. Rev. Lett., 100 (2008) 096407. Fig. 3: (color online) WTDs versus τ /τ of electrons flowing out of the interface of a N/S (gray line) and a N/TS (red line). The dashed and solid line correspond to two different width of the states as mentioned in the legend. we have shown that the non topological situation (ABS) is immune to spin filtering in sharp contrast with the topological one due to SESAR effect. This conclusion is valid for the WTD which makes it a clear fingerprint of MBS but is also true for other quantities like the average current or higher moments of the FCS which can be easier to measure in actual experiments. Extension of this work could be the study of the influence of Coulomb repulsion when the superconducting part is not grounded but floating or the poisoning by another Majorana [58,59] and temperature or disorder effects. ∗ ∗ ∗ We are grateful to G. Candela, G. Haack and J. Klino- vaja for useful discussions and remarks. The research of D. C. was supported by the Swiss NSF and NCCR QSIT. Appendix A: Finite width of the outgoing states. -- We study here the effect of a finite energy width of the Majorana bound state and show that it does not qual- itatively change the waiting time distribution. Due to the finite hopping strength at the interface, the Majorana bound state located at zero energy has a finite width Γ. The main consequence of this is the energy dependence of the Andreev reflection coefficient at the interface which becomes [60] RA(E) = . (7) (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)2 Γ E + iΓ However, we can recover the same states (2) and (3) as in the main text when taking the limit eV (cid:28) Γ. Using this assumption, we can calculate the ITP and thus the WTD when the states have a finite width (see Fig. 3) using the energy dependent coefficients [46]. In fig. 3, we can see that the effect of the broadening of the states has no dramatic effects on the WTD. For this reason, we do not focus on this effect in the main text. p-5 0 0.5 1 0 1 2 3 τ/τW(τ)MBSΓ≫eVMBSΓ=eVABSΓ≫eVABSΓ=eVD. Chevallier et al. [2] Sato M. and Fujimoto S., Phys. Rev. B, 79 (2009) 094504. [32] Fu L. and Kane C., Phys. Rev. B, 79 (2009) 161408(R). [33] Rubbert S., Akhmerov A. R., Phys. Rev. B, 94 (2016) [3] Lutchyn R.M., Sau J. D. and Das Sarma S., Phys. Rev. 115430. Lett., 105 (2010) 077001. [34] Beenakker C. W. J. and Kowenhoven L., Nat. Phys., [4] Oreg Y., Refael G. and v. Oppen F., Phys. Rev. Lett., 12 (2016) 618. 105 (2010) 177002. [5] Alicea J., Phys. Rev. B, 81 (2010) 125318. [6] Potter A. C. and Lee P. A., Phys. Rev. B, 83 (2011) 094525. [7] Klinovaja J., Gangadharaiah S. and Loss D., Phys. Rev. Lett., 108 (2012) 196804. [8] Chevallier D., Sticlet D., Simon P. and Bena C., Phys. Rev. B, 85 (2012) 235307. [35] He J. J., Ng T. K., Lee P. A. and Law K. T., Phys. Rev. Lett., 112 (2014) 037001. [36] Haim A., Berg E., von Oppen F. and Oreg Y., Phys. Rev. Lett., 114 (2015) 166406. [37] Brandes T., Ann. Phys. (Berlin), 17 (2008) 477. [38] Albert M., C. Flindt C. and M. Buttiker M., Phys. Rev. Lett., 107 (2011) 086805. [39] Albert M., Haack G., Flindt C. and Buttiker M., [9] Sticlet D., Bena C., and Simon P., Phys. Rev. Lett., Phys. Rev. Lett., 108 (2012) 186806. 108 (2012) 096802. [10] Klinovaja J., Stano P. and Loss D., Phys. Rev. Lett, 109 (2012) 236801. [40] Thomas K. H. and C. Flindt C., Phys. Rev. B, 87 (2013) 121405(R). [41] Dasenbrook D., Flindt C., and M. Buttiker M., [11] Nadj-Perge S.,Drozdov I. K.,Bernevig B. A., and Yazdani A., Phys. Rev. B, 88 (2013) 020407(R). Phys. Rev. Lett., 112 (2014) 146801. [42] Rajabi L., Poltl C. and Governale M., Phys. Rev. [12] Klinovaja J., Stano P., Yazdani A., and Loss D., Lett., 111 (2013) 067002. Phys. Rev. Lett., 111 (2013) 186805. [43] Thomas K. H. and Flindt C., Phys. Rev. B, 89 (2014) [13] Braunecker B. and Simon P., Phys. Rev. Lett., 111 245420. (2013) 147202. [14] Vazifeh M. M. and Franz M., Phys. Rev. Lett., 111 [44] Sothmann B., Phys. Rev. B, 90 (2014) 155315. [45] Albert M. and Devillard P., Phys. Rev. B, 90 (2014) (2013) 206802. 035431. [15] Pientka F., Glazman L. I. and v. Oppen F., Phys. [46] Haack G., Albert M. and C. Flindt C., Phys. Rev. B, Rev. B, 88 (2013) 155420. [16] Poyhonen K., Weststrom A., Rontynen J. and Oja- nen T., Phys. Rev. B, 89 (2004) 115106. 90 (2014) 205429. [47] Hofer P. P., Dasenbrook D. and C. Flindt C., Phys- ica E (2015), doi: 10.1016/j.physe.2015.08.034, () . [17] Maier F., Klinovaja J. and Loss D., Phys. Rev. B, 90 [48] Albert M., Chevallier D. and Devillard P., Physica (2014) 195421. E, 76 (2016) 209. [18] Vernek E., Penteado P. H., Seridonio A. C. and [49] Dasenbrook D. and C. Flindt C., Phys. Rev. B, 93 Egues J. C., Phys. Rev. B, 89 (2014) 165314. [19] Alicea J., Rep. Prog. Phys., 75 (2012) 076501. [20] Beenakker C. W. J., Annu. Rev. Con. Mat. Phys., 4 (2013) 113. [21] Mourik V., Zuo K., Frolov S. M., Plissard S. R., Bakkers E. P. A. M. and Kouwenhoven L. P., Science, 336 (2012) 1003. [22] Churchill H. O. H., Fatemi V., Grove-Rasmussen K., Deng M. T., Caroff P., Xu H. Q. and Marcus C. M., Phys. Rev. B, 87 (2013) 241401. [23] Das A., Ronen Y., Most Y., Oreg Y., Heiblum M. and Shtrikman H., Nat. Phys., 8 (2012) 887. [24] Rokhinson L. P., Liu X. and Furdyna J. K. , Nat. Phys., 8 (2012) 795. [25] Lee E. J. H., Jiang X., Aguado R., Katsaros G., Lieber C. M. and De Franceschi S., Phys. Rev. Lett., 109 (2012) 186802. [26] Liu J., Potter A. C., Law K. T. and Lee P. A., Phys. Rev. Lett., 109 (2012) 267002. (2016) 245409. [50] Brandes T. and Emary C., Phys. Rev. E, 93 (2016) 042103. [51] Pikulin D. I. and Nazarov Y., JETP Lett., 94 (2012) 693697. [52] Pikulin D. I. and Nazarov Y., Phys. Rev. B, 87 (2013) 235421. [53] San Jose P., Cayao J., Prada E. and Aguado R., Sci. Rep., 6 (2016) 21427. [54] Blanter Ya. M. and Buttiker M., Phys. Rep., 336 (2000) 1. [55] Kitaev A. Y., Phys. Usp., 44 (2001) 131. [56] E. Bocquillon et al, Annalen der Physik, 526 (2014) 1. [57] Haack G., Steffens A., J. Eisert J. and Hubener R., New J. Phys., 17 (2015) 113024. [58] Beri B. and Cooper N. R., Phys. Rev. Lett., 109 (2012) 156803. [59] Plugge S., Zazunov A., Erikson E., Tsvelik A. M. and Egger R., Phys. Rev. B, 93 (2016) 104524. [27] Bagrets D. and Altland A., Phys. Rev. Lett., 109 [60] Ioselevich. P, and Feigel'man M. V., New. J. Phys., (2012) 227005. [28] Pientka F., Kells G., Romito A., Brouwer P. W. and von Oppen F., Phys. Rev. Lett., 109 (2012) 227006. [29] Pikulin D. I., Dahlhaus J. P., Wimmer M., Schome- rus H. and Beenakker C. W. J., New. J. Phys., 14 (2012) 125011. [30] Klinovaja J. and Loss D., Phys. Rev. B, 86 (2012) 085408. [31] Prada E., San-Jose P. and Aguado R., Phys. Rev. B, 86 (2012) 180503(R). 15 (2013) 055011. [61] Chevallier. D, Rech J., Jonckheere T. and Martin T., Phys. Rev. B, 83 (2011) 125421. [62] Recher. P, Sukhorukov E. V. and Loss D., Phys. Rev. B, 63 (2001) 165314. p-6 

1911.12016

1

1911

Quantum dot exciton dephasing by Coulomb interaction. A fermionic analogue of the independent boson model

The time evolution of a quantum dot exciton in Coulomb interaction with wetting layer carriers is treated using an approach similar to the independent boson model. The role of the polaronic unitary transform is played by the scattering matrix, for which a diagrammatic, linked cluster expansion is available. Similarities and differences to the independent boson model are discussed. A numerical example is presented.

cond-mat.mes-hall

cond-mat

a Quantum dot exciton dephasing by Coulomb interaction. A fermionic analogue of the independent boson model 1 National Institute of Materials Physics, Atomistilor 405A, Magurele 077125, Romania and I.V. Dinu1,2, M. T¸ olea1, and P. Gartner2 2 Centre International de Formation et de Recherche Avanc´ees en Physique - NIMP, Atomistilor 407, Magurele 077125, Romania (Dated: November 28, 2019) The time evolution of a quantum dot exciton in Coulomb interaction with wetting layer carriers is treated using an approach similar to the independent boson model. The role of the polaronic unitary transform is played by the scattering matrix, for which a diagrammatic, linked cluster expansion is available. Similarities and differences to the independent boson model are discussed. A numerical example is presented. PACS numbers: I. INTRODUCTION Quantum dots (QD) in semiconductor heterostruc- tures are sometimes regarded as artificial atoms, but one aspect in which they differ from real atoms is that they live in an invasive environment. QD carriers interact with both acoustic and optical phonons, and sometimes with free carriers in the wetting layer (WL) or in the bulk. Such interactions play an important role in the QD opti- cal properties, and especially in the dissipative behavior, like carrier population redistribution and polarization de- phasing. In this respect, the role of the phonons enjoyed a much larger attention in the literature, one reason being the availability of the independent boson model (IBM)1 -- 3, which is both simple and in certain circumstances exact. The popularity of the method cannot be overestimated, it being used in various contexts like the theory of exci- ton dephasing and absorption4 -- 6, phonon-assisted exci- ton recombination7, phonon-mediated off-resonant light- matter coupling in QD lasers8,9, generation of entan- gled phonon states10 or phonon-assisted adsorption in graphene11, to cite only a few. The IBM relies on the carrier-phonon interaction be- ing diagonal in the carrier states. If not, or if additional interactions are present, the method ceases to be exact, but it is still helpful as a way to handle a part of the interaction, responsible for the polaron formation. This diagonality implies that the bosons see an occupation number, either zero or one, according to whether the QD excitation is present or not. The difference between the two cases amounts, in a linear coupling, to a displace- ment of the phonon oscillation centers, without changing their frequency. In other words one has just a change of basis, performed by a unitary operator, the polaronic transform. The diagonality requirement means that the method is particularly suited for calculating pure dephas- ing, i.e. polarization decay without population changes, and absorption spectra4,12, where it produces exact ana- lytic results. In view of this wide range of applications it is legiti- mate to ask whether such a unitary transform does not exist for the interaction with the free carriers too. We show that the answer is positive, if we frame the prob- lem as above, i.e. if the interaction is diagonal in the QD states. We illustrate the situation in the case of a two-level system where the electron-hole vacuum acts as the ground state and an exciton pair as the excited state. The charge distribution of the latter acts as a scattering center for the carriers in the continuum. It is known13,14 that the free and the scattered continua are unitary equivalent, with the transform provided by the scattering matrix. In many ways this approach re- sembles the IBM, and can be regarded as its fermionic analogue. Many similarities between the two can be seen, but we also point out the differences. Most importantly, the method is not exact any more, but it lends itself to a diagrammatic expansion. In a previous paper15 this approach has been already used in the context of a nonresonant Jaynes-Cummings (JC) model. The cavity feeding was assisted by the fermionic bath of WL carriers, which compensated the energy mismatch. The situation there is more compli- cated due to the QD states getting mixed by the JC in- teraction with the photonic degrees of freedom. In the present paper, we address a simpler, more clearcut situation, involving only the exciton-continuum interaction. We illustrate the method by calculating the polarization decay and absorption line shape, as func- tions of bath temperature and carrier concentration. We also discuss in more detail the similarities and differences with respect to the IBM. II. THE MODEL The system under consideration consists of a QD ex- citon interacting with a fermionic thermal bath, repre- sented by the WL carriers. The latter are taken inter- actionless and in thermal equilibrium. The Hamiltonian describing the problem is ( = 1) with S generated by the scattering potential W 2 H = εX X †X + (1 − X †X) h0 + X †X hX , h0 = Xλ=e,hXk hX = h0 + W . kλ† ελ kλk , (1) Here X †, X are the excitonic creation and annihilation operators. Specifically, considering in the QD one s-state in each band, with operators e†, e and h†, h for elec- trons and holes respectively, we have X = he. Limit- ing the QD configurations to the neutral ones, one has X †X = e†e = h†h. The exciton energy is εX . The WL Hamiltonian is given by h0, with the subscripted λ sym- bol meaning either e or h WL-continuum operators, and the momentum index k including tacitly the spin. In the presence of the exciton the WL Hamiltonian hX gets ad- ditionally a term W , describing the interaction with the QD carriers. This is expressed in terms of the matrix elements V λλ′ ij,kl =Xq Vq hϕλ i eiqr ϕλ l ihϕλ′ j e−iqr ϕλ′ k i (2) of the Coulomb potential Vq, between one-particle states, with λ, λ′ = e, h. Only two of these indices are needed since band index conservation is, as usual, assumed. The WL-QD interaction has the form X †X W = Xk,k′ Xk,k′ e† kek′(cid:2)V ee khk′(cid:2)V hh h† sk,k′se†e − V he sk,k′sh†h − V ee sk,k′sh†h − V eh sk,k′se†e − V hh sk,sk′e†e(cid:3) + sk,sk′ h†h(cid:3) . (3) It shows that the WL carriers are scattered by an external field produced by the exciton, having the form W = Xλ=e,hXk,k′ W λ k,k′λ† kλk′ . (4) In each W λ k,k′ the first two terms describe direct, elec- trostatic interaction between WL-carriers with the QD electron and hole respectively. The difference between repulsion and attraction is nonzero due to the different charge densities of these two. The exciton is globally neutral, but local charges are usually present and gen- erate scattering. The strength of the scattering depends on the degree of charge compensation within the exci- ton. The third term is the exchange contribution, and it is not expected to be large, since around q = 0 the ma- trix elements in Eq.(2) become overlap integrals between orthogonal states. The idea of the present method relies on the unitary equivalence between the WL Hamiltonians h0 and hX provided by the scattering matrix S(0,−∞) (see e.g. 13,14). One has S(−∞, 0) hX S(0,−∞) = h0 , (5) S(t1, t2) = S †(t2, t1) = T exp(cid:20)−iZ t1 t2 W (t) dt(cid:21) , (6) t1 > t2 . T is the time ordering operator and the interaction rep- resentation of the perturbation W (t) with respect to h0 is used. For the full WL-QD Hamiltonian we formally eliminate the interaction part using the unitary transform U = 1 − X †X + X †X S(0,−∞) , (7) which switches on the action of the S-matrix only when the exciton is present. It follows that U †(cid:2)(1 − X †X) h0 + X †X hX(cid:3) U = (1 − X †X) h0+X †X S(−∞, 0) hX S(0,−∞) = h0 . The excitonic operators are changed, according to eX = U † X U = X S(0,−∞) = S(0,−∞) X and similarly eX † = X †S(−∞, 0), but X †X remains invariant. There- fore in the transformed problem (8) (9) eH = U † H U = εX X † X + h0 , the QD and the WL become uncoupled. This follows faithfully the effect produced by the polaronic transform in the bosonic bath case, as described by the IBM. Assuming an instantaneous excitation at t = 0, the linear optical polarization of the QD is expressed in terms of the exciton retarded Green's function P(t) = −i θ(t)hX(t)X †i = −i θ(t)Tr{ρ X(t)X †} . (10) In the unitary transformed picture P(t) = −i θ(t)Tr{eρ eX(t)eX †} , the problem is interaction-free, and QD and WL opera- tors evolve independently. Therefore (11) eX(t) = e−iεX tXeih0tS(0,−∞)e−ih0t = e−iεX tXS(t,−∞) , (12) and as a consequence P(t) = −i θ(t)e−iεX t hXX †ihS(t, 0)i . (13) Essentially, besides a trivial exciton energy oscillation, the problem boils down to evaluating the thermal bath average of the scattering matrix S(t, 0) for positive times. This is again formally similar to the IBM case, but dif- fering in the details, as will be discussed below. In the present case one makes use of the linked cluster (cumu- lant) expansion for hS(t, 0)i1,16, in which a lot of resum- mation has been performed. As a consequence hS(t, 0)i is expressed as an exponential, exp[Φ(t)], where Φ(t) is the (a) (b) k t 1 W λ kk L 1 W λ kk ' k k ' 1/2 k ' W λ k ' k W λ kk ' t 3 W λ k ' ' k t 1 t 2 k k L 2 W λ k ' k ' ' 1/3 k ' ' t 2 k ' t 1 W λ kk ' L 3 3 Initially it decays quadratically with time, as obvious from the double integral from 0 to t in Eq.(14), and also reflected in Eq.(16) by the subtraction of the first two terms in the expansion of the exponential. More relevant is the long-time behavior of the real part k,k′(cid:12)(cid:12)(cid:12) ReL2(t) = − Xλ,k,k′(cid:12)(cid:12)(cid:12)W λ 2 (1 − f λ k )f λ k′ kk′ t) 1 − cos(ελ kk′)2 (ελ , (17) which controls the polarization decay. Indeed, using the large t asymptotics of (1 − cos ωt)/ω2 ∼ π δ(ω) t, one finds an exponential attenuation P(t) ∼ exp(−Γt) with the decay rate given by Figure 1: (a) Elementary interaction vertex. (b) First three connected diagrams L1, L2 and L3 of the linked cluster expan- sion in the evaluation of the thermal average of the S-matrix hS(t, 0)i. Γ = π Xλ,k,k′(cid:12)(cid:12)(cid:12)W λ k,k′(cid:12)(cid:12)(cid:12) 2 (1 − f λ k )f λ k′ δ(ελ kk′ ) . (18) sum of all connected diagrams with no external points Φ(t) = Pn Ln(t), where the diagram Ln , n = 1, 2, 3 . . . of order n comes with a factor 1/n. Its internal points are time-integrated from 0 to t. In our case the interac- tion is an external potential, not a many-body one and the elementary interaction vertex in the diagrams is as in Fig. 1(a). The first diagrams of the expansion are represented in Fig. 1(b). One has 0 dt1W λ Z t L1(t) = −Xλ,k k,k′(cid:12)(cid:12)(cid:12) 2 Xλ,k,k′(cid:12)(cid:12)(cid:12)W λ L2(t) = − ×Z t dt1Z t 1 0 0 k,k G0 λk(t1, t+ 1 ) 2 (14) dt2 G0 λk(t1, t2) G0 λk′ (t2, t1) , where G0 λk. λk is the free Green's function for the WL state For the first-order contribution one obtains an imagi- nary, linear time dependence, which amount to a correc- tion to the exciton energy. L1(t) = −iXλ,k W λ k,k f λ k t (15) with f λ same indices. k the Fermi function for the WL level carrying the More important is the second order diagram (ελ kk′ de- notes the difference ελ k − ελ k′ ) 2 (1 − f λ k )f λ k′ k,k′(cid:12)(cid:12)(cid:12) L2(t) = Xλ,k,k′(cid:12)(cid:12)(cid:12)W λ ×(cid:16)e−i ελ kk′ t − 1 + i ελ kk′ t(cid:17)(cid:16)ελ kk′(cid:17)−2 (16) . A comparative discussion with the IBM is in order. The dephasing process does not imply a change of pop- ulation (pure dephasing) and therefore the decay rate Γ does not involve energy transfer, as seen by the presence of the δ-function. In the case of IBM that means only zero-energy phonons are involved. Then all the discus- sion takes place around the spectral edge, and depends on the density of states and coupling constants there. Usu- ally they vanish as a higher power of energy and overcome the singularity of the Bose-Einstein distribution, with the result that Γ = 0. This leads to the problem of the zero- phonon line (ZPL) appearing as an artificial pure δ-peak in the spectrum. This is a weak point and several ways out have been devised, like including a phenomenological line broadening4, a phonon-phonon interaction17, or con- sidering a lower dimensionality5 to enhance the density of states. The fermionic case is free from this problem, since Γ relies on quantities around the chemical potential. Also, it is worth noting that limiting the expansion to L2 gives the exact result in the IBM, while here it is only an approximation. One may plead in favor of neglecting higher diagrams by arguing that a lot of compensation takes place between the direct terms in Eq.(3) and the exchange terms are small, in other words the QD-WL coupling is weak. Nevertheless, this is not sufficient, since it might turn out that higher order diagrams behave as higher powers in time, and thus asymptotically overtake the second order one. We argue below that this is not the case. Indeed, the structure of the diagrams is such that the θ-functions contained in the Green's functions splits the expression into integrals of the form In(t) =Z t 0 dt1eiω1t1Z t1 0 dt2eiω2t2 . . .Z tn−1 0 dtneiωntn . (19) The frequency appearing in each time integration is the difference of the energies of the Green's functions meet- ing at the corresponding internal points. Summing these pairwise differences around the diagram entails the rela- tion ω1 + ω2 + . . . + ωn = 0. On the other hand, the Laplace transform of In(t) can be easily calculated and gives Jn(s) = 1 s 1 s + iω1 1 s + i(ω1 + ω2) ··· 1 s + i(ω1 + ω2 + . . . + ωn) ··· (20) The last factor is actually 1/s, like the first, so that Jn(s) ∼ 1/s2 around s = 0. This corresponds to a behavior In(t) ∼ t as t → ∞ for all n. For instance, the n = 2 case, discussed above, can be recovered from J2(s) = 1/s2 · 1/(s + iω1). The low-s asymptotics of its real part generates the linear large time behavior times the πδ(ω1) factor, with ω1 = ελ k,k′. Using g as a generic notation for the coupling strength, we conclude that the contribution of the diagram of order n at large times is ∼ gn t. This is in agreement with the so-called weak interaction limit18, stating that when g → 0 and simultaneously t → ∞ so that g2 t remains constant, the Born-Markov dissipative evolution becomes exact. Indeed, here all n > 2 contributions vanish in this scaling limit. III. NUMERICAL EXAMPLE As an illustration we consider an InAs/GaAs het- erostructure, with a self-assembled QD on a WL of L = 2.2 nm width. The relevant material parame- ters are those of Vurgaftman et al.19. We assume the wavefunctions to be factorized into the square-well solu- tion χλ(z) along the growth direction and the in-plane function. The latter are taken as oscillator ground- state gaussians for the QD s-states for electrons and holes, and as plane waves, orthogonalized on the for- mer, for the WL extended states. The gaussians are defined by their width αλ in the reciprocal space, i.e. ϕλ λr2/2) with r here the in-plane position. These parameters depend on many geometric and composition features of the QD, so that they can reach a broad set of values. For the sake of our example we take αe = 0.2/nm and αh = 0.1/nm. s (r) = αλ/√π exp(−α2 The phonon-induced dephasing is expected to be less important at low temperatures. The Coulomb-assisted dephasing depends on both temperature and WL-carrier concentration, therefore lowering the temperature and in- creasing the concentration it has a chance to compete with the phononic processes. We consider only the neu- tral charging, with electrons and holes in the WL at the same concentration n. In Fig. 2 the time evolution of the real part of Φ(t) is plotted for two different temperatures. The inital quadratic behavior is followed by a linear decrease, whose slope Γ is the dephasing rate predicted by Eq.(18). It in- creases with temperature, as confirmed by Fig. 3, which shows the values of Γ at various temperatures and carrier concentrations. 4 (b) 0.000 -0.002 ] ) t ( [ e R -0.004 0.000 (a) -0.002 -0.004 -0.006 T = 5 K -0.006 T = 10 K -0.008 0.0 1.0 0.5 Time (ps) -0.008 0.0 1.5 1.0 0.5 Time (ps) 1.5 Figure 2: (color online) Time evolution of the real part of Φ for two temperatures, 5K (a) and 10K (b) at the same carrier concentration n = 1012/cm2. The dephasing reaches a linear decay whose rate increases with temperature. ) s p / 1 ( 0.018 0.016 0.014 0.012 0.010 0.008 0.006 0.004 0.002 0.000 n = 1012 / cm2 n = 5 1011 / cm2 n = 1011 / cm2 0 20 40 60 80 Temperature (K) 100 120 Figure 3: (color online) Dephasing rates Γ as functions of temperature for different carrier concentrations. The range of those values is such that Γ is of the or- der of a few µeV. This is comparable with results for dephasing by phonons at low temperatures both in theo- retical simulations4,20 and in experimental data21,22. Ex- perimental data obtained by four-wave mixing22 do not separate phonon and injected carrier contributions to de- phasing, but their total effect is still in the µeV range. For an increase of temperatures from 5K to 120K the dephasing grows by roughly one order of magnitude. In the same conditions the rate of dephasing by phonons gains two orders of magnitude4,21, showing a higher sen- sitivity to temperature. Yet in the case of the fermionic bath the decay is enhanced also by increasing a second ) s t i n u . b r a ( m u r t c e p S 1.0 0.8 0.6 0.4 0.2 0.0 T = 10 K T = 30 K T = 50 K T = 100 K -0.0375 -0.0300 -0.0225 -0.0075 0.0000 0.0075 -0.0150 ( ps) 5 physical pure δ-spike in the frequency domain. This is not the case with the fermionic bath, as also seen in Fig. 4. The main feature of the spectra is their Lorentzian shape, as a consequence of the exponential decay in the time domain. Still, the quadratic initial behavior replaces the cusp at t = 0 of a pure exponential by a smooth matching. In the frequency domain this leads to a departure from the Lorentzian, in the sense of a faster decay at large frequencies. In Fig. 5 we also consider the dependence of dephasing on other, more geometric parameters. It is seen that Γ is not very sensitive to the WL width L, but it is sig- nificantly influenced by the spatial extension of the QD s-states. The broader the states, the stronger the de- phasing, due to a more efficient scattering. Figure 4: (color online) Absorption spectra for n = 1012/cm2 at T=10, 30, 50, 100K IV. CONCLUSIONS 0.030 0.025 0.020 ) s p / 1 ( 0.015 0.010 0.005 L = 4.0 nm L = 2.2 nm e = 0.15 / nm, h = 0.075 / nm e = 0.20 / nm, h = 0.100 / nm e = 0.30 / nm, h = 0.150 / nm 0.000 0 20 40 60 80 Temperature (K) 100 120 Figure 5: (color online) Temperature dependence of the de- phasing rate for different WL width L and space extension parameters α. All curves for n = 1012/cm2 controllable parameter, the carrier concentration. As mentioned above, the description of the phonon de- phasing by the IBM runs into the ZPL problem. As seen in Refs.6,17 the slope of the long-time linear asymptotics is zero, for reasons discussed Sec.II. This leads to an un- 1 G. D. Mahan, Many Particle Physics (Springer, 2000), 3rd ed., ISBN 0306463385. 2 K. Huang and A. Rhys, Proc. R. Soc. Ser. A 204, 406 (1950). 3 C. B. Duke and G. D. Phys. Rev. 139, A1965 (1965), Mahan, URL In conclusion, we have shown that a fermionic counter- part of the popular IBM is possible. It describes the QD exciton interaction with the fermionic bath consisting of injected carriers in the bulk or WL. Similarities and dif- ferences to the IBM are pointed out. For instance, the present solution takes the form of a diagrammatic series expansion, while the IBM is exact, but this advantage is lost as soon as other interactions are present. Also, our case is free from the ZPL problem inherent to the bosonic case. The dephasing process is controlled not only by temperature but also by the chemical potential of the bath. The numerical illustration shows that at low temperatures and higher carrier concentrations the dephasing times are comparable with those produced by the phonon interaction. But, of course, this is also depen- dent on the parameters of the particular case considered. The dephasing gets stronger at higher temperature and concentration, as well as with broader charge distribution of QD states. Acknowledgments The authors acknowledge financial from CNCS-UEFISCDI Grant No. PN-III-P4-ID-PCE-2016- 0221 (I.V.D. and P.G.) and from the Romanian Core Pro- gram PN19-03, Contract No. 21 N/08.02.2019, (I.V.D. and M. T¸ .). support https://link.aps.org/doi/10.1103/PhysRev.139.A1965. 4 R. Zimmermann and E. Runge, in Proceedings of the 26th ICPS, Edinburgh, UK, edited by A. Long and J. Davies (IOP Publishing, Bristol, 2002), p. M 3.1. 5 G. Lindwall, A. Wacker, C. Weber, and A. Knorr, URL Phys. Rev. (2007), 087401 Lett. 99, https://link.aps.org/doi/10.1103/PhysRevLett.99.087401. 6 X. Li, T. Wang, and C. Dong, IEEE J. Quantum Electron. F. Jahnke, Phys. Rev. B 89, 161302 (2014), URL https://link.aps.org/doi/10.1103/PhysRevB.89.161302. 6 50, 548 (2014). 7 L. Dusanowski, A. Musia l, A. Mary´nski, P. Mrowi´nski, J. Misiewicz, Reithmaier, (2014), URL J. Andrzejewski, P. Machnikowski, A. et https://link.aps.org/doi/10.1103/PhysRevB.90.125424. Hofling, al., Phys. Rev. B 90, J. 125424 Somers, P. S. 8 M. Florian, P. Gartner, C. Gies, and F. Jahnke, New J. Phys. 15, 035019 (2013). 9 U. Hohenester, Phys. Rev. B 81, 155303 (2010), URL https://link.aps.org/doi/10.1103/PhysRevB.81.155303. and D. Wig- (2019), URL 10 T. Hahn, D. Groll, T. Kuhn, ger, Phys. Rev. B 100, https://link.aps.org/doi/10.1103/PhysRevB.100.024306. 024306 11 S. Sengupta, Phys. Rev. B 100, 075429 (2019), URL https://link.aps.org/doi/10.1103/PhysRevB.100.075429. 12 L. Besombes, K. Kheng, L. Marsal, 155307 riette, Phys. Rev. B 63, https://link.aps.org/doi/10.1103/PhysRevB.63.155307. and H. Ma- (2001), URL 13 J. R. Taylor, Scattering Theory (Dover Publications, N.Y., 2006). 14 A. G. Sitenko, Scattering Theory (Springer, 1991), ISBN 3642840361. 15 M. Florian, P. Gartner, A. Steinhoff, C. Gies, and 16 A. A. Abrikosov, Methods of Quantum Field Theory in Sta- tistical Physics (Dover Publications, N.Y., 1975), revised ed., ISBN 0486632288. Muljarov 17 E. and A. R. 237401 Zimmermann, URL Phys. Rev. https://link.aps.org/doi/10.1103/PhysRevLett.93.237401. (2004), Lett. 93, 18 H. Spohn, Rev. Mod. Phys. 52, 569 (1980), URL https://link.aps.org/doi/10.1103/RevModPhys.52.569. 19 I. Vurgaftman, J. R. Meyer, Mohan, Appl. Phys. Rev. 89, https://doi.org/10.1063/1.1368156. and L. R. Ram- 5815 (2001), URL 21 P. Borri, W. Langbein, 20 T. Takagahara, Phys. Rev. B 60, 2638 (1999), URL https://link.aps.org/doi/10.1103/PhysRevB.60.2638. S. Schneider, U. Wog- and D. Bim- 157401 (2001), URL gon, R. L. Sellin, D. Ouyang, berg, Phys. Rev. Lett. 87, https://link.aps.org/doi/10.1103/PhysRevLett.87.157401. 22 P. Borri, W. Langbein, gon, R. L. Sellin, D. Ouyang, berg, Phys. Rev. Lett. 89, https://link.aps.org/doi/10.1103/PhysRevLett.89.187401. S. Schneider, U. Wog- and D. Bim- 187401 (2002), URL 

1811.04390

2

1811

Unconventional quantum optics in topological waveguide QED

The discovery of topological materials has challenged our understanding of condensed matter physics and led to novel and unusual phenomena. This has motivated recent developments to export topological concepts into photonics to make light behave in exotic ways. Here, we predict several unconventional quantum optical phenomena that occur when quantum emitters interact with a topological waveguide QED bath, namely, the photonic analogue of the Su-Schrieffer-Hegger model. When the emitters frequency lies within the topological band-gap, a chiral bound state emerges, which is located at just one side (right or left) of the emitter. In the presence of several emitters, it mediates topological, long-range tunable interactions between them, that can give rise to exotic phases such as double N\'eel ordered states. On the contrary, when the emitters' optical transition is resonant with the bands, we find unconventional scattering properties and different super/subradiant states depending on the band topology. We also investigate the case of a bath with open boundary conditions to understand the role of topological edge states. Finally, we propose several implementations where these phenomena can be observed with state-of-the-art technology.

cond-mat.mes-hall

cond-mat

Unconventional quantum optics in topological waveguide QED M. Bello,1 G. Platero,1 J. I. Cirac,2 and A. Gonz´alez-Tudela2, 3 1Instituto de Ciencia de Materiales de Madrid, CSIC, 28049, Spain 2Max-Planck-Institut fur Quantenoptik, Hans-Kopfermann-Strasse 1, 85748 Garching, Germany 3Instituto de F´ısica Fundamental IFF-CSIC, Calle Serrano 113b, Madrid 28006, Spain.∗ (Dated: July 31, 2019) The discovery of topological materials has challenged our understanding of condensed matter physics and led to novel phenomena. This has motivated recent developments to export topological concepts into photonics to make light behave in exotic ways. Here, we predict several unconven- tional quantum optical phenomena that occur when quantum emitters interact with a topological waveguide QED bath, namely, the photonic analogue of the Su-Schrieffer-Heeger model. When the emitters' frequency lies within the topological band-gap, a chiral bound state emerges, which is located at just one side (right or left) of the emitter. In the presence of several emitters, this bound state mediates topological, tunable interactions between them, that can give rise to exotic many-body phases such as double N´eel ordered states. Furthermore, when the emitters' optical transition is resonant with the bands, we find unconventional scattering properties and different su- per/subradiant states depending on the band topology. Finally, we propose several implementations where these phenomena can be observed with state-of-the-art technology. I. INTRODUCTION Even though the introduction of topology in condensed matter was originally motivated to explain the integer Quantum Hall effect [1], its implications were more far- reaching than expected. On a fundamental level, topo- logical order resulted in a large variety of new phenom- ena, as well as new paradigms for classifying matter phases [2]. On practical terms, topological states can be harnessed to achieve more robust electronic devices or fault-tolerant quantum computation [3]. This spec- tacular progress motivated the application of topological ideas to photonics, for example, to engineer unconven- tional light behaviors. The starting point of the field was the observation that topological bands also appear with electromagnetic waves [4]. Soon after that, many experimental realizations followed using microwave pho- tons [5], photonic crystals [6, 7], coupled waveguides [8] or resonators [9 -- 11], exciton-polaritons [12] or metamateri- als [13], to name a few (see [14] and references therein for an authoritative review). Nowadays, topological photon- ics is a burgeoning field with many experimental and the- oretical developments. Among them, one of the current frontiers of the field is the exploration of the interplay between topological photons and quantum emitters [15 -- 17]. In this manuscript, we show that topological pho- tonic systems cause a number of unprecedented phe- nomena in the field of quantum optics, namely, when they are coupled to quantum emitters. We analyze the simplest model consisting of two-level quantum emit- ters (QEs) interacting with a one-dimensional topolog- ical photonic bath described by the Su-Schrieffer-Heeger (SSH) model [18] (see Fig. 1). When the QE frequency ∗ [email protected] FIG. 1. System schematic. (A) Schematic picture of the present setup: Ne two-level quantum emitters interact with the photonic analogue of the SSH model. This model is char- acterized by having alternating hopping amplitudes J(1 ± δ), where J defines their strength, while δ, the so-called dimeriza- tion parameter, controls the asymmetry between them. The interaction with photons (in transparent red) induces non- (B) Bath's energy trivial dynamics between the emitters. bands for a system with a dimerization parameter δ = 0.2. The main spectral regions of interest for this manuscript are the middle band-gap (green) and the two bands (blue). 9 1 0 2 l u J 9 2 ] l l a h - s e m . t a m - d n o c [ 2 v 0 9 3 4 0 . 1 1 8 1 : v i X r a BAlies between the two bands (green region in Fig. 1B) we predict the emergence of chiral photon bound states (BS), that is, BSs which localize to the left/right side of the QEs depending on the topology of the bath. In the many-body regime (i.e., with many emitters) those BSs mediate tunable, chiral, long-range interactions, leading to a rich phase diagram at zero temperature, e.g., with double N´eel-ordered phases. Furthermore, when the QEs are resonant with the bands (blue regions in Fig. 1B), we also find unusual dissipative dynamics. For example, for two equal QEs separated a given distance, we show that both the super/subradiance conditions [19] and the scat- tering properties depend on the parameter that governs the bath topology even though the energy dispersion ω(k) is insensitive to it. This might open avenues to probe the topology of these systems in unconventional ways, e.g., through reflection/transmission experiments. II. LIGHT-MATTER INTERACTIONS WITH ONE-DIMENSIONAL TOPOLOGICAL BATHS The system that we study along this manuscript is sketched in Fig. 1A: one or many QEs interact through a common bath which behaves as the photonic analogue of the SSH model [18]. This bath model is described by two interspersed photonic lattices A/B of size N with alter- nating nearest neighbour hoppings J(1±δ) between their photonic modes. Assuming periodic boundary conditions and defining V † = (a†k, b†k), the bath Hamiltonian can be k V † HB(k)V , with (setting  = 1): written in momentum space as HB = (cid:80) (cid:19) where f (k) = −J(cid:2)(1 + δ) + (1 − δ)e−ik(cid:3) = ω(k)eiφ(k) tween the A (B) modes, ak = (cid:80) (cid:80) j aje−ikj/√N (bk = j bje−ikj/√N ). Here a†j/b†j (aj/bj) are the creation (with ω(k) > 0) is the coupling in momentum space be- (cid:18) ωa f (k) f∗(k) ωa HB(k) = (1) , (annihilation) operators of the A/B photonic mode at the j'th unit cell. We assume that the A/B modes have the same energy, ωa, that from now on will be the reference energy of the problem, i.e., ωa ≡ 0. This Hamiltonian can be easily diagonalized introducing the eigenoperators, uk/lk =(cid:2) (cid:80) (cid:112) k ωk(u†kuk − l†klk), leading to two bands with energy: (2) (cid:3) /√2, as HB = 2(1 + δ2) + 2(1 − δ2) cos(k) . Let us now summarize the main bath properties: ±ak + eiφ(k)bk ± ω(k) = ±J • The bath possesses sublattice (chiral) symme- try [18], such that all eigenmodes can be grouped in chiral-symmetric pairs with opposite energies. Thus, the two bands are symmetric with respect to ωa, spanning from [−2J,−2δJ] (lower band) and [2δJ, 2J] (upper band). The middle gap is 2 4δJ, such that it closes when δ = 0, recovering the normal 1D tight-binding model. • This bath supports topologically non-trivial phases, belonging to the BDI class in the topological classi- fication of phases [20]. More concretely, both bands can be characterized by a topological invariant, the Zak phase [18] Z, such that Z = 0 corresponds to a trivial insulator, while Z = π implies a non-trivial insulator. For the parametrization we have chosen this occurs for δ > 0 and δ < 0, respectively. Notice that for an infinite system (i.e., in the bulk), this definition depends on the choice of the unit cell and the role of δ can be reversed by shifting the unit cell by one site. In the bulk the band topology man- ifests in the fact that one can not transform from one phase to the other without closing the gap (as long as the symmetry is preserved). • With finite systems, however, the sign of δ deter- mines whether the chain ends with weak/strong hoppings, which leads to the appearance (or not) of topologically robust edge states [21]. Now, let us finally describe the rest of the elements of our setup. For the Ne QEs, we consider they all have a single optical transition g-e, with a detuning ∆ respect to ωa, and they couple to the bath locally. Thus, their free and interaction Hamiltonian read: Ne(cid:88) (cid:88) (cid:0)σm egcxm + H.c.(cid:1) , σm ee , m=1 HS = ∆ HI = g (3) (4) m where cxm ∈ {axm , bxm} depends on the sublattice and the unit cell xm at which the m'th QEs couples to the bath. We use the notation σm µν = µ(cid:105)m(cid:104)ν, µ, ν ∈ {e, g}, for the m'th QE operator. We highlight that we use a rotating-wave approximation, such that only excitation- conserving terms appear in HI . Methods. In the next sections, we study the dy- namics emerging from the global QE-bath Hamiltonian H = HS + HB + HI using several complementary ap- proaches. When one is only interested in the QE dynam- ics, and the bath can be effectively traced out, the fol- lowing Born-Markov master equation [22] describes the evolution of the reduced density matrix ρ of the QEs: ρ = i[ρ, HS] + i J αβ mn egσn ge (cid:2)ρ, σm (cid:3) geρσm eg − σm egσn geρ − ρσm egσn ge (cid:3) . (5) (cid:88) (cid:2)2σn n,m (cid:88) n,m + Γαβ mn 2 mn, Γαβ The functions J αβ mn, which ultimately control the QE coherent and dissipative dynamics, respectively, are the real and imaginary part of the collective self-energy Σαβ 2 . This collective self-energy mn(∆ + i0+) = J αβ mn mn − i Γαβ depends on the sublattices α, β ∈ {A, B} to which the m'th and n'th QE couple respectively, as well as on their relative position xmn = xn − xm. Remarkably, for our model they can be calculated analytically in the thermo- dynamic limit (N → ∞) yielding: (cid:104) (cid:112) yxmn+ Θ+(y+) − yxmn − Θ−(y+) z4 − 4J 2(1 + δ2)z2 + 16J 4δ2 (cid:105) , ΣAA/BB mn (z) = − (6) g2J [Fxmn (y+)Θ+(y+) − Fxmn (y−)Θ−(y+)] , g2z (cid:112) ΣAB mn(z) = z4 − 4J 2(1 + δ2)z2 + 16J 4δ2 (7) where Fn(z) = (1 + δ)zn + (1 − δ)zn+1, Θ±(z) = Θ(±1 ∓ z), Θ(z) is Heaviside's step function, and (cid:104) (cid:112) z2 − 2J 2(1 + δ2) z4 − 4J 2(1 + δ2)z2 + 16J 4δ2 1 2J 2(1 − δ2) ± (cid:105) . (8) y± = However, since we have a highly structured bath this perturbative description will not be valid in certain regimes, e.g., close to band-edges, and we will use re- solvent operator techniques [23] or fully numerical ap- proaches to solve the problem exactly for infinite/finite bath sizes, respectively. Since those methods were ex- plained in detail in other works, here we focus on the results and leave the details for the Supp. Material. III. BAND-GAP REGIME In this Section we assume that the QEs are in the band- gap regime, that is, their transition frequency lies outside of the two bands of the photonic bath. From here on, we only discuss results in the thermodynamic limit (when N → ∞) such that the edge states [21] play no role in the QE dynamics. We refer the interested reader to Refs. [24] and Supp. Material to see some of the consequences the edge states have on the QE dynamics. A. Single QE: dynamics Let us start considering the dynamics of a single ex- cited QE, i.e., ψ(0)(cid:105) = e(cid:105)vac(cid:105), where vac(cid:105) denotes the vacuum state of the lattice of bosonic modes. Since H conserves the number of excitations, the global wavefunc- tion at any time reads: Ce(t)σeg + N(cid:88) (cid:88) j=1 α=a,b g(cid:105)vac(cid:105) . (9) Cj,α(t)α†j ψ(t)(cid:105) = In both perturbative and exact treatments, the dy- namics of Ce(t) can be shown (see Refs. [23] and Supp. 3 FIG. 2. Single QE dynamics. Probability to find the emitter in the excited state, Ce(t)2, for different values of δ. The other parameters are ∆ = 0 (middle of the band gap) and g = 0.4J. As the band gap closes, i.e., δ → 0, the decay becomes stronger. Dashed lines mark the value of Ce(t → ∞)2 =(cid:0)1 + g2 4J 2δ (cid:1)−2. (cid:112) Material) to depend only on the single QE self-energy: Σe(z) = g2z sign(y+ − 1) , (10) z4 − 4J 2(1 + δ2)z2 + 16J 4δ2 obtained from Eq. 6 defining: Σe(z) ≡ ΣAA nn (z). From i) here, we can already extract several conclusions: Σe(z) is independent of the sign of δ, which means that the spontaneous emission dynamics is insensitive to the topology of the bands. ii) Perturbative approaches, like the Born-Markov approximation of Eq. 5, predict an exponential decay of excitations at a rate Γe(∆) = −2ImΣe(∆ + i0+), which is strictly zero in the band-gap regime. Thus, one expects that the excitation remains localized in the QE at any time. However, in Fig. 2 we compute the exact dynamics Ce(t) for several δ's and ob- serve that this perturbative limit is only recovered in the limit of δ → 1. On the contrary, when δ (cid:28) 1 and δ (cid:54)= 0 the dynamics displays fractional decay and oscillations. As it happens with other baths [25], the origin of this dy- namics stems from the emergence of photon bound states (BSs) which localize around the QEs [26 -- 28]. However, the BSs appearing in the present topological waveguide bath have some distinctive features with no analogue in other systems, and therefore deserve special attention. B. Single QE: Bound states The energy and wavefunction of the BSs in the single- excitation subspace can be obtained by solving the secu- lar equation HΨBS(cid:105) = EBSΨBS(cid:105), with EBS lying out 020406080100time[J−1]0.00.20.40.60.81.0Ce(t)2δ=0.70.50.30.10.054 FIG. 3. Bound state properties. (A) BS wavefunction for a QE placed at j = 0 that couples to the A sublattice; δ = 0.2 and g = 0.4J. Probability amplitudes Cj,a are shown in blue, while the Cj,b are shown in red. The QE frequency is set to ∆ = 2.2J (top row), ∆ = 0 (middle row) and ∆ = −2.2J (bottom row). The first column corresponds to the model without disorder, the second to the model with disorder in the couplings between cavities and the third to the model with disorder in the cavities resonant frequencies. In both cases with disorder its strength is set to w = 0.5J. For each case, the value of the bound state's energy is shown at the bottom of the plots. (B) Inverse bound state localization length for the two different models of disorder as a function of the disorder strength. Parameters: g = 0.4J, δ = 0.5. The dots correspond to the average value computed with a total of 104 instances of disorder, and the error bars mark the value of one standard deviation above and below the average value (the blue curves are slightly displaced to the right for better visibility). Two cases are shown which correspond to ∆ (cid:39) 2.06J (triangles, outer band gaps), and ∆ = 0 (circles, inner band gap). (C) Absolute value of the dipolar coupling for ∆ = 0 and g = 0.4J; Markov (solid line), exact (dots). The insets show the shape of the bound states in the topological and the trivial phases. The situation for the BA configuration is the same, reversing the role of δ. of the bands, and ΨBS(cid:105) in the form of Eq. (9), but with time-independent coefficients. Without loss of gen- erality, we assume that the QE couples to sublattice A at the j = 0 cell. After some algebra, one can find that the energy of the BS is given by the pole equation: EBS = ∆ + Σe(EBS). Irrespective of ∆ or g, there exist always three BS solutions of the pole equation (one for each band-gap region). This is because the self-energy di- verges in all band-edges, which guarantees finding a BS in each of the band-gaps [29, 30]. The main difference with respect to other BSs [26 -- 30] appears in the wavefunction amplitudes, which read (cid:90) π (cid:90) π −π dk −π eikj dk E2 BS − ω2(k) ω(k)ei[kj−φ(k)] , E2 BS − ω2(k) , (11) (12) gEBSCe Cj,a = 2π gCe 2π Cj,b = where Ce is a constant obtained from the normalization condition that is directly related with the long-time pop- (cid:1)−2 . 4J 2δ ulation of the excited state in spontaneous emission. For example, in Fig. 2 where ∆ = 0, it can be shown to be Ce(t → ∞)2 = Ce4 =(cid:0)1 + g2 From Eqs. 11-12, we can extract several properties of the spatial wavefunction distribution. On the one hand, above or below the bands (outer band gaps) the largest contribution to the integrals is that of k = 0, thus all the Cj,α have the same sign (see left column of Fig. 3A top and bottom row). In the lower (upper) band-gap, Cj,α of the different sublattices has the same (opposite) sign. On the other hand, in the inner band gap, the main contribution to the integrals is that of k = π. This gives an extra factor (−1)j to the coefficients Cj,α (see Fig. 3A middle row). Furthermore, the probability amplitudes of the sublattice which the QE couples to are symmetric with respect to the position of the QE, whereas they are asymmetric in the other sublattice, that is, the BSs are chiral. Changing δ from positive to negative results in a spatial inversion of the BS wavefunction. The asymmetry ABCof the BS wavefunction is more extreme in the middle of the band-gap (∆ = 0). For example, if δ > 0, the BS wavefunction with EBS = 0 is given by Cj,a = 0 and  gCe(−1)j J(1+δ) 0 , (cid:17)j (cid:16) 1−δ 1+δ Cj,b = , j ≥ 0 j < 0 , (13) the usual behavior for 1D baths λBS ∼ 1/(cid:112) whereas for δ < 0 the wavefunction decays for j < 0 while being strictly zero for j ≥ 0. At this point, the BS decay length diverges as λBS ∼ 1/(2δ) when the gap closes. Away from this point, the BS decay length shows ∆edge, with ∆edge being the smallest detuning between the QE and the band-edge frequencies. (cid:17) HB → HB +(cid:80) HB → HB +(cid:80) The physical intuition of the appearance of such chiral BS at EBS = 0 is that the QE with ∆ = 0 acts as an effective edge in the middle of the chain, or equivalently, as a boundary between two semi-infinite chains with dif- ferent topology. This picture provides us with an insight useful to understand other results of the manuscript: de- spite considering the case of an infinite bath, the local QE-bath coupling inherits information about the under- lying bath topology. In fact, one can show that this chi- ral BS has the same properties as the edge-state which appears in a semi-infinite SSH chain in the topologi- cally non-trivial phase, for example, inheriting its ro- bustness to disorder. To illustrate it, we study the ef- fect of two types of disorder: one that appears in the cavities bare frequencies (diagonal), and another one that appears in the tunneling amplitudes between them (off-diagonal). The former corresponds to the addition of random diagonal terms to the bath's Hamiltonian and breaks the chi- ral symmetry of the original model, while the latter cor- responds to the addition of off-diagonal random terms and pre- serves it. We take the
ν,j, ν = a, b, 1, 2, from a uni- form distribution within the range [−w/2, w/2] for each j'th unit cell. To prevent changing the sign of the cou- pling amplitudes between the cavities, w is restricted to w/2 < (1 − δ) in the case of off-diagonal disorder. In the middle (right) column of Fig. 3A we plot the shape of the three BS appearing in our problem for a situation with off-diagonal (diagonal) disorder with w = 0.5J. There, we observe that while the upper and lower BS get modified for both types of disorder, the chiral BS has the same protection against off-diagonal disorder as a regular SSH edge-state: its energy is pinned at EBS = 0 as well as keeping its shape with no amplitude in the sublattice to which the QE couples to. On the contrary, for diagonal disorder the middle BS is not protected any more and may have weight in both sublattices.
1,jb†jaj +
2,ja†j+1bj + H.c.
a,ja†jaj +
b,jb†jbj (cid:16) (cid:16) (cid:17) j j Finally, to make more explicit the different behaviour with disorder of the middle BS compared to the other ones, we compute their localization length λBS as a func- tion of the disorder strength w averaging for many re- alizations. In Fig. 3B we plot both the average value 5 (markers) of λ−1 BS and its standard deviation (bars) for the cases of the middle (blue circles) and upper (pur- ple triangles) BSs. Generally, one expects that for weak disorder, states outside the band regions tend to delo- calize, while for strong disorder all eigenstates become localized (see, for example, Ref. [31]). In fact, this is the behaviour we observe for the upper BS for both types of disorder. However, the numerical results suggest that for off-diagonal disorder the chiral BS never delocalizes (on average). Furthermore, the chiral BS localization length is less sensitive to the disorder strength w manifested in both the large initial plateau region as well as the smaller standard deviations compared to the upper BS results. Summing up, a QE coupled locally to an SSH bath: i) localizes a photon only at one side of the emitter de- pending on the sign of δ, ii) with no amplitude in the sublattice where the QE couples to, iii) with the same properties as the topological edge states, e.g., robustness to disorder. As we discuss in more detail in the Supp. Material, the SSH bath is the simplest one-dimensional bath that provides all these features simultaneously. C. Two QEs Let us now focus on the consequences of such exotic BS when two QEs are coupled to the bath. For concreteness, we focus on a parameter regime where the Born-Markov approximation is justified, although we have performed an exact analysis in the Supp. Material. From Eq. 5, it is easy to see that in the band-gap regime, the interac- tion with the bath leads to an effective unitary dynamics governed by the following Hamiltonian: (cid:0)σ1 ge + H.c.(cid:1) . Hdd = J αβ 12 egσ2 (14) That is, the bath mediates dipole-dipole interactions be- tween the QEs. One way to understand the origin of these interactions is that the emitters exchange virtual photons through the bath, which in this case are local- ized around the emitter. In fact, these virtual photons are nothing but the photon BS that we studied in the previous Section. Thus, these interactions J αβ mn inherit many properties of the BSs. For example, the interac- tions are exponentially localized in space, with a localiza- tion length that can be tuned and made large by setting ∆ close to the band-edges, or fixing ∆ = 0 and letting the middle band-gap close (δ → 0). Moreover, one can also change qualitatively the interactions by moving ∆ to different band-gaps: for ∆ > 2J all the J αβ mn have the same sign, while for ∆ < 2δJ they alternate sign as xmn increases. Also, changing ∆ from positive to neg- ative changes the sign of J AA/BB , but leaves unaltered J AB/BA are insensitive to mn the bath's topology, the J AB/BA mimic the dimerization of the underlying bath, but allowing for longer range cou- plings. The most striking regime is again reached for ∆ = 0. In that case J AA/BB identically vanish, and thus . Furthermore, while J AA/BB mn mn mn mn the QEs only interact if they are coupled to different sub- lattices. Furthermore, in such a situation the interactions have a strong directional character, i.e., the QEs only in- teract if they are in some particular order. Assuming that the first QE at x1 couples to sublattice A, and the second one at x2 couples to B, we have (cid:17)x12 (cid:16) 1−δ 1+δ sign(δ) g2(−1)x12 0 Θ(δ) J(1+δ) g2 J(1+δ) J AB 12 = if δ · x12 > 0 . if δ · x12 < 0 . if x12 = 0 . (15) In Fig. 3C we plot the absolute value of the coupling for this case computed exactly, and compare it with the Markovian formula. Apart from small deviations at short distances, it is important to highlight that the directional character agrees perfectly in both cases. D. Many QEs: Spin models with topological long-range interactions One of the main interests of having a platform with BS-mediated interactions is to investigate spin models with long-range interactions [32, 33]. The study of these models has become an attractive avenue in quantum sim- ulation because long-range interactions are the source of non-trivial many-body phases [34] and dynamics [35], and are also very hard to treat classically. Let us now investigate how the shape of the QE in- teractions inherited from the topological bath translate into different many-body phases at zero temperature as compared to those produced by long-range interactions appearing in other setups such as trapped ions [34, 35], or standard waveguide setups. For that, we consider hav- ing Ne emitters equally spaced and alternatively coupled to the A/B lattice sites. After eliminating the bath, and adding a collective field with amplitude µ to control the number of spin excitations, the dynamics of the emitters (spins) is effectively given by: (cid:2)J AB mn (cid:0)σm,A eg σn,B (cid:88) m,n ge + H.c.(cid:1) (cid:0)σm,A µ 2 − (cid:1)(cid:105) , (16) z + σn,B z Hspin = ν denoting by σn,α , ν = x, y, z, the corresponding Pauli matrix acting on the α ∈ {A, B} site in the n'th unit cell. The J αβ mn are the spin-spin interactions derived in the previous subsection, whose localization length, denoted by ξ, and functional form can be tuned through system parameters such as ∆. For example, when the lower (upper) BS mediates the interaction, the J αβ mn has negative (alternating) sign for all sites, similar to the ones appearing in standard waveguide setups. When the range of the interactions is short (near- est neighbor), the physics is well described by the ferro- magnetic XY model with a transverse field [36], which 6 goes from a fully polarized phase when µ dominates to a superfluid one in which spins start flipping as µ decreases. In the case where the interactions are long- ranged the physics is similar to that explained in Ref. [34] for power-law interactions (∝ 1/r3). The longer range of the interactions tends to break the symmetry between the ferro/antiferromagnetic situations and leads to frustrated many-body phases. Since similar interactions also appear in other scenarios (standard waveguides or trapped ions), we now focus on the more different situation where the middle BS at ∆ = 0 mediates the interactions, such that the coefficients J AB mn have the form of Eq. (15). In that case, the Hamiltonian Hspin of Eq. 16 is very unusual: i) spins only interact if they are in different sublattices, i.e., the system is bipartite ii) the interaction is chiral in the sense that they interact only in case they are properly sorted: the one in lattice A to the left/right of that in lattice B, depending on the sign of δ. Note that δ also controls the interaction length ξ. In particular, for δ = 1 the interaction only occurs between nearest neighbors, whereas for δ → 0, the interactions become of infinite range. These interactions translate into a rich phase diagram as a function of ξ and µ, which we plot in Fig. 4A for a small chain with Ne = 20 emitters (obtained with exact diagonalization). Let us guide the reader into the different parts: • The region with maximum average magnetization (in white) corresponds to the places where µ dom- inates such that all spins are alligned upwards. • Now, if we decrease µ from this fully polarized phase in a region where the localization length is short, i.e., ξ ≈ 0.1, we observe a transition into a state with zero average magnetization. This be- haviour can be understood because in that short- range limit J AB mn only couples nearest neighbor AB sites, but not BA sites as shown in the scheme of the lower part of the diagram for δ > 0 (the oppo- site is true for δ < 0). Thus, the ground state is a product of nearest neighbor singlets (for J > 0) or triplets (for J < 0). This state is usually referred to as Valence-Bond Solid in the condensed matter literature [37]. Note, the difference between δ ≷ 0 is the presence (or not) of uncoupled spins at the edges. • However, when the bath allows for longer range in- teractions (ξ > 1), the transition from the fully polarized phase to the phase of zero magnetiza- tion does not occur abruptly but passing through all possible intermediate values of the magnetiza- tion. Besides, we also plot in Fig. 4B the spin-spin correlations along the x and z directions (note the symmetry in the xy plane) for the case of µ = 0 to evidence that a qualitatively different order ap- pears as ξ increases. In particular, we show that the spins align along the x direction with a dou- ble periodicity, which we can pictorially represent 7 FIG. 4. Spin models: phase diagram and correlations. (A) Ground state average polarization obtained by exact diagonalization for a chain with Ne = 20 emitters with frequency tuned to ∆ = 0 as a function of the chemical potential µ and the decay length of the interactions ξ. The different phases discussed in the text, a Valence-Bond Solid (VBS) and a Double N´eel ordered phase (DN) are shown schematically below, on the left and right respectively. Interactions of different sign are marked with links of different color. For the VBS we show two possible configurations corresponding to δ < 0 (top) and δ > 0 (bottom). In the topologically non-trivial phase (δ < 0) two spins are left uncoupled with the rest of the chain. (B) Correlations Cν (r) = (cid:104)σ9 (cid:105), ν = x, y, z [Cx(r) = Cy(r)] for the same system as in (A) for different interaction lengths, fixing µ = 0 (left column). Correlations for different chemical potentials fixing ξ = 5, darker colors correspond to lower chemical potentials (right column). Note we have defined a single index r that combines the unit cell position and the sublattice index. The yellow dashed line marks the value of 1/2 expected when the interactions are of infinite range. ν(cid:105)(cid:104)σ9+r (cid:105) − (cid:104)σ9 ν σ9+r ν ν by ↑↑↓↓↑↑ . . .(cid:105)x, and that we label as double N´eel order states. Such orders have been predicted as a consequence of frustration in classical and quan- tum spin chains with competing nearest and next- nearest neighbour interactions [38 -- 40], introduced to describe complex solid state systems such as multiferroic materials [41]. In our case, this order emerges in a system which has long-range inter- actions but no frustration as the system is always bipartite regardless the interaction length. To gain analytical intuition of this regime, we take the limit ξ → ∞, where the Hamiltonian (16) reduces to z A/B =(cid:80) eg H(cid:48)spin = U HspinU† (cid:39) J(S+ A S−B + H.c.) , (17) n∈Zodd where S+ σn,A z σn,B tary transformation U = (cid:81) n σn,A/B , and we have performed a uni- , to cancel the alternating signs of J AB mn . Equality in Eq. (17) occurs for a system with periodic boundary conditions, while for finite systems with open boundary conditions some corrections have to be taken into account due to the fact that not all spins in one sublattice couple to all spins in the other but only to those to their right/left depending on the sign of δ. The ground state is symmetric under (independent) permutations in A and B. In the thermo- dynamic limit we can apply mean field, which predicts symmetry breaking in the spin xy plane. For instance, if J < 0 and the symmetry is broken along the spin direc- tion x, the spins will align so that (cid:104)(Sx B)2(cid:105) = (cid:104)Sx A)2(cid:105) = (cid:104)(Sx B(cid:105)2 = (Ne/2)2 . B(cid:105) = (Ne/2)2, and (cid:104)Sx A(cid:105)2 = (cid:104)Sx ASx Since Ne is finite in our case, the symmetry is not broken, but it is still reflected in the correlations, so that (cid:104)σm,A ν σn,A ν (cid:105) (cid:39) (cid:104)σm,A ν σn,B ν (cid:105) (cid:39) 1/2 , (18) with ν = x, y. In the original picture with respect to U , we obtain the double N´eel order observed in Fig. 4B. As can be understood, the alternating nature of the inter- actions is crucial for obtaining this type of ordering. Fi- nally, let us mention that the topology of the bath trans- lates into the topology of the spin chain in a straightfor- ward manner: regardless the range of the effective inter- actions, the ending spins of the chain will be uncoupled to the rest of spins if the bath is topologically non-trivial. This discussion shows the potential of the present setup to act as a quantum simulator of exotic many-body phases not possible to simulate with other known setups. The full characterization of such spin models with topo- logical long-range interactions is interesting on its own and we will present it elsewhere. IV. BAND REGIME Here, we study the situation when QEs are resonant with one of the bands. For concreteness, we only present two results where the unconventional nature of the bath plays a prominent role, namely, the emergence of unex- pected super/subradiant states, and their consequences when a single-photon scatters into one or two QEs. ABA. Dissipative dynamics: super/subradiance The band regime is generally characterized by inducing non-unitary dynamics in the QEs. However, when many QEs couple to the bath there are situations in which the interference between their emission may enhance or sup- press (even completely) the decay of certain states. This phenomenon is known as super/subradiance [19], respec- tively, and it can be used, e.g., for efficient photon stor- age [42] or multiphoton generation [43]. Let us illustrate this effect with two QEs: In that case, the decay rate of a symmetric/antisymmetric combination of excitations is Γe ± Γ12. When Γ12 = ±Γe, these states decay at a rate that is either twice the individual one or zero. In this lat- ter case they are called perfect subradiant or dark states. In standard one-dimensional baths Γ12(∆) = (cid:1), so the dark states are such that Γe(∆) cos(cid:0)k(∆)xmn the wavelength of the photons involved, k(∆), allows for the formation of a standing wave between the QEs when both try to decay, i.e., when k(∆)xmn = nπ, with n ∈ Z. Thus, the emergence of perfect super/subradiant states solely depends on the QE frequency ∆, bath energy dispersion ω(k), and their relative position xmn, which is the common intuition for this phenomenon. This common wisdom gets modified in the bath con- sidered along this manuscript, where we find situations in which, for the same values of xmn, ω(k) and ∆, the induced dynamics is very different depending on the sign of δ. In particular, when two QEs couple to the A/B sublattice respectively, the collective decay reads: 12 (∆) = Γesign(∆) cos(cid:0)k(∆)x12 − φ(∆)(cid:1) , (cid:104) ∆2−2J 2(1+δ2) (cid:105) which depends both on the photon wavelength mediating the interaction k(∆) = arccos , an even function of δ, and on the phase φ(∆) ≡ φ(k(∆)), sensitive to the sign of δ. This φ-dependence enters through the system-bath coupling when rewriting HI in Eq. 4 in terms of the eigenoperators uk, lk. The intuition behind it is that even though the sign of δ does not play a role in 2J 2(1−δ2) ΓAB (19) the bath properties of an infinite system, when the QEs couple to it, the bath embedded between them is different for δ ≷ 0, making the two situations inequivalent. 8 Using Eq. 19, we find that to obtain a per- fect a super/subradiant state it must be satisfied: k(∆s)x12 − φ(∆s) = nπ, n ∈ N. They come in pairs: If ∆s is a superradiant (subradiant) state in the upper band, −∆s is a subradiant (superradiant) state in the lower band. In particular, it can be shown that when δ < 0, the super/subradiant equation has solutions for n = 0, . . . , x12, while if δ > 0, the equation has solutions for n = 0, . . . , x12 + 1. Besides, the detunings, ∆s at which the subradiant states appear also satisfy that J AB 12 (∆s) ≡ 0, which guarantees that these subradiant states survive even in the non-Markovian regime (with a correction due to retardation which is small as long as x12Γe(∆)/(2vg(∆)) (cid:28) 1). Apart from inducing different decay dynamics, these different conditions for super/subradiance at fixed ∆ also translate in different reflection/transmission coefficients when probing the system through photon scattering, as we show next. B. Single-photon scattering The scattering properties of a single photon imping- ing into one or several QEs in the ground state can be obtained by solving the secular equation with energies HΨk(cid:105) = ±ωkΨk(cid:105), with the ± sign depending on the band we are probing [44]. Here, we focus on the study of the transmission amplitude t (see scheme of Fig. 5A) for two different situations: i) a single QE coupled to both cavity A and cavity B in the same unit cell with coupling constants gα and g(1 − α), such that we can interpolate between the case where the QE couples only to sublattice A (α = 1) or B (α = 0), ii) and a pair of emitters in the AB configuration separated x12 unit cells. After some algebra, we find the exact formulas for the transmission coefficients for the two situations: 2iJ(1 − δ) sin(k)(cid:2)J(1 + δ)(±ωk − ∆) − g2α(1 − α)(cid:3) kei2(kx12−φ) − [g2ωk ± i2J 2(1 − δ2)(±ωk − ∆) sin(k)]2 . (cid:2)2J 2(1 − δ2)(±ωk − ∆) sin(k)(cid:3)2 2iJ 2(1 − δ2)(±ωk − ∆) sin(k) + g2ωk [2α(1 − α)(e−iφ ∓ 1) ± 1] g4ω2 t1QE = t2QE = , (20) (21) In Fig. 5B, we plot the single-photon transmission probability t2 as a function of the frequency of the in- cident photon for the single (left) and two QE (right) situations. Let us now explain the different features ob- served: Single QE: We first plot in dashed orange the results for α = 0, 1, showing well known features for this type of systems [44], namely, a perfect transmission dip (t2 = 0) when the frequency of the incident photon matches exactly that of the QEs. This is because the Lamb-shift induced by the bath in this situation is δωe = 0. The dip has a band-width defined by the individual decay 9 12 in the single QE scenario since now the responses are also qualitatively different: While the case δ > 0 fea- tures a single transmission dip at the QEs frequency, for δ < 0 the transmission dip is followed by a win- dow of frequencies with perfect photon transmission, i.e, t2QE2 = 1. A convenient picture to understand this behaviour is depicted in Fig. 5A, where we show that a single photon only probes the symmetric/antisymmetric states in the single excitation subspace (S/A) with the following energies (linewidths) renormalized by the bath ωS,A = ∆ ± J AB (ΓS,A = Γe ± Γ12). For the pa- rameters chosen (see caption) it can be shown that for δ > 0 the QEs are in a perfect super/subradiant config- uration in which one of the states decouples while the other one has 2Γe decay rate. Thus, at this configura- tion the two QEs behave like a single two-level system with an increased linewidth. On the other hand, when δ < 0 both the (anti)symmetric states are coupled to the bath, such that the system is analogous to a V-type system where perfect transmission occurs for an incident frequency ±ωk,EIT = (ωSΓA−ωAΓS)/(ΓA−ΓS) [45] (de- picted in dashed black). In both the single and two QE situations the different response can be intuitively understood as the QEs cou- ple locally to a different bath for δ ≷ 0. However, this different response of t2 can be thought as an indirect way of probing topology in these systems. V. IMPLEMENTATIONS One of the attractive points of our predictions is that they can be potentially observed in several platforms by combining tools which, in most of the cases, have already been experimentally implemented independently. Some candidate platforms are: • Photonic crystals. The photonic analogue of the SSH model has been implemented in several pho- tonic platforms [6, 10 -- 12], including some recent photonic crystal realizations [7]. The latter are particularly interesting due to the recent advances in their integration with solid-state and natural atomic emitters (see Refs. [46, 47] and references therein). • Circuit QED. Superconducting metamaterials mimicking standard waveguide QED are now be- ing routinely built and interfaced with one or many qubits in experiments [48, 49]. The only missing piece is the periodic modulation of the couplings to obtain the SSH model, for which there are already proposals using circuit superlattices [50]. • Cold-atoms. Quantum optical phenomena can be simulated in pure atomic scenarios by using state- dependent optical lattices. The idea is to have two different trapping potentials for two atomic metastable states, such that one state mostly local- izes, playing the role of QEs, while the other state (cid:1) /√2 e(cid:105)1g(cid:105)2 ± g(cid:105)1e(cid:105)2 mon ground state, while S, A(cid:105) = (cid:0) FIG. 5. Single-photon scattering. (A) Pictorial represen- tation of the scattering process: An incident photon impinges into a scatterer, part of which is reflected (transmitted) with probability amplitude r (t). Lower row: Relevant level struc- ture for the single photon scattering for both scatterers con- sidered: one and two QEs. gg(cid:105) ≡ g(cid:105)1g(cid:105)2 denotes the com- denotes the symmetric (antisymmetric) excited state combi- nation of the two QEs. (B) Transmission probability for a single emitter coupled to both A and B cavities of the same unit cell (left panel) and two emitters in the AB configura- tion separated a total of x12 = 2 unit cells (right panel). The parameters in the single emitter case are: g = 0.4J, δ = ±0.5 and ∆ = 1.5J. The dashed line corresponds to the case where the emitter couples to a single sublattice (α = 0, 1) (does not depend on the sign of δ). When the emitter couples to both sublattices (α = 0.3), the perfect-reflection resonance experi- ences a shift that is different for δ > 0 (purple line) or δ < 0 (blue line). The parameters for the two emitter case are: g = 0.1J, δ = ±0.5, and ∆ (cid:39) 1.65J for which the two QE are in a subradiant configuration if δ > 0. rate Γe. Besides, it also shows t2 = 0 at the band- edges due to the divergent decay rate at these frequencies, also predicted for standard waveguide setups [44]. The situation becomes more interesting for 0 < α < 1, since the QE energy is shifted by δωe = g2α(1− α)/[J(1 + δ)], which is different for ±δ. This is why the dips in t1QE2 appear at different frequencies for δ = ±0.3. Notice t1QE is invariant under the transformation α → 1 − α (this is not true for the reflection coefficient, which acquires a δ-dependent phase shift for α = 0 but not for α = 1). Two QEs: In the right panel of Fig. 5B we plot t2QE2 for two QEs coupled equally to a bath (same energy, dis- tance, and coupling strength), and where the only dif- ference is the sign of δ of the bath. The distance cho- sen is small such that retardation effects do not play a significant role. The differences between δ > 0 and δ < 0 in the t2QE2 are even more pronounced that AScatterertrinB10 VII. ACKNOWLEDGEMENTS Funding: This work has been supported by the Span- ish Ministry of Economy and Competitiveness through grants No. MAT2017-86717-P and BES-2015-071573. AGT and JIC acknowledge the ERC Advanced Grant QENOCOBA under the EU Horizon 2020 program (grant agreement 742102). MB, GP, and AGT acknowl- edge support from CSIC Research Platform on Quantum Technologies PTI-001. Competing Interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data available from authors upon request. Author con- tributions: MB and AGT conceived the original idea, MB did the analytical and numerical analysis under the supervision of AGT. All authors discussed and analyzed the results. MB and AGT wrote the manuscript with input from all authors. propagates as a matter-wave. This proposal [51] has been recently used [52] to explore the physics of standard waveguide baths. Replacing their po- tential by an optical superlattice made of two laser fields with different frequencies, one would be able to probe the physics of the topological SSH bath. In fact, these cold-atoms superlattices have already been implemented in an independent experiment to measure the Zak phase of the SSH model [53]. Beyond these platforms, the bosonic analogue of the SSH model has also been discussed in the context of metamaterials [54] or plasmonic and dielectric nanopar- ticles [55, 56], where the predicted phenomena could as well be potentially observed. VI. CONCLUSIONS & OUTLOOK Summing up, we have presented several phenomena appearing in a topological waveguide QED system with no analogue in other optical setups. When the quantum emitter frequencies are tuned to the middle band-gap, we predict the appearance of chiral photon bound states which inherit the topological robustness of the bath. Fur- thermore, we also show how these bound states mediate directional, long-range spin interactions, leading to ex- otic many-body phases, e.g., double-N´eel ordered states, which cannot be obtained to our knowledge with other bound-state mediated interactions. Besides, we study the scattering and super/subradiant behaviour when one or two emitters are resonant with one of the bands, finding that transmission amplitudes can depend on the param- eter which controls the topology even though the band energy dispersion is independent of it. Except for the many-body physics, the rest of the phe- nomena discussed in this article, that is, the formation of chiral bound states and the peculiar scattering proper- ties, could also be observed in classical setups, since these results are derived within the single-excitation regime. Given the simplicity of the model and the variety of plat- forms where it can be implemented, we foresee that our predictions can be tested in near-future experiments. As an outlook, we believe our work opens complemen- tary research directions on topological photonics, which currently focuses more on the design of exotic light prop- erties [10 -- 12, 57, 58]. For example, the study of the emer- gent spin models with long-range topological interactions is interesting on its own and might lead to the discovery of novel many-body phases. Moreover, the scattering- dependent phenomena found along the manuscript can provide alternative paths for probing topology in pho- tonic systems. On the fundamental level, the analytical understanding we develop for one-dimensional systems provides a solid basis to understand quantum optical ef- fects in higher dimensional topological baths [59, 60]. Supplemental Material: Unconventional quantum optics in topological waveguide QED. 11 In this Supp. Material, we provide more details on: i) the exact integration of the quantum emitter (QE) dynamics using resolvent operator techniques, in Sec- tion S1; ii) the study of asymptotic long-time decay, in Section S2; iii) the exact integration of the two QEs dy- namics, in Section S3; iv) the derivation of the exact conditions of existence of two QE bound states, in Sec- tion S4; v) the effect of the edge states on the QE dy- namics when they are coupled to finite size baths, in Sec- tion S5; vi) a review of bipartite one-dimensional baths and the properties of the middle bound-states, in Sec- tion S6. S1. INTEGRATION OF THE DYNAMICS FIG. S1. Schematics showing the contour of integration. At the band edges the path changes from the first to the second Riemann sheet of the Green function (shaded areas). should subtract the detours taken at the branch cuts. Their contribution can be computed as CBC,j(t) = ± 1 2π× (cid:104) (cid:90) ∞ 0 (cid:105) dy GI e(xj − iy) − GII e (xj − iy) e−i(xj−iy)t , (S3) with xj ∈ {±2J,±2δJ}. The sign has to be chosen positive if when going from xj + 0+ to xj − 0+ the inte- gration goes from the first to the second Riemann sheet, and negative if it is the other way around. Plotting the absolute value of the different contribu- tions at time t = 0, we can deduce the relevant physics involved in the QE dynamics, see Fig. S2(a). Not surpris- ingly, when the emitter's transition frequency lays in the bands of allowed bath modes it will decay emitting a pho- ton into the bath. In Fig. S2(b), we compare the actual decay rate with the prediction given by the Markovian approximation. On the other hand, when it lays outside the bands, a bound state will form in which the emit- ter is mostly in the excited state and part of the photon remains trapped around it. This is what we observe in Fig. 2 in the main text, where the long-term dynamics is dominated by the bound state at zero energy, whose residue can be computed as R0 = [1 + g2/(J 24δ)]−1. Since the global Hamiltonian H conserves the number of excitations, if a QE is initially excited with no pho- tons in the bath, i.e., ψ(0)(cid:105) = e(cid:105)vac(cid:105) (vac(cid:105) denotes the vacuum state of the lattice of bosonic modes), the wavefunction at any time has the form: Ce(t)σeg + N(cid:88) (cid:88) j=1 α=a,b ψ(t)(cid:105) = g(cid:105)vac(cid:105) . Cj,α(t)α†j (S1) The probability amplitude Ce(t) can be computed [23, 61] as the Fourier transform of the Green function of the emitter Ge(z) = [z − ∆ − Σe(z)]−1: dE Ge(E + i0+)e−iEt , (S2) (cid:90) ∞ −∞ Ce(t) = −1 2πi To compute the integral in (S2), we use residue integra- tion closing the contour of integration in the lower half of the complex plane. Since the QE Green function has branch cuts in the real axis along the regions where the bands of the bath are defined (the continuous spectrum of H), it is necessary to detour at the band edges to other Riemann sheets of the function, see Fig. S1. The formula for the self-energies presented in the main text, Eq. (10), corresponds to the Green function in the first Riemann sheet GI e(z). We can analytically continue it to the second Riemann sheet GII e (z) by simply changing √· → −√· in the denominator of Σe(z). e(E + i0+) and GII Since the imaginary part of GI e (E − i0+) is nonzero in the band regions, we should only take into account the real poles of GI e outside the band re- gions (zBS) and the complex poles of GII e with real part inside band regions (zUP). The residue at both the real and complex poles can be computed as R(z0) = [1 − Σ(cid:48)e(z0)]−1, where Σ(cid:48)e(z) denotes the first derivative of the appropriate function ΣI e (z). Finally, we e(z) or ΣII 12  y1/2e−yt i(2 − x2 0 + 2δ2) x0 + O(y) i(2 − x2 0 + 2δ2) x0 t−3/2 + O(t−5/2) . (S7) series around y = 0, (cid:115)  4 (cid:115) g2 K0(t) = ±1 2π dy (cid:90) ∞ (cid:39) ±1(cid:112) 0 πg2 Therefore, to leading order D(t)2 ∼ t−3. In Fig. S3 it is shown an example of this algebraic decay when ∆ is placed at the lower band edge of the bath's spectrum. FIG. S3. Decaying part of the dynamics of a single emitter with parameters ∆ = −2J, δ = 0.5 and g = 0.2J. eg ± σ2 eg ± = (cid:2)σ1 (cid:3)/√2 couple to orthogonal The dynamics of two emitters are not much harder to analyze than that of a single emitter. It can be shown that the symmetric and antisymmetric combi- nations σ† bath modes [62]. Therefore, the two-emitter problem can be split in two independent single-emitter problems. The Green functions associated to the probability am- plitudes to find the 1st or the 2nd emitter in the ex- cited state C1,2(t) can be obtained form the Green func- tions associated to the symmetric/antisymmetric com- bination of excitations as G1,2(z) = [G+(z) ± G−(z)]/2, with G±(z) = [z − ∆ − Σ±(z)]−1. Rewriting the interaction Hamiltonian in the bath's eigenmode basis, substituting σm ±, and pairing the terms that go with opposite momentum, we obtain for the case where the two QEs are on the sublat- tice A eg in terms of σ† Kj(t)e−ixj t , (S4) S3. TWO QE DYNAMICS IN THE NON-MARKOVIAN REGIME FIG. S2. Non-Markovian dynamics. (a) Absolute value of the different contributions to the single QE dynamics at time t = 0 as a function of the emitter detuning ∆; bound state residues R(zBS) (circles), unstable pole residues R(zUP) (squares) and branch-cut contributions CBC,j(0) (crosses). The system parameters are δ = 0.5 and g = 0.4J. (b) Com- parison between the exact decay rate given by the imaginary part of the complex poles of Ge (diamonds) and the approxi- mate Markovian decay rate (black lines) for the same param- eters as in (a) Defining D(t) ≡ Ce(t) − times we have zBS R(zBS)eizBSt, at long S2. SUB-EXPONENTIAL DECAY (cid:80) (cid:88) (cid:88) lim t→∞ D(t) (cid:39) CBC,j(t) = j j with Kj(t) = ±1 2π (cid:90) ∞ dy 0 2Σe(xj − iy)e−yt (xj − iy − ∆)2 − Σ2 e(xj − iy) . (S5) The long-time average of the decaying part of the dy- namics can be computed as (cid:90) t 0 (cid:88) j D(t)2 ≡ lim t→∞ 1 t dt(cid:48)D(t(cid:48))2 = Kj(t)2 . (S6) If the emitter's transition frequency is close to one of the band edges, ∆ (cid:39) x0, then D(t)2 (cid:39) K0(t)2. In the long-time limit, we can expand the integrand in power 00.20.40.60.81(a)00.20.4−3−2−10123(b)R(zj),Cj,BC(0)00.20.40.60.81Γe∆/J00.20.4−3−2−1012310−810−610−410−2101102103∝t−3D(t)2time[J−1]10−810−610−410−21011021031 + β cos(kx12)(uk,β + lk,β)σ†β + H.c. , k>0 (cid:112) (cid:88) (cid:88) g ei(kx1+φ) ± ei(kx2+φ)(cid:17) (cid:104)(cid:16) √N ei(kx1+φ) ± ei(kx2+φ)(cid:17) (cid:104)(cid:16) β=± (cid:112) (cid:112) 2 2 uk,± = lk,± = H AA I = 1 1 ± cos(kx12) 1 1 ± cos(kx12) (cid:88) (cid:104)(cid:112) β=± 1 k>0 (cid:88) (cid:112) (cid:112) (cid:17) 1 + β cos(kx12 − φ) uk,βσ†β + (cid:17) (cid:104)(cid:16) (cid:104)(cid:16) ei(kx1+φ) ± eikx2 ei(kx1+φ) ∓ eikx2 1 ± cos(kx12 − φ) 1 1 ∓ cos(kx12 − φ) H AB I = g √N uk,± = lk,± = 2 2 uk + lk + e−i(kx1+φ) ± e−i(kx2+φ)(cid:17) (cid:16) e−i(kx1+φ) ± e−i(kx2+φ)(cid:17) (cid:16) (cid:105) (cid:112) (cid:17) (cid:16) 1 − β cos(kx12 − φ) lk,βσ†β (cid:17) (cid:16) e−i(kx1+φ) ± e−ikx2 e−i(kx1+φ) ∓ e−ikx2 lk + uk + Here, xn refers to the unit cell where the n'th QE is located, and x12 = x2 − x1 is the signed distance between the two QEs. For the case where the two QEs are on a different sublattice 13 (S8) (S9) (S10) , (cid:105) (cid:105) u−k l−k , + H.c. , (cid:105) (cid:105) u−k , l−k . (S11) (S12) (S13) (cid:88) The prefactors in the definition of uk,± and lk,± come Importantly, these modes are or- from normalization. thogonal, they satisfy [uk,α, u†k(cid:48),α(cid:48)] = [lk,α, l†k(cid:48),α(cid:48)] = δkk(cid:48)δαα(cid:48) . (S14) Since ω(k) = ω(−k), we have that the bath Hamilto- nian is also diagonal in this new basis. The two other configurations can be analyzed analogously. From these expressions for the interaction Hamilto- nian, it is possible to obtain the self-energy for the sym- metric/antisymmetric states of the two QE. As it turns out, they have a very simple form: Σαβ 12 , with Σαβ 12 : ± = Σe±Σαβ ΣAA/BB mn (z; xmn) = ΣAB mn(z; xmn) = g2 N g2 N zeikxmn z2 − ω2(k) ω(k)ei[kxmn−φ(k)] k (cid:88) k z2 − ω2(k) , , (S15) (S16) where xmn = xn − xm. It can be shown that ΣBA mn(z; δ, xmn) = ΣAB nm(z; δ,−xmn) = ΣAB mn(z;−δ, xmn − 1) . (S17) S4. EXISTENCE CONDITIONS OF TWO QE BOUND STATES We can integrate the dynamics in the same way as we did for the single QE case, but there are some subtleties particular to the two QE case. First, the cancellation of divergences of Σe and Σαβ 12 at some of the band edges results in critical transition frequencies above (or below) (cid:27) (cid:26) which some bound states cease to exist. For example, in the symmetric subspace we have that the lower bound state (EBS < −2J) always exists, while the upper bound state (EBS > 2J) exists only for ∆ > ∆out , c ∆out c = 2J − g2(2x12 + 1 − δ) 2J(1 − δ2) . (S18) For the middle bound state there are two possibilities: either the divergence vanishes at −2δJ, in which case the bound state will exist for ∆ > ∆mid , or the divergence vanishes at 2δJ, then the middle bound state exists for ∆ < ∆mid . In both cases ∆mid takes the same form c c c c = (−1)x12 ∆mid 2δJ + g2[(2x12 + 1)δ − 1] 2J(1 − δ2) , (S19) The situation in the antisymmetric subspace can be readily understood realizing that ReΣAB − (z) = −ReΣAB + (−z), which implies that if z = EBS is a solu- tion of the pole equation for Σαβ + with a particular value of ∆, then z = −EBS is a solution of the pole equation for Σαβ − with the opposite value of the emitter transition frequency. Fig. S4 summarizes at a glance the different possibilities and the dependence on the bath's topology. S5. FINITE-BATH DYNAMICS It is well known that a finite bath with open boundary conditions in the topologically non-trivial phase (δ < 0) supports a pair of edge states ES±(cid:105), with opposite ener- gies HBES±(cid:105) = ±
ES±(cid:105), given by
(cid:39) J(1 − δ)e−N/λ. These states are exponentially localized at the edges of the bath with the same localization length λ as the BSs at 14 nentially localized state in one of the ends of the chain. Due to this, the photon oscillates between the QE and the edge whose ending mode is in the sublattice to which the QE is coupled [see Fig. S5(a)]. The oscillation fre- quency given by the effective model overestimates the actual frequency, which can be calculated exactly using the resolvent operator formalism. We do so by extending the bath, adding the two edge states, which are orthogo- nal to all other bath modes. The emitter Green function becomes now Ge = z2 −
2 (z − ∆ − Σe)(z2 −
2) − 2g2z . (S22) The long-term dynamics is given just by the real poles of this modified Green function. In particular, for ∆ = 0 the denominator is and odd function with three real roots around the middle of the band gap: z = 0 and z = ±ω0. It can be shown that the largest contribution to the dynamics comes from these real poles, such that Ce(t) (cid:39) R0 + 2R+ cos(ω0t), where R0 denotes the residue at the pole z = 0, and R+ = R− is the residue at the poles z = ±ω0. In Fig. S5(d, e), we show the QE population dy- namics when two QEs are coupled to the A/B lattices symmetrically with respect to the middle of the chain, and for two different situations, i.e., with fixed δ = 0.3 but different sign. As in the individual behaviour, the collective dynamics is very different depending on the topological nature of the bath. In the topologically trivial bath, the BS mediates perfect coherent transfer of excitations between the two QEs [see Fig. S5(d)]. In the topologically non-trivial bath, however, the edge states become largely populated since they are quasi-resonant with the QE oscillation, leading to additional oscillatory behaviour. Interestingly, perfect coherent transfer is still possible at certain times [see Fig. S5(e)], even though the induced dipolar coupling is zero. This dynamics can again be captured by a simple effective Hamiltonian, which written in the basis {e1(cid:105)g2(cid:105)vac(cid:105),g1(cid:105)e2(cid:105)vac(cid:105),g1(cid:105)g2(cid:105)ES+(cid:105),g1(cid:105)g2(cid:105)ES−(cid:105)} reads  ∆ J AB J AB 12 g g g 12 g ∆ g −g 0 g −g 0 −

 , FIG. S4. Bound states for the symmetric (continuous) and the antisymmetric (dashed) subspaces as a function of the QEs transition frequency in the topological (a) and the trivial (b) cases; g = 0.8J and x12 = 0. The shaded areas mark the bath's band regions, where no bound states can be found. zero energy mentioned in the main text. So far, our cal- culations have been done in baths large enough such that the contributions of the topological edge-states could be neglected. In this section, we consider the effect they can have in systems with moderate sizes. In Fig. S5(a -- c) we compare the dynamics of an initially excited QE coupled to a finite bath (N = 40) in the topologically non-trivial and trivial phases with the same δ = 0.3. The induced dynamics is very different: while most of the QE excitation remains localized around the QE for a topologically trivial bath, in the non-trivial case the QE exchanges non-locally the excitation with one of the edges of the bath. This emergent dynamics can be captured by a simple effective Hamiltonian considering only the excited state of the QE and the two edge states (with the QE in the ground state):  ∆ g+ g− g+
0 g− 0 −
 , Heff = (S20) Heff = (S23) written here in the basis {e(cid:105)vac(cid:105),g(cid:105)ES+(cid:105),g(cid:105)ES−(cid:105)}. The coupling constants are g± = g(cid:104)ES±c†x1vac(cid:105) (c†x1 is equal to a†x1 or b†x1 depending on the sublattice to which the emitter is coupled) and satisfy g− = g+ ≡ g. Ex- actly when ∆ = 0, the QE couples more strongly to the edge states. In that case, the excited-state probability amplitude can be computed as
2 + 2g2 cos(ω0t) Ce(t) (cid:39)
2 − 2g2. Note that a (anti)symmetric su- with ω0 = perposition of the edge states corresponds to an expo-
2 + 2g2 (S21) , (cid:112) using the definitions of the edge-states and the coupling constants g± for each QE that we use in the single QE dynamics. Solving this Hamiltonian with ∆ = 0, and assuming
(cid:28) g, the excited state occupation probability of the 1st (2nd) emitter can be well approximated by: C1(t) ≈ cos(
t/2) cos(cid:0)√2gt(cid:1) , C2(t) ≈ sin(
t/2) cos(cid:0)√2gt(cid:1) . (S24) (S25) which captures qualitatively the double oscillatory be- haviour of Fig. S5(e). In order to quantitatively capture −303−303−303−4−3−2−101234−303−4−3−2−101234EBS[J]−303(a)δ=−0.5∆outc∆midcEBS[J]−303(a)δ=−0.5∆outc∆midcEBS[J]∆[J]−303−4−3−2−101234(b)δ=0.5∆outc∆midcEBS[J]∆[J]−303−4−3−2−101234(b)δ=0.5∆outc∆midc15 FIG. S5. Finite-size effects. (Left panel) Bath dynamics in the topological, δ = −0.3, (a) and trivial, δ = 0.3 (b) regimes, for a lattice with N = 40 unit cells and open boundary conditions. A single QE with ∆ = 0 is coupled with strength g = 0.2J to the middle of the bath. The color shows the probability to find the photon in each site of the lattice. Brighter colors correspond to a higher probability. We have used a different logarithmic scale in each case for clarity. Below (c), it is shown the probability to find the emitter in the excited state for both the topological (blue) and trivial (red) cases. The dashed black line is a cosine function with frequency 2ω0, as obtained by a more precise treatment using Green functions. (Right panel) Dynamics for two QEs coupled to the A/B lattices respectively, placed symmetrically around the middle of the bath (∆ = 0) with parameters N = 10, g = 0.1J, x12 = 3, and δ = 0.3 (d), δ = −0.3 (e). The dashed black line is a cosine with a frequency obtained from the exact treatment with Green functions. the frequencies of the transfer exactly, one can use resol- vent operator techniques, which yields the dashed black line of Fig. S5(e). In this case, the extended Green func- tions are given by G± = (z − ∆ − Σαβ z ±
± )(z ±
) − 2g2 . (S26) For ∆ = 0, the real poles of G+ around the middle of the band gap, z±, are the same as those of G− with opposite sign. The residues are the same in both cases, therefore C±(t) (cid:39) R+e±iz+t + R−e±iz−t. Since C1,2(t) = [C+(t) ± C−(t)]/2, the relevant frequencies are ω± = z+ ± z−. It should be noted, however, that for really small systems or situations in which the emitters are placed close to the edges, the results given by these modified Green functions will not be accurate, as they use the thermodynamic self-energies Σαβ ± , which are obtained for infinite systems. S6. MIDDLE BOUND STATES IN ONE-DIMENSIONAL BATHS appears in the middle band-gap when an emitter with energy ∆ couples to the bath. In particular, this bound state has the following properties: A) it is chiral, in the sense that it localizes preferentially in one side of the emitter; B) when ∆ = 0, the bound state has its energy in the middle of the gap, EBS = 0, being fully direc- tional and with no amplitude in the sublattice to which the QE couples; C) it inherits the topological robustness to disorder from the bath. To make evident that the photonic SSH model is the simplest one-dimensional model where all these proper- ties are satisfied, and connect it with the topological fea- tures of the bath, let us consider a general bipartite bath Hamiltonian defined by: (cid:18) GA(k) F (k) F ∗(k) GB(k) (cid:19) HB(k) = . (S27) a plethora of Depending on the functions GA/B(k), F (k), relevant this Hamiltonian covers one- dimensional models with a middle band-gap where an ex- tra bound-state appears. In Table S6 we review the topo- logical properties of several of these models, and whether middle bound states show the features A -- C discussed above: The central part of the manuscript analyzes the prop- erties and consequences of a peculiar bound state which 1. When GA/B(k) = ωa±δω, F (k) = −J, we have the simple coupled-cavity array model with staggered (d)(e)Model Quantized Zak Phase Coupled cavity array with staggered energies SSH SSH with staggered energies SSH with next-nearest neighbours No Yes No Yes Bulk-Boundary Correspondence Not applicable Yes Not applicable No A B C No Yes Yes Yes No Yes Yes No No Yes No Yes* TABLE S6. Topological properties of several one-dimensional baths, and their corresponding bound state features A -- C (see text for discussion) when an emitter couples to them. 16 energies. The staggered energies break the symme- try between the A/B sublattices, opening a middle band-gap. In this case, an extra bound-state can be found in the middle of the bands but with none of the features A -- C. to go from one band-configuration to the other without closing the gap, such that the Zak phase is not quantized anymore. The middle bound states can be chiral, but they do not get the robustness to disorder since the model is topologically trivial. 2. Setting 4. Finally, we consider the SSH model adding next- GA/B(k) = ωa , F (k) = −J[(1 + δ) + (1 − δ)e−ik] , we recover the SSH model whose topological and bound state properties were already discussed in the main text. 3. One can combine the SSH model with staggered energies, i.e., GA/B(k) = ωa ± δω , F (k) = −J[(1 + δ) + (1 − δ)e−ik] , which still shows a middle band-gap but chiral sym- metry is broken. The staggered energies allow one nearest neighbour hoppings, i.e., GA/B(k) = ωa − 2J2 cos(2k) , F (k) = −J(cid:2)(1 + δ) + (1 − δ)e−ik(cid:3) . This model does not preserve chiral symmetry, but it does preserve spatial inversion symmetry. The later still leads to a quantized Zak phase, but bulk- boundary correspondence is not guaranteed any more (the number of edge states is not linked to the topological invariant) [31]. Even though the bound states are chiral, they are not fully direc- tional, and they are only robust to disorder of very restricted type (one which respects spatial inversion symmetry). [1] K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [2] X.-G. Wen, Rev. Mod. Phys. 89, 041004 (2017). [3] A. Y. Kitaev, Physics-Uspekhi 44, 131 (2001). [4] F. D. M. Haldane and S. Raghu, Phys. Rev. Lett. 100, 013904 (2008). [5] Z. Wang, Y. Chong, J. D. Joannopoulos, and M. Soljaci´c, Nature 461, 772 (2009). [6] N. Malkova, I. Hromada, X. Wang, G. Bryant, and Z. Chen, Opt. Lett. 34, 1633 (2009). [7] X.-D. Chen, D. Zhao, X.-S. Zhu, F.-L. Shi, H. Liu, J.-C. Lu, M. Chen, and J.-W. Dong, Phys. Rev. A 97, 013831 (2018). [8] M. C. Rechtsman, J. M. Zeuner, Y. Plotnik, Y. Lumer, D. Podolsky, F. Dreisow, S. Nolte, M. Segev, and A. Sza- meit, Nature 496, 196 (2013). Bandres, J. Ren, M. C. Rechtsman, M. Segev, D. N. Christodoulides, and M. Khajavikhan, Phys. Rev. Lett. 120, 113901 (2018). [12] P. St-Jean, V. Goblot, E. Galopin, A. Lemaıtre, and T. Ozawa, L. Le Gratiet, I. Sagnes, J. Bloch, A. Amo, Nature Photonics 11, 651 (2017). [13] W.-J. Chen, S.-J. Jiang, X.-D. Chen, B. Zhu, L. Zhou, J.-W. Dong, and C. T. Chan, Nature Communications 5, 5782 (2014). [14] T. Ozawa, H. M. Price, A. Amo, N. Goldman, M. Hafezi, L. Lu, M. C. Rechtsman, D. Schuster, J. Simon, O. Zil- berberg, and I. Carusotto, Rev. Mod. Phys. 91, 015006 (2019). [15] J. Perczel, J. Borregaard, D. E. Chang, H. Pichler, S. F. Yelin, P. Zoller, and M. D. Lukin, Phys. Rev. Lett. 119, 023603 (2017). [9] M. Hafezi, E. A. Demler, M. D. Lukin, and J. M. Taylor, [16] R. J. Bettles, J. c. v. Min´ar, C. S. Adams, I. Lesanovsky, Nature Physics 7, 907 (2011). and B. Olmos, Phys. Rev. A 96, 041603 (2017). [10] H. Zhao, P. Miao, M. H. Teimourpour, S. Malzard, R. El- Ganainy, H. Schomerus, and L. Feng, Nature Commu- nications 9, 981 (2018). [17] S. Barik, A. Karasahin, C. Flower, T. Cai, H. Miyake, W. DeGottardi, M. Hafezi, and E. Waks, Science 359, 666 (2018). [11] M. Parto, S. Wittek, H. Hodaei, G. Harari, M. A. [18] J. K. Asb´oth, L. Oroszl´any, and A. P. P´alyi, A 17 Short Course on Topological Insulators: Band Struc- ture and Edge States in One and Two Dimensions, Lec- ture Notes in Physics (Springer International Publishing, 2016) Chap. 1. [19] R. H. Dicke, Phys. Rev. 93, 99 (1954). [20] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. Ludwig, New Journal of Physics 12, 065010 (2010). 88, 333 (1992). [41] Y. Qi, Q. Yang, N. sen Yu, and A. Du, Journal of Physics: Condensed Matter 28, 126006 (2016). [42] A. Asenjo-Garcia, M. Moreno-Cardoner, A. Albrecht, H. J. Kimble, and D. E. Chang, Phys. Rev. X 7, 031024 (2017). [43] A. Gonz´alez-Tudela, V. Paulisch, H. J. Kimble, and J. I. [21] S. Ryu and Y. Hatsugai, Phys. Rev. Lett. 89, 077002 Cirac, Phys. Rev. Lett. 118, 213601 (2017). (2002). [44] L. Zhou, Z. R. Gong, Y.-x. Liu, C. P. Sun, and F. Nori, [22] G. W. Gardiner and P. Zoller, Quantum Noise, 2nd ed. Phys. Rev. Lett. 101, 100501 (2008). (Springer-Verlag, Berlin, 2000). [45] D. Witthaut and A. S. Sørensen, New Journal of Physics [23] C. Cohen-Tannoudji, J. Dupont-Roc, G. Grynberg, and P. Thickstun, Atom-photon interactions: basic processes and applications (Wiley Online Library, 1992). [24] F. Ciccarello, Phys. Rev. A 83, 043802 (2011). [25] S. John and T. Quang, Physical Review A 50, 1764 (1994). [26] V. P. Bykov, Soviet Journal of Quantum Electronics 4, 861 (1975). [27] S. John and J. Wang, Phys. Rev. Lett. 64, 2418 (1990). [28] G. Kurizki, Phys. Rev. A 42, 2915 (1990). [29] G. Calaj´o, F. Ciccarello, D. Chang, and P. Rabl, Phys. 12, 043052 (2010). [46] P. Lodahl, S. Mahmoodian, and S. Stobbe, Rev. Mod. Phys. 87, 347 (2015). [47] D. E. Chang, J. S. Douglas, A. Gonz´alez-Tudela, C.-L. Hung, and H. J. Kimble, Rev. Mod. Phys. 90, 031002 (2018). [48] Y. Liu and A. A. Houck, Nature Physics 13, 48 (2017). [49] M. Mirhosseini, E. Kim, V. S. Ferreira, M. Kalaee, A. Sipahigil, A. J. Keller, and O. Painter, Nature com- munications 9 (2018), 10.1038/s41467-018-06142-z. [50] T. Goren, K. Plekhanov, F. Appas, and K. Le Hur, Phys. Rev. A 93, 033833 (2016). Rev. B 97, 041106 (2018). [30] T. Shi, Y.-H. Wu, A. Gonz´alez-Tudela, and J. I. Cirac, [51] I. de Vega, D. Porras, and J. Ignacio Cirac, Phys. Rev. Phys. Rev. X 6, 021027 (2016). Lett. 101, 260404 (2008). [31] B. P´erez-Gonz´alez, M. Bello, A. G´omez-Le´on, and [52] L. Krinner, M. Stewart, A. Pazmino, J. Kwon, and G. Platero, Phys. Rev. B 99, 035146 (2019). D. Schneble, Nature 559, 589 (2018). [32] J. S. Douglas, H. Habibian, C.-L. Hung, A. Gorshkov, H. J. Kimble, and D. E. Chang, Nature Photonics 9, 326 (2015). [33] A. Gonz´alez-Tudela, C.-L. Hung, D. E. Chang, J. I. Cirac, and H. Kimble, Nature Photonics 9, 320 (2015). [34] P. Hauke, F. M. Cucchietti, A. Muller-Hermes, M.-C. Banuls, J. I. Cirac, and M. Lewenstein, New Journal of Physics 12, 113037 (2010). [35] P. Richerme, Z.-X. Gong, A. Lee, C. Senko, J. Smith, and M. Foss-Feig, S. Michalakis, A. V. Gorshkov, C. Monroe, Nature 511, 198 (2014). [36] S. Katsura, Phys. Rev. 127, 1508 (1962). [37] A. Auerbach, Interacting electrons and quantum mag- [53] M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kitagawa, E. Demler, and I. Bloch, Nature Physics 9, 795 (2013). [54] W. Tan, Y. Sun, H. Chen, and S.-Q. Shen, Scientific Reports 4, 3842 (2014). [55] S. Kruk, A. Slobozhanyuk, D. Denkova, A. Poddubny, and I. Kravchenko, A. Miroshnichenko, D. Neshev, Y. Kivshar, Small 13, 1603190 (2017). [56] S. R. Pocock, X. Xiao, P. A. Huidobro, and V. Giannini, ACS Photonics 5, 2271 (2018). [57] S. Mittal, V. V. Orre, and M. Hafezi, Optics Express 24, 15631 (2016). [58] M. C. Rechtsman, Y. Lumer, Y. Plotnik, A. Perez-Leija, netism (Springer Science & Business Media, 2012). A. Szameit, and M. Segev, Optica 3, 925 (2016). [38] T. Morita and T. Horiguchi, Physics Letters A 38, 223 (1972). [59] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [60] N. P. Armitage, E. J. Mele, and A. Vishwanath, Rev. [39] P. Sen and B. K. Chakrabarti, Phys. Rev. B 40, 760 Mod. Phys. 90, 015001 (2018). (1989). [40] P. Sen, S. Chakraborty, S. Dasgupta, and B. K. Chakrabarti, Zeitschrift fur Physik B Condensed Matter [61] H. Nakazato, M. Namiki, and S. Pascazio, International Journal of Modern Physics B 10, 247 (1996). [62] A. Gonz´alez-Tudela and J. I. Cirac, Phys. Rev. A 96, 043811 (2017). 

1309.6983

1

1309

Phase diagrams of magnetopolariton gases

The magnetic field effect on phase transitions in electrically neutral bosonic systems is much less studied than those in fermionic systems, such as superconducting or ferromagnetic phase transitions. Nevertheless, composite bosons are strongly sensitive to magnetic fields: both their internal structure and motion as whole particles may be affected. A joint effort of ten laboratories has been focused on studies of polariton lasers, where non-equilibrium Bose-Einstein condensates of bosonic quasiparticles, exciton-polaritons, may appear or disappear under an effect of applied magnetic fields. Polariton lasers based on pillar or planar microcavities were excited both optically and electrically. In all cases a pronounced dependence of the onset to lasing on the magnetic field has been observed. For the sake of comparison, photon lasing (lasing by an electron-hole plasma) in the presence of a magnetic field has been studied on the same samples as polariton lasing. The threshold to photon lasing is essentially governed by the excitonic Mott transition which appears to be sensitive to magnetic fields too. All the observed experimental features are qualitatively described within a uniform model based on coupled diffusion equations for electrons, holes and excitons and the Gross-Pitaevskii equation for exciton-polariton condensates. Our research sheds more light on the physics of non-equilibrium Bose-Einstein condensates and the results manifest high potentiality of polariton lasers for spin-based quantum logic applications.

cond-mat.mes-hall

cond-mat

1 Phase Diagrams of Magnetopolariton Gases Vladimir P. Kochereshko1,2, Mikhail V. Durnev1,2, Lucien Besombes3, Henri Mariette3, Victor F. Sapega1,2, Alexis Axitopoulos4, Ivan G. Savenko9,10, Timothy C.H. Liew5, Ivan A. Shelykh5, Alexey V. Platonov1,2, Simeon I. Tsintzos7, Z. Hatzopoulos7, Pavlos Lagoudakis4, Pavlos G. Savvidis6,7, Christian Schneider8, Matthias Amthor8, Christian Metzger8 , Martin Kamp8, Sven Hoefling8, Alexey Kavokin1,3 1 Spin Optics Laboratory, Saint--‐Petersburg State University, 1, Ulianovskaya, 198504, St--‐Petersburg, Russia 2 Ioffe Physical--‐Technical Institute, Russian Academy of Sciences, 26, Politechnicheskaya, 194021, St--‐Petersburg, Russia 3 Institut Néel, CNRS/UJF 25, avenue des Martyrs --‐ BP 166, Fr--‐38042 Grenoble Cedex 9, France 4 Physics and Astronomy School, University of Southampton, Highfield, Southampton, SO171BJ, UK 5 Division of Physics and Applied Physics, Nanyang Technological University 637371, Singapore 6 Department of Materials Science & Technology, University of Crete, Greece 7 IESL--‐FORTH, P.O. Box 1527, 71110 Heraklion, Crete, Greece 8 Technische Physik and Wilhelm--‐Conrad--‐Röntgen--‐Research Center for Complex Material Systems, Universität Würzburg, D--‐97074 Würzburg, Am Hubland, Germany. 9 Science Institute, University of Iceland, Dunhagi--‐3, IS--‐107, Reykjavik, Iceland 10 Department of Applied Physics/COMP, Aalto University, PO Box 14100, 00076 Aalto, Finland The magnetic field effect on phase transitions in electrically neutral bosonic systems is much less studied than those in fermionic systems, such as superconducting or ferromagnetic phase transitions1,2. Nevertheless, composite bosons are strongly sensitive to magnetic fields: both their internal structure and motion as whole particles may be affected3. A joint effort of ten laboratories has been focused on studies of polariton lasers4, where non--‐equilibrium Bose--‐Einstein condensates of bosonic quasiparticles, exciton--‐polaritons, may appear or disappear under an effect of applied magnetic fields. Polariton lasers based on pillar or planar microcavities were excited both optically and electrically. In all cases a pronounced dependence of the onset to lasing on the magnetic field has been observed. For the sake of comparison, photon lasing (lasing by an electron--‐hole plasma) in the presence of a magnetic field has been studied on the same samples as polariton lasing. The threshold to photon lasing is essentially governed by the excitonic Mott transition5 which appears to be sensitive to magnetic fields too. All the observed experimental features are qualitatively described within a uniform model based on coupled diffusion equations for electrons, holes and excitons and the Gross--‐Pitaevskii equation for exciton--‐polariton condensates. Our research sheds more light on the physics of non--‐equilibrium Bose--‐Einstein condensates and the results manifest high potentiality of polariton lasers for spin--‐based quantum logic6 applications. 2 Mixed light--‐matter quasiparticles in the form of exciton--‐polaritons7 demonstrate remarkable collective properties including polariton lasing8, Josephson oscillations9, vortices10, solitons11, optical spin Hall12 and, possibly, spin Meissner13 effects. Clearly, microcavities present a laboratory rich in complex many--‐body processes. Electrons, holes, excitons and exciton--‐polaritons coexist and interact giving rise to new phases or pseudo--‐phases14. Magnetic field appears to be an efficient tool of switching between some of these phases. Here we present the first detailed study of the magnetic field effect on the (out of equilibrium) phase transitions in microcavities. So far, polariton lasers are the only existing example of bosonic lasers, where coherent light is emitted spontaneously by a bosonic condensate15. In the meanwhile, magnetic field has been recently shown to be instrumental for the achievement of polariton lasing in electrically pumped microcavities16,17. Besides, it affects the second phase transition towards a photonic laser, which takes place when stimulated emission of light from electron--‐hole plasma starts. The switch from polariton to photon lasing18 is associated with the exciton Mott transition19: the phase transition between a bosonic gas (exciton--‐polariton gas) and neutral plasma (electron--‐hole plasma). Previous experimental studies evidenced a Mott transition in semiconductor quantum wells manifested by sharp changes of the photoluminescence (PL) energy and linewidth20,21. This paper summarises a coordinated effort of ten laboratories aimed at understanding magnetic field induced transitions to bosonic lasing (polariton lasing) and fermionic lasing (photon lasing). A variety of state of the art structures coming from two molecular beam epitaxy facilities, including both planar and pillar microcavities with optical and electrical injection, have been studied by means of polarisation resolved magneto--‐photoluminescence under continuous wave (cw) and pulsed excitations. Figure 1(a--‐e) shows the photoluminescence spectra of a round pillar microcavity with embedded GaAs/AlGaAs quantum wells taken at different pumping power in the presence of the external magnetic field applied along the axis of the structure (the Faraday geometry). Details of the sample structure appear in the Methods section. The structure has been excited through a micro--‐objective by short--‐duration light pulses focused to a spot with 1.5 µm diameter and having a high energy compared to the exciton transition and the cavity mode energies. These pulses create electron--‐hole pairs, which cool down and form excitons. The latter relax further down in energy, and eventually form a condensate of exciton polaritons, in the polariton lasing regime. At low pumping intensities (Figure 1(a)) and relatively high magnetic fields (over 7 T) we observe a narrow and intense polariton lasing mode showing a characteristic diamagnetic shift22. The diamagnetic shift is a signature of the exciton--‐polariton state: the exciton energy enhances proportionally to B2. Increasing the pumping strength or lowering the field results in dramatic modifications of the spectra (presented in Figure 1(b)): the exciton--‐polariton mode abruptly disappears, while a strong laser line at a higher energy emerges. This line is clearly pinned to the cavity photon mode whose energy is not affected by magnetic fields. The sharp transition from the polariton to photon lasing regime is a manifestation of the abrupt excitonic Mott transition in our system. Interestingly enough, this photon laser emission line disappears again at very weak magnetic fields at the intermediate pumping power (Figure 1(c,d)). Figure 2(a) shows how the spectra of this pillar change when modifying the pump power at a fixed magnetic field B=6 T. One can clearly distinguish the polariton and photon lasing regimes. The thresholds are identified from the peak intensity dependencies on the pumping power for the lower polariton (LP) and cavity (C) photon modes (Figure 2(b)). This analysis has been done for the full range of available pumping and magnetic field strengths which allowed the obtaining of a phase diagram shown in Figure 2(c). One can see that the threshold to polariton lasing decreases with the magnetic field 3 increase. This tendency is confirmed by experiments done on the pillar samples of different diameters (see the Supplementary information). In contrast, the behaviour of the threshold to photon lasing is strongly non--‐monotonic: initially it sharply decreases, and then increases (Figure 2(c)). The switching between exciton--‐polariton gas and electron--‐hole plasma has all the features of a first order phase transition, if this term can be applied to a non--‐equilibrium optically driven system. We interpret the experimental observations both for pillar and planar geometries by an interplay of the exciton Bohr radius shrinkage in the presence of the magnetic field and field--‐controlled diffusion of electrons, holes and excitons away from the excitation spot. The magnetic field and phase space filling both bring the exciton transition into resonance with the cavity mode, which switches on the photon lasing effect and fully suppresses exciton--‐polaritons in the system in the course of an avalanche Mott transition. Our model accounts for the diffusion of electrons and holes away from the excitation spot, formation of the exciton reservoir from the electron--‐hole gas and formation of the polariton condensate once the density of excitons in the reservoir achieves a critical value (see the scheme in Figure 3(b)). The details of the model are summarised in the Supplementary information. The results of calculation are shown by dashed lines in the phase diagram (Figure 2(c)). The decrease of the polariton laser threshold with increase of magnetic field is due to the suppression of in--‐plane diffusion of electrons and holes away from the excitation spot, because the magnetic field results in an orbital movement of the carriers. This diffusion leads to the dilution of the concentration of electrons, holes and excitons at zero field. In the presence of magnetic field normal to the quantum well planes, electrons and holes cannot spread further away from the excitation area than by distances comparable with their cyclotron orbit. The suppression of in--‐plane diffusion of carriers has also an important impact on the excitonic Mott transition, which is governed by the condition23 𝜅𝑛!𝑎!!=1 where nx is the in--‐plane concentration of excitons, aB is the exciton Bohr radius, and κ is a coefficient depending on the geometry of the structure. The increase of nx under the pumping spot because of the magnetic field effect leads to the decrease of the threshold power for photon lasing at low magnetic fields seen in Figure 1(c). At high fields, due to the shrinkage of the exciton Bohr radius, the critical pumping power increases again. In order to check the validity of the model and the universality of the observed effects, we have studied magneto--‐photoluminescence spectra of a series of planar microcavity samples (see Methods). The results of these studies are summarised in Figure 3. The samples are pumped non--‐resonantly by a cw laser light. The threshold to polariton lasing has been observed very clearly. It appeared to be strongly sensitive to the magnetic field (Figure 3(a)) but also to the size of the excitation spot, as the phase diagrams in the lower panels of Figure 3 show. In the case of a small (10 µm) spot, the threshold non--‐monotonically depends on the field. For larger spots, the initial decrease disappears, and the threshold demonstrates a steady and monotonic increase with the increase of the magnetic field. This behaviour is consistent with our kinetic model (see the scheme in Figure 3(b)). The monotonic decrease of the threshold to polariton lasing in pillar samples is replaced by a non--‐monotonic behaviour in planar samples and returns to a monotonic behaviour as we increase the excitation spot size (see Figure 3(d--‐f)). At zero field, the exciton concentration at the center of the excitation spot appears to be strongly diluted due to the diffusion of carriers if the pump spot is narrow (Figure 3(c) and Supplementary Figure 4 S2). On the other hand, for large spots, the diffusion has little effect on the exciton concentration, which is why the suppression of diffusion by a magnetic field does not so notably affect the threshold of polariton lasing. The increase of the threshold with a magnetic field observed for all spot sizes is due to the shrinkage of the exciton Bohr radius which leads to the reduction of the exciton radiative life--‐time and, consequently, emptying of the exciton reservoir. Polariton lasers with electrical injection strongly differ from the lasers with optical pumping from the point of view of the dynamics of formation of polariton condensates. Indeed, in our samples, electron and hole injection is nearly homogeneous over the whole area of the sample, which is why the diffusion effects play little or negligible role. On the other hand, the lifetime of excitons in the structure is shortened in the presence of the magnetic field24 which leads to the increase of the threshold for polariton lasing. Besides, the shrinkage of the exciton Bohr radius is also responsible for the increase of the threshold to photon lasing associated with the Mott transition. This tendency is experimentally confirmed for the polariton lasing (Figure 4(a)). Figure 4(b) shows the thresholds to polariton and photon lasing at B=5 T as functions of temperature. Both thresholds decrease in full agreement with the theory, which accounts for acceleration of the acoustic phonon relaxation processes leading to electron and hole relaxation to the exciton reservoir and exciton relaxation to the polariton condensate. To summarize, we observed magnetic--‐field controlled transitions to polariton and photon lasing, which can be considered as out of equilibrium Bose--‐Einstein condensation and Mott phase transitions. Magnetic field effects on composite bosons are of key importance for realisation of quantum logic devices based on spin properties of bosonic liquids25, in particular, magnetic control of the Bose--‐Einstein condensation of exciton polaritons may be used in quantum and classical optical memories. Acknowledgments We thank Jacqueline Bloch for many helpful discussions. V.K., M.D., V.F.S. and A.K. acknowledge support from the Russian Ministry of Science and Education, contract (contract No. 11.G34.31.0067). P.G.S. acknowledges support from Greek GSRT program Aristeia (grant No. 1978). C.S., M. A. J.F., M.K and S.H. acknowledge support from the state of Bavaria. Methods Summary Sample design and fabrication. All structures have been grown by molecular beam epitaxy. The planar sample is a high--‐finesse Q>16000 mircocavity formed by 5/2 λ Al0.3Ga0.7As cavity surrounded by 32 (35) period AlAs/Al0.15Ga0.85As top (bottom) DBR mirrors. Four sets of three 10nm GaAs quantum wells are embedded inside the cavity at the antinodes of the electric field producing a Rabi splitting of ΩR= 9.2 meV. For the micropillars studies, reactive ion etching has been applied to sculpt circular mesas with diameter ranging from 1 to 40 μm. The data shown in Figures 1,2 are taken at the positive exciton--‐photon detuning of about 2 meV. The data in Figure 3 are taken at the negative detuning of 7 meV. The sample for a polariton laser with electrical injection is the same as in Ref. 16. Experimental set--‐up. The optically excited spectra were pumped by a Ti:Sa laser with pulse duration of 2 picoseconds and energy 1.62 eV, which corresponds to a range of transparency in the Bragg mirrors. Time integrated micro PL spectra of the pillar samples were recorded by a CCD detector in magnetic fields up to 11 T in the Faraday geometry with spatial resolution of about 1 µm. The planar microcavities were mounted in a gas flow sample chamber kept at T = 3 K of superconducting cryostat operating in 5 magnetic fields up to 5 T. The PL signal was integrated in a solid angle determined by numerical aperture (NA) of the collimating lens (NA = 0.5 for 10 μm spot and NA = 0.08 for 100 μm spot). All PL experiments were performed in back scattering geometry and at about normal incidence of the light on the sample surface. Figures Fig. 1. Magneto photoluminescence of the 5m diameter micropillar sample. Panels represent different pumping powers: 0.09 mW (a), 0.58 mW (b), 0.85 mW (c), 1 mW (d) and 1.5 mW (e). Both polariton lasing (a) and a very sharp transition to photon lasing (c--‐e) are observed. Fig. 2. Phase transitions in micropillar samples. a, b, Emission pattern and the integrated PL intensity of the 5 µm pillar at B = 6 T. The arrows indicate the onset and offset of the polariton laser (LP1 and LP2) and the onset of the photon laser (C). c, Phase diagram of a 5 µm pillar. Note that the polariton lasing transition at zero field corresponds to a pump power intermediate between those considered in Figure 1(a,b). The lines show the results of simulation (see Supplementary information). The part with horizontal white bands marks the polariton gas regime, beyond the offset of polariton lasing. µ6 Fig. 3. Magneto--‐polariton lasing of the planar microcavity sample. a, Integrated PL intensity as a function of pump power at different magnetic fields. The laser is focused to a 10 µm spot. A clear threshold to polariton lasing is observed in the whole range of magnetic fields. b, The scheme of polariton formation under non--‐resonant pumping. c, Theoretical curves for the exciton density at the center of the excitation spot for a small spot (d = 10 µm, red curve) and a large spot (d = 100 µm, black curve). The dramatic difference in their behaviours reflects the crucial role of diffusion suppression by magnetic field in the case of small spots. d--‐f, Phae diagrams of polariton emission for three different excitation spot sizes (d = 10, 40 and 100 µm). Red squares represent the experimental data, and the results of modeling are shown by dashed curves. 7 Fig. 4. Phase diagrams of the electrically pumped polariton laser. a,b, Magnetic and temperature dependent phase diagrams. Dashed lines show the results of our kinetic modelling. c, Linewidth of the PL emission as a function of injection current. Dashed lines indicate the onset of the polariton and photon lasing. 8 1 Giamarchi, Th., Rüegg, C. & Tchernyshyov, O. Bose--‐Einstein condensation in magnetic insulators. Nature Physics 4, 198 --‐ 204 (2008). 2 Moskalenko, S.A., Liberman, M.A. and Dumanov, E.V. Exciton condensation under high magnetic field. Journal of Nanoelectronics and Optoelectronics 6, 1--‐27 (2011). 3 Gor'kov, L.P., Dzyaloshinskii, I.E. Contribution to the Theory of the Mott Exciton in a Strong Magnetic Field. Sov. Phys. JETP 26 (2), 449--‐451 (1968) [ZhETF 53 (2), 717--‐722 (1967)]. 4 Ĭmamoğlu, A., Ram, R.J., Pau, S. & Yamamoto, Y. Nonequilibrium condensates and lasers without inversion: exciton--‐polariton lasers. Phys. Rev. A 53, 4250 -- 4253 (1996). 5 Timofeev, V.B., Larionov, A.V., Grassi--‐Alessi, M., Capizzi, M., and Hvam, J.M. Phase diagram of a two--‐dimensional liquid in GaAs/AlxGa1--‐xAs biased double quantum wells. Phys. Rev. B 61, 8420--‐8426 (2000). 6 Paraïso, T.K., Wouters, M., Léger, Y., Morier--‐Genoud, F. and Deveaud--‐Plédran, B. Multistability of a coherent spin ensemble in a semiconductor microcavity. Nature Materials 9, 655 (2010). 7 Hopfield, J. J. Theory of the contribution of excitons to the complex dielectric constant of crystals. Phys. Rev. 112, 1555 (1958). 8 Christopoulos, S., von Hoegersthal G. Baldassarri Hoeger, Grundy, A. J. D., Lagoudakis, P. G., Kavokin, A. V., Baumberg, J. J., Christmann, G., Buttè, R., Feltin, E., Carlin, J.--‐F. and Grandjean, N. Room--‐temperature polariton lasing in semiconductor microcavities. Phys. Rev. Lett. 98, 126405 (2007). 9 Lagoudakis, K. G., Pietka, B., Wouters, M., André, R., and Deveaud--‐Plédran, B. Coherent Oscillations in an Exciton--‐Polariton Josephson Junction. Phys. Rev. Lett. 105, 120403 (2010). 10 Lagoudakis, K. G., Ostatnický, T., Kavokin, A. V., Rubo, Y. G., André, R., Deveaud--‐Plédran, B. Observation of Half--‐Quantum Vortices in an Exciton--‐Polariton Condensate. Science 326, 974 (2009). 11 Amo A. et al, Polariton Superfluids Reveal Quantum Hydrodynamic Solitons, Science, 332, 1167--‐1170 (2010). 12 Leyder, C., Romanelli, M., Karr, J. Ph., Giacobino, E., Liew, T. C. H., Glazov, M. M., Kavokin, A. V., Malpuech, G. and Bramati, A. Observation of the optical spin Hall effect. Nature Physics 3, 628 (2007). 13 Larionov, A. V. et al. Polarized nonequilibrium Bose--‐Einstein condensates of spinor exciton polaritons in a magnetic field. Phys. Rev. Lett. 105, 256401 (2010). 14 Ogawa T., Exciton Mott transition and quantum condensation in electron--‐hole systems, Phys. Stat. Sol. (c), 6, 28--‐33 (2009). 15 Kavokin, A. and Malpuech, G. Polariton lasers, in "Radiation--‐matter interaction in confined systems" edited by L.C. Andreani, G. Benedek, E. Molinari, Società Italiana di Fisica, Bologna, 287--‐301 (2002). 16 Schneider, C., Rahimi--‐Iman, A., Kim, N. Y., Fischer, J., Savenko, I. G., Amthor, M., Lermer, M., Wolf, A., Worschech, L., Kulakovskii, V. D., Shelykh, I. A., Kamp, M., Reitzenstein, S., Forchel, A., Yamamoto, Y. and Hofling, S. An Electrically Pumped Polariton Laser, Nature 497, 348--‐352 (2013). 17 Bhattacharya P., Xiao B., Das A., Bhowmick S., Heo J., Solid State Electrically Injected Exciton--‐Polariton Laser. Physical Review Letters 110, 206403 (2013). 18 Kammann, E., Ohadi, H., Maragkou, M., Kavokin, A. V. & Lagoudakis, P. G. Crossover from photon to exciton--‐polariton lasing. N. J. Phys. 14, 105003 (2012). 19 Mott, N.F. Metal Insulator Transition, Taylor and Francis, London, 1990. 20 Kappei, L., Szczytko, E., Morier--‐Genoud, F. and Deveaud, B. Direct Observation of the Mott Transition in an Optically Excited Semiconductor Quantum Well. Phys. Rev. Lett. 94, 147403 (2005). 21 Stern, M., Garmider, V., Umansky, V., and Bar--‐Joseph, I. Mott Transition of Excitons in Coupled Quantum Wells. Phys. Rev. Lett. 100, 256402 (2008). 22 Rahimi--‐Iman, A., Schneider, C., Fischer, J., Holzinger, S., Amthor, M., Höfling, S., Reitzenstein, S., Worschech, L., Kamp, M. and Forchel, A. Zeeman splitting and diamagnetic shift of spatially confined quantum--‐well exciton polaritons in an external magnetic field. Phys. Rev. B 84, 165325 (2011). 23 Nikolaev, V.V and Portnoi, M.E. Theory of excitonic Mott transition in quasi--‐two--‐dimensional systems, Superlattices and Microstructures 43 (5--‐6), 460--‐464 (2008). 24 Berger J.D., Lyngnes O., Gibbs H.M., Khitrova G., Nelson T.R., Lindmark E.K., Kavokin A.V., Kaliteevski M.A., and Zapasskii V.V., Magnetic Field Enhancement of the Exciton--‐Polariton Splitting in a 9 Semiconductor Quantum Well Microcavity: The Strong Coupling Threshold, Phys. Rev. B 54, 1975--‐1981 (1996). 25 Liew, T. C. H., Kavokin, A. V., and Shelykh, I. A. Optical circuits based on polariton neurons in semiconductor microcavities. Phys. Rev. Lett. 101, 016402 (2008). SUPPLEMENTARYMATERIALBasicequationsThecoherentpolaritonstatewithinourmodelisformedduringthethree-stageprocessillustratedschematicallyinFig.3bofthemaintext.Thefirststageisthenonresonantpumpingofelectronsandholeswithlargevaluesofwavevectors;thesecondstageistheenergyrelaxationofchargedcarriersfollowedbytheformationofanexcitonreservoir,andthefinalstageistheformationofpolaritonsfollowedbyitsresonantscatteringtothegroundstate.Thefirsttwoprocessescanbedescribedbythefollowingsetofdynamicaldiffusionequations:∂∂tne=De∆ne−wnenh+Je;∂∂tnh=Dh∆nh−wnenh+Jh;∂∂tnx=Dx∆nx+wnenh−nxτx.(1)Herene,nhandnxarethedensitiesofelectrons,holesandexcitons,respectively;Diarethediffusioncoefficients,wdescribestheexcitonformationrate;JeandJharethepumpratesforelectronsandholesandτxistheexcitonlifetime.Theformationofexcitonsinourmodelisdescribedbythetermwnenhcorrespondingtothenon-geminate(orbimolecular)mechanismbecausethecontributionfromthegeminatemechanismistypicallysmall[1,2].Sinceinastrongcouplingregimenon-radiativelossesarenegligible,weassumethatexcitonsdecaymainlyduetoradiativeprocesses,sothatτxcorrespondstotheradiativelifetime.Todescribetheformationofacondensatethethirdequationin(1)shouldbesupplementedwiththescatteringterms:∂nx∂t=Dx∆nx+wnenh−nxτx−Γin(ψ+2+ψ−2)nx.(2)HereΓinisthetransitionratefromthereservoirtothecondensate.Theequationsforthecondensatewavefunctionsψ±read:i∂ψ±∂t+2∇22mψ±=−i2τψ±+(α1ψ±2+α2ψ∓2±∆Z2)ψ±+i2Γinnxψ±.(3)Heremistheeffectivemassofapolaritonatzerowavevector,τisalifetimeofthecondensate,∆ZistheZeemansplittingandα1,2arepolariton-polaritoninteractionconstantsfortheparallelandanti-parallelspinconfigurations.Theequationforthecondensatewavefunctionisbasedonamean-fieldtreatment[3],modifiedforspin[4]andincoherentexcitation[5].Wenotethatinthepresentequationsweneglectthedriftterms,arisingfromthedifferentspatialdistributionofelectronsandholesduetothepossibledifferenceinthediffusioncoefficientsDe,h.Comparisonwithmoreaccuratesimulations(includingthedriftcurrents)showedthesamebehaviorsforthephasediagramswithonlyminorquantitativedifferences,thereforewewillfurtheruseDe=Dh≡Dandomitthepossibilityforelectricalcurrentsinsidethestructure.LowandhighpumpingregimesLetus,first,focusonthesetofequations(1)fortheexcitonreservoirassumingtheonsetofpolaritonlasingatnx=n(th)x,wheren(th)xisacriticalexcitonconcentrationcorrespondingtothepopulationofpolaritongroundstateequaltounity.InthefollowingweassumethattheelectronandholecomponentsarepumpedwithequalratesbytheGaussian-shapedlaserbeamJe=Jh≡Je−r2/R2.Theabsenceofthedriftcurrentsmeansequalconcentrationsforelectronsandholes,ne=nh≡n,satisfyingthefollowingequationinthestationarylimit:D∆n−wn2+J(r)=0.(4)NonlinearEq.(4)canbesolvednumerically,howeverasimplevariationalprocedurewiththecarrierdensitytakenintheformn(r)=n0e−r2/a2givesagoodapproximationpreservingallthephysicalinsight.Moreover,therearetwolimitingcaseswhenEq.(4)allowsforanalyticalsolutions.Thefirstonecorrespondstoalowpumpingregime:inthatcaseonedealswithlowconcentrationsandthenonlinearterminEq.(4)issmall.Theeffectivelifetimeofaparticleisthenτ=1/(wn0),andthespreadofcarrierdensityisa=√Dτ.IntegrationofEq.(4)overspacedomaingives:a=D/(R√wJ).Thiseffectiveradiusobviouslyshouldbegreaterthantheradiusofthepumpspot,R,sothata(cid:29)RandJ(cid:28)J∗=2D2/(wR4).Theexcitonicdensityatr=0isnx(0)=wn20τx(ifonetotallyneglectsthediffusionofexcitons,i.e.setsDx≡0)anditbehavesquadraticallywiththepumpintensity:nlowpumpx(0)∝wR42D2J2τx=J2J∗τx.(5)Intheoppositecaseofahighpumpregime,wecanneglectthediffusionterminEq.(4)resultinginaveryshortlinearinpumpexpressionfortheexcitondensity:nhighpumpx(0)=Jτx.(6)SincethetransitionbetweenthediscussedregimesisdefinedbyJ∗whichisinverselyproportionaltothefourthpowerofR,oneshouldexpectthattheexcitonicdensitydistributiondependenceonthepumppowerwithandwithoutamagneticfieldsignificantlydependsontheexcitationspotradius.ThiseffectisillustratedinFig.S1a,wherenx(0)isplottedfortwospotradiidifferentbyoneorderofmagnitude.Onecanseethatforthesharperspot(redcurve)J∗significantlyincreases.Thus,forthefixedexcitondensityweswitchbetweenthehigh-pumpingregime(blackcurve)andthelow-pumpingregime(redcurve).ApplicationofamagneticfieldaltersparametersDandτx.MagneticfieldleadstosuppressionofthediffusionofchargedcarrierswhichcanbedescribedusingtheEinsteinrelationandtheHallexpressionforconductivity:D(B)=kBTµ0/e1+B2/B20,B0=cµ0.(7)Hereµ0=eτp/misthemobilityofelectron/holegasatB=0andTisitstemperature.TheexperimentaldatashowthatTmightbesignificantlyhigherthanthelatticetemperature(e.g.T≈50K).ExcitoniclifetimedecreaseswiththemagneticfieldduetothedecreaseoftheexcitonBohrradius(τx∝a2Bin2Dstructures[6],seeFig.S1b).AreasonabledependenceshouldbetheLorentzian:τx(B)=τ01+B2/B21(8)withB1sufficientlylargerthanB0[7].Itbecomesobviousnow,thatforthetworegimesdiscussedaboveoneshouldexpectcompletelydifferentbehaviorofexcitons'densityinthecenterofthespotasafunctionofmagneticfield.Indeed,inthehighpumpingregime,nx(0)∝τx(B)decreasesmonotonouslyasafunctionofmagneticfield,whileinthelowpumpingregime,nx(0)∝τx(B)/D2(B)which(inthecaseB1>B0)resultsintheenhancementofexcitonformationatr=0.Further,sinceparameterJ∗dependsonthemagneticfieldviathediffusioncoefficient,themagneticfieldactsasaswitchbetweenthetworegimesleadingtoanon-monotonousbehavioroftheexcitons'density.ThisisillustratedinFig.3cinthemaintextoftheLetter.ResultsofcalculationsandparametersetsusedAsithasbeenalreadymentioned,theonsetofpolaritonlasingisdescribedbyaconditionnx=n(th)x,wheren(th)xcanbesetmanually.ThephotonlasingonsetcoincideswiththeMotttransitioninthesystemanditisdescribedbythefollowingcondition[8]:κnxa2B=1,(9)whereκisthecoefficientoftheorderofunity.Inthesimulationswesetκ=1.Usingthosetwoconditions,onecanfindthethresholdpumpintensityasafunctionofmagneticfield.ThisfunctionisplottedinFig.2c,Fig.2dandFig.3d-finthemaintext.Theexperimentalvaluesforwfoundinliteratureare:w=6±2cm2/s[9],w=15cm2/s[1],w≥0.5cm2/s[10].Thetheoreticalcalculations(accountingforrelaxationonopticalandacousticphonons[2])resultinw=0.1÷10cm2/s.Thefollowingsetsofparameterswereemployedtocalculatethephasediagramsshowninthemaintext:1.Fig.2cD0=1000cm2/s,B0=1T,B1=9T,w=1cm2/s,nth=5×1010cm−2,aB(B=0)=13nm;2.Fig.2dD0=1500cm2/s,B0=2.5T,B1=10T,w=0.1cm2/s,nth=5×1010cm−2;3.Fig.3d-fD0=900cm2/s,B0=1T,B1=7T,w=1cm2/s,nth=3×109cm−2.Accountforthecouplingbetweentheexcitonicreservoirandtheexciton-polaritoncondensateToinvestigatetheroleofthemagneticfieldontheformationoftheexciton-polaritoncondensatefromthereservoirofexcitons,weusednumericalanalysisofthecoupledEqs.(1),(2),(3).Theresultsofmodelingofthecondensatereal-spaceprofile,ψ2,arepresentedinFig.S2.Onecanseethat,inagreementwiththecalculationsfortheexcitonicreservoir,amagneticfieldleadstolocalizationofthecondensedensity.Thecalculatedphasediagramsdescribingpolaritonlasingshowthefollowingbehavior:monotonousthresholdincreaseinthecaseofthewidelaserspotandnon-monotonousbehaviorinthecaseofthesharpfocusing.Thus,itisimportanttomention,thatthisaccuratemodelvalidatesthesimplecalculationspresentedabovewhichaccountfortheevolutionoftheexcitoniccomponentonly.EquationsforthelaserwithcurrentinjectionThesituationisquitedifferentifweconsideranelectricallypumpedlaser.Inthiscaseanelectronandaholeareexcitedindifferentspotsofthesample,thusleadingtosmallvaluesoftheparameterw.Wewillassumethatinoursampletheelectricalinjectionishomogeneousalongthequantumwellplanesmeaningthatthediffusionisneglectedinthecaseofcurrentinjection.Ontheotherhand,therearelotsofpossiblemechanismsofelectronandholedecaysnotleadingtotheexcitonformation(includingthe'fly-through'oftheparticles,non-radiativeprocessesatthepillarsurface,etc.),sothatthenon-radiativedecaytimesτeandτhshouldbeintroduced.ThekineticequationssimilartoEq.(1)thereforeread:∂∂tne=−wnenh−neτe+Je;∂∂tnh=−wnenh−nhτh+Jh;∂∂tnx=wnenh−nxτx.(10)Inthestationarylimit,weobtainthealgebraicsystemofequations,thesolutionofwhichreads:ne=−12wτh−12(Jh−Je)τe+s14(cid:20)1wτh+(Jh−Je)τe(cid:21)2+Jeτewτh,(11a)nh=neτhτ2+(Jh−Je)τh.(11b)Wewillfurthermakesomesimplifications,namely:τe=τh≡τandJe=Jh≡J,sothatne=nh≡n.Inarealisticlimitofsmallcarrierlifetimes,Jwτ2<1,thecarrierconcentrationisn=Jτ,andfortheexcitondensityoneobtains:nx=wJ2τ2τx.(12)Wewillassumeausualdependenceofτxonmagneticfield[seeEq.(8)],whichgivesthefollowingexpressionsforthepolaritonandphotonthresholds:J(pol)th(B)=Jth,0s1+B2B21,(13a)J(ph)th(B)=Jth,0pntha2B(0)(cid:18)1+B2B21(cid:19).(13b)HereJth,0=pnth/(wτx(0)τ2)isthepolaritonthresholdatzeromagneticfield.ThecalculatedphasediagramispresentedinFig.4ainthemaintextwithJth,0=85A/cm2,B1=7Tandntha2B(0)=0.65.Thetemperaturedependenceoflasingthresholdsisruledbythestronglytemperature-dependentexcitonformationrate,w.Weassumethattheexcitonformationrateinthelow-temperatureregionincreaseswithtemperature,following:w(T)=w0exp(−T0/T),whereT0issomeeffectivetemperature.ItleadstoadecreaseofthethresholdswhichisillustratedinFig.4bofthemaintext.ThevalueofT0ischosenT0=5K.nx(0) (cm-2)11031061091012J (cm-2/s)101410161018102010221024Diffusion coefficient (cm2/s)02004006008001000Bohr radius aB (nm)78910111213Magnetic field (T)02468101214abFig.S1:Switchbetweenthelinearandnonlinearregimes.a,Excitonicdensityinthecenterofthespotasafunctionofpumpintensity.Redcurvecorrespondstoasharperfocusing,theradiiaredifferentbyoneorderofmagnitude.ThedashedlinesindicatethepositionofJ∗.b,BohrradiusanddiffusioncoefficientdependencesonmagneticfieldasusedincalculationspresentedinFig.2cinthemaintext.B = 0 TB = 3 TFig.S2:Real-spaceprofilesofthecondensatedensityψ(x,y)2.Onecansee,thatapplicationofamagneticfieldleadstoastrongerlocalizationofthecondensate.Theexcitationspotdiameteris1.5µm.Theotherparametersusedare:Γin≈0.005µeV,w=1cm2/s,aB=10nm,τx=100ps,τ=10/(1+B2/B21)ps,B0=1T,B1=7T,D0=1000cm2/s,Dx=100cm2/s.SupplementaryexperimentaldataThissectionsummarizesadditionalexperimentaldatanotpresentedinthemaintextoftheLetter.Letusstartwiththemicropillarsamples.FigureS3presentsthedataonthe8µmdiameterpillarmanifestingtheroleofmagneticfieldonthein-planemotionofthecarriers:duetodiffusionquenchingthecenterregionofthepillarisdepletedwithcarriersathighmagneticfieldswhichresultsinlasingsuppression(indicatedbythearrow).Ontheotherhand,theemissionintensityincreasesmonotonouslywhenthepillarisexcitedatthecenter.Fig.S4presentstheexperimentaldataonthresholdstopolaritonlasingmeasuredfortwomicropillarsampleswithdiameters8and14µm(thedataona5µmsampleispresentedinthemaintext).Bothpillarsclearlyillustratethesuppressionofdiffusionbymagneticfieldandshowthethresholdbehaviorwhichcorrelateswellwiththetheory.Howevertheinitialriseofthethresholdin14µmsampleisstillunclearandneedsfurtherinvestigation.TheadditionaldataontheplanarsamplesissummarizedinFigs.S5andS6.Fig.S5showsPLspectraofthesamplewithaplanarmicrocavityforvaryingpumpintensitiesatafixedmagneticfieldB=5Tandnegativedetuningδ=−7meV.Onecanseethattheonsetofapolaritonlasingisclearlyobservedforbothexcitationspots.WeconjecturethatpronouncedPLpeaksobservedathighenergysideoflowpolaritonicbranchfor10µmspotandabovethresholdpoweroriginatefromthecavitymodescharacterizedbylargek-vectors.Indeed,thelenswithalargerNA(10µmspot)integratesthePLsignalinawiderk-spacethanthatwithasmallerNA(100µmspot).Ouradditionalstudydemonstratesthatthephasediagrams(seeFig.3d-finthemaintext)arealsosensitivetothedetuning.FigureS6showsthephasediagramsobtainedforlarge100µm(upperpanel)andsmall10µm(lowerpanel)spotsmeasuredatoptimalnegativedetuningδ=−7meV(thehighestefficiencyoflasing)andδ=−3.3meV.Onecanseethatdeviationofthedetuningfromtheoptimaltolessnegativevaluessignificantlydecreasestheeffectofmagneticfieldonthresholdforlargespot.Incontrasttheeffectofdetuningvariationforsmallspotisalmostnegligible.FigureS7presentstheInput-outputcharacteristicsandtheemissionlinewidthoftheelectricallypumpedmicrocavitysample.Thepolaritonlaserthresholdismanifestedbysharpincreaseintheintensityat2Tand4TasshowninFig.S7a,whiletheweakcouplinglaserthresholdischaracterizedbyasmoothS-curveathighercurrentvalues.Theonsetofpolaritonlasingisalsoaccompaniedbyasharpdropinthelinewidth(seeFig.S7b).Fig.S3:MagnetoPLpatternsofthe8µmdiametermicropillarsampleintheregimeofpolaritonlaser.Thepillarisexcitedatthecenter(leftpanel)andattheedge(rightpanel),whilethesignalisdetectedfromthecenterinbothcases.Polariton lasingP/Pth(B=0)00.20.40.60.81.01.21.4Magnetic field (T)0246810Polariton lasingMagnetic field (T)0246810Mesa 8μMesa 14μFig.S4:Experimentallyobservedphasediagramsforpolaritonemissionofthemicropillarsamples.Leftpanel,pillardiameteris8µm,rightpanel,pillardiameteris14µm. 153515401545155015550.05.0x1031.0x1041.5x1042.0x1042.5x1043.0x1043.5x104Intensity (arb. units)Photon Energy (meV)100mkmB=5TP=1P=0.067P=0.067 15361538154015421544154615481550155215540.05.0x1031.0x1041.5x1042.0x1042.5x104P=0.067P=0.067P=1B=5TIntensity (arb. units)Photon Energy (meV) 100 mkmB = 5 T10 mkmB = 5 TabFig.S5:PLspectraoftheplanarmicrocavityillustratingtheonsetofapolaritonlasing.a,Excitationspotdiameteris100µm,b,excitationspotdiameteris10µm. 0.91.01.11.21.31.41.51.61.7 δ=--‐3.3meV δ=--‐7meV100µm Threshold Power (arb. units)02460.40.60.81.01.2 δ=--‐3.3meV δ=--‐7meV10 µm Magnetic Field (T)Fig.S6:Polaritonthresholdoftheplanarmicrocavitysample.Thedatafortwodifferentdiametersoftheexcitationspotandtwovaluesofdetuningispresented.(cid:48)(cid:46)(cid:49)(cid:49)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:48)(cid:48)(cid:69)(cid:76)(cid:73)(cid:110)(cid:116)(cid:101)(cid:110)(cid:115)(cid:105)(cid:116)(cid:121)(cid:40)(cid:97)(cid:114)(cid:98)(cid:46)(cid:117)(cid:46)(cid:41)(cid:32)(cid:48)(cid:46)(cid:49)(cid:49)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:48)(cid:48)(cid:49)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:46)(cid:49)(cid:49)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:49)(cid:48)(cid:48)(cid:48)(cid:48)(cid:48)(cid:32)(cid:67)(cid:117)(cid:114)(cid:114)(cid:101)(cid:110)(cid:116)(cid:32)(cid:68)(cid:101)(cid:110)(cid:115)(cid:105)(cid:116)(cid:121)(cid:32)(cid:40)(cid:65)(cid:47)(cid:99)(cid:109)(cid:50)(cid:41)(cid:66)(cid:61)(cid:48)(cid:84)(cid:66)(cid:61)(cid:50)(cid:84)(cid:66)(cid:61)(cid:52)(cid:84)101000.011100100001000000Current Density (A/cm2)EL Intensity (arb.u.)150200250300350400Linewidth (µeV)abFig.S7:Input-outputcharacteristicsandtheemissionlinewidthoftheelectricallydrivenpolaritonlaser.a,Input-outputcharacteristicsatamagneticfieldof0T,2Tand4T.b,Input-outputcharacteristicsandtheemissionlinewidthat2T.[1]D.Robart,X.Marie,B.Baylac,T.Amand,M.Brousseau,G.Bacquet,G.Debart,R.Planel,andJ.M.Gerard,SolidStateCommunications95,287(1995).[2]C.Piermarocchi,F.Tassone,V.Savona,A.Quattropani,andP.Schwendimann,Phys.Rev.B55,1333(1997),URLhttp://link.aps.org/doi/10.1103/PhysRevB.55.1333.[3]I.CarusottoandC.Ciuti,Phys.Rev.Lett.93,166401(2004),URLhttp://link.aps.org/doi/10.1103/PhysRevLett.93.166401.[4]I.A.Shelykh,Y.G.Rubo,G.Malpuech,D.D.Solnyshkov,andA.Kavokin,Phys.Rev.Lett.97,066402(pages4)(2006),URLhttp://link.aps.org/abstract/PRL/v97/e066402.[5]M.WoutersandI.Carusotto,Phys.Rev.Lett.99,140402(2007),URLhttp://link.aps.org/doi/10.1103/PhysRevLett.99.140402.[6]E.L.Ivchenko,Opticalspectroscopyofsemiconductornanostructures(AlphaScience,HarrowUK,2005).[7]S.N.WalckandT.L.Reinecke,Phys.Rev.B57,9088(1998),URLhttp://link.aps.org/doi/10.1103/PhysRevB.57.9088.[8]V.NikolaevandM.Portnoi,SuperlatticesandMicrostructures43,460(2008),ISSN0749-6036,pro-ceedingsofthe7thInternationalConferenceonPhysicsofLight-MatterCouplinginNanostructures,URLhttp://www.sciencedirect.com/science/article/pii/S0749603607002121.[9]R.Strobel,R.Eccleston,J.Kuhl,andK.Kohler,Phys.Rev.B43,12564(1991),URLhttp://link.aps.org/doi/10.1103/PhysRevB.43.12564.[10]R.Kumar,A.S.Vengurlekar,S.S.Prabhu,J.Shah,andL.N.Pfeiffer,Phys.Rev.B54,4891(1996),URLhttp://link.aps.org/doi/10.1103/PhysRevB.54.4891.

1905.11571

1

1905

Sustained biexciton emission in colloidal quantum wells assisted by dopant-host interaction

Biexcitons have been considered as one of the fundamental building blocks for quantum technology because of its overwhelming advantages in generating entangled photon pairs. Although many-body complexes have been demonstrated recently in mono-layer transition metal dichalcogenides (TMDs), the low emission efficiency and scale up capability hinder their applications. Colloidal nanomaterials, with high quantum efficiency and ease of synthesis/processing, are regarded to be an appealing complement to TMDs for biexciton sources. However, a progress towards biexciton emission in colloidal nanomaterials has been challenging largely by small binding energy and ultrafast non-radiative multiexciton recombination. Here, we demonstrate room-temperature biexciton emission in Cu-doped CdSe colloidal quantum wells (CQWs) under continuous-wave excitation with intensity as low as ~10 W/cm2. The characteristics of radiative biexciton states are investigated by their super linear emission with respect to excitation power, thermal stability and transient photophysics. The interaction between the quantum confined host carriers and the dopant ions increases biexciton binding energy by two folds compared to the undoped CQWs. Such strong binding energy together with suppressed Auger recombination and efficient, spectrally narrow photoluminescence in a quasi-2D semiconductor enables sustained biexciton emission at room temperature, providing a potential solution for efficient, scalable and stand-alone quantum devices.

cond-mat.mes-hall

cond-mat

Sustained biexciton emission in colloidal quantum wells assisted by dopant-host interaction Junhong Yu+1, Manoj Sharma+1, Mingjie Li+2, Pedro Ludwig Hernández-Martínez1, Savas Delikanli1, Ashma Sharma1, Yemliha Altintas3, Chathuranga Hettiarachchi5, TzeChien Sum, Hilmi Volkan Demir*1,2,3, Cuong Dang*1,4,5 1LUMINOUS! Centre of Excellence for Semiconductor Lighting and Displays, School of Electrical and Electronic Engineering, Nanyang Technological University, 50 Nanyang Avenue, 639798, Singapore 2School of Physical and Mathematical Sciences, Nanyang Technological University, 639798, Singapore 3Department of Electrical and Electronics Engineering and Department of Physics, UNAM-Institute of Materials Science and Nanotechnology, Bilkent University, Bilkent, Ankara, 06800, Turkey 4CINTRA UMI CNRS/NTU/THALES 3288, Research Techno Plaza, 50 Nanyang Drive, Border X Block, Level 6, 637553, Singapore 5Centre for OptoElectronics and Biophotonics, School of Electrical and Electronic Engineering & The Photonics Institute, Nanyang Technological University, 50 Nanyang Avenue, 639798, Singapore + Junhong Yu, Manoj Sharma and Mingjie Li contribute equally to this work * Email: [email protected]; [email protected] Abstract: Biexcitons have been considered as one of the fundamental building blocks for quantum technology because of its overwhelming advantages in generating entangled photon pairs. Although many-body complexes have been demonstrated recently in mono-layer transition metal dichalcogenides (TMDs), the low emission efficiency and scale up capability hinder their applications. Colloidal nanomaterials, with high quantum efficiency and ease of synthesis/processing, are regarded to be an appealing complement to TMDs for biexciton sources. However, a progress towards biexciton emission in colloidal nanomaterials has been challenging largely by small binding energy and ultrafast non-radiative multiexciton recombination. Here, we demonstrate room-temperature biexciton emission in Cu-doped CdSe colloidal quantum wells (CQWs) under continuous-wave excitation with intensity as low as ~10 W/cm2. The characteristics of radiative biexciton states are investigated by their super linear emission with respect to excitation power, thermal stability and transient photophysics. The interaction between the quantum confined host carriers and the dopant ions increases biexciton binding energy by two folds compared to the undoped CQWs. Such strong binding energy together with suppressed Auger recombination and efficient, spectrally narrow photoluminescence in a quasi-2D semiconductor enables sustained biexciton emission at room temperature, providing a potential solution for efficient, scalable and stand-alone quantum devices. Entangled photon pairs are natural candidates for basic entities ('qubits') in quantum information and quantum computation1-3. Commonly, parametric down conversion (PDC) in nonlinear optics is utilized to create entangled photon pairs4. However, PDC mechanism suffers from non-guaranteed single photon pair as well as extremely low non-linear efficiency5. Utilizing biexciton states in quantum emitters is a promising strategy to circumvent the limitations since the radiative decay of a biexciton has two channels: either emit a horizontally (H) polarized photon or a vertically (V) polarized photon1-3,6. If fine-structure splitting (FSS) is negligible, the generated two photons are 7. Benefiting from the indistinguishable in any other properties except the polarization: 1 2HHVVstrong many-body interaction in atomically thin transition metal dichalcogenides (TMDs), the first observation of biexciton emission at 10 K using femtosecond pulse excitation was achieved 3 years ago8. Since then, lots of investigations have been conducted to demonstrate continuous-wave pumped biexciton states at room temperature in TMDs7,9-14. However, the synthesis (i.e. mechanical exfoliation, chemical exfoliation, and chemical vapor deposition) of high quality TMDs is still suffering from various issues15-17 such as synthesizing cost, upscaling capability (curbed by flake size and film uniformity), emission efficiency (quantum yield: < 5%) or limited spectral tunability (constrained by the valley band). The use of colloidal nanomaterials offers an avenue for fully solution-processed, highly efficient, bottom-up biexciton sources, which in turn enable the easy integration of quantum emitters into a wide variety of optoelectronic quantum architectures18,19. Till now, the demonstration of biexciton states in colloidal nanomaterials can only be achieved under intense excitation with ultrashort pulsed lasers and cryogenic condition due to the two key challenges: very fast Auger decay20,21, and small biexciton binding energy22,23. The former relates to enhanced non-radiative recombination of multi-exciton states in nanometer confinement with a typical time constant of tens to hundred picoseconds, which leads to ultrafast deactivation of the high-order correlated excitonic states. The latter relates to Coulomb interaction to form a bounded four-body complex from two excitons. The biexciton binding energy in colloidal nanomaterials22 is typically smaller than 30 meV, comparable to the thermal energy at room temperature and much smaller than the inhomogeneous broadening of exciton emission. Insufficient binding energy causes the thermal dissociation of biexciton states at room temperature and makes biexcitonic transition inseparable from exciton states. Suppression of Auger recombination has been investigated extensively for colloidal nanocrystals to achieve optical gain practically. Large-volume nanocrystals leveraging the well-known 'volume scaling'21,24 or interfacial alloying core/shell nanocrystals20,25 utilizing the smoothed confinement potential exhibit relatively long Auger recombination (nanosecond timescales). Unfortunately, large-size nanomaterials which lose the benefits of strong quantum confinement (i.e. large binding energy and reduced dielectric screening)24,26 and smooth interface nanocrystals which associate with broad photoluminescence (PL) spectrum (FWHM: > 30 nm, caused by the interface composition gradient)20,27 are hard to be ideal biexciton sources. The requirements for colloidal nanomaterials which reconcile suppressed Auger recombination, quantum confinement and narrow emission spectrum can be satisfied with colloidal quantum wells (CQWs)18,28,29. In this quasi-2D system, the exciton wave functions are quantized only in vertical direction, which makes Auger process hard to meet the momentum conservation compared with the situation in colloidal quantum dots18,22. Also, the exciton emission in CQWs exhibits extremely narrow linewidth at room temperature (FWHM of 4-monolayer CdSe CQWs: ~40 meV) due to vertical thickness control at the single atomic layer level and weak phonon coupling30-34. However, the highest biexciton binding energy reported for CQWs (~ 34 meV) is still smaller than their emission bandwidth, and thus the biexciton emission peak is only resolved at cryogenic temperature22,31,33. Brovelli and colleagues35 have shown that in Ag doped CdSe colloidal quantum dots (CQDs), by exploiting the effect of excitonic carriers in host nanomaterials on dopant ions, the nonmagnetic dopants can exhibit optically switchable magnetism. Ag+ dopants are not magnetic ([Kr]4d10, there is no unpaired electron spin in full 4d shell), after capturing the photo-generated hole in host CdSe, Ag+ is oxidized to Ag2+ ([Kr]4d9) and the unpaired electron spin can exhibit paramagnetic behavior. Reversibly, we propose that in doped CdSe CQWs, through utilizing the effect of dopant ions on the excitonic carriers in host nanomaterials, the biexciton binding energy can be enhanced. The dopants strongly capturing the photo-generated holes dilute the carrier density in the host nanomaterial, thus increases Debye screening length (reduces the dielectric screening) leading to increased biexciton binding energy36-38. In this work, we have demonstrated Cu dopants in CQWs enhances the biexciton binding energy up to 64 meV, a factor of 2 greater than the values reported in undoped counterparts. The boosted biexciton binding energy, suppressed Auger, and narrow emission spectrum enable continuous-wave biexciton emission at room temperature. A comprehensive analysis of this sustainable biexciton emission in Cu-doped CdSe CQWs (Cu:CdSe CQWs) is presented via the carrier dynamics (time-resolved photoluminescence and transient absorption) with temperature and excitation dependent emission spectra. Cu:CdSe CQWs as a dopant-host system We prepared Cu-doped CdSe CQWs according to the existing recipe39 and optimized the procedures to achieve highly efficient and reproducible synthesis (details of the synthesis and material characterization are described in 2 Supplementary Note 1). The structure information of Cu:CdSe CQWs is shown in Supplementary Fig. S1: the four- monolayer (4 ML) CQWs are 1.2 nm thick18 and tens of nanometers along each side laterally. The PL quantum yield (PLQY) of Cu:CdSe CQWs is in the range of 40% to 80% depends on the amount of Copper precursor added in the synthesis (Fig. S2). In terms of biexcitonic emission behavior, CQW with ~30 Cu dopants (see Methods) is tested in this work unless otherwise mentioned. Figure 1. The dopant-host system in a Cu:CdSe CQW. CE: copper-related emission; BE: band edge emission. (a) Schematic of carrier dynamics in the Cu-doped CQWs with photo excitation. After capturing of a VB hole (process 1), Cu+ is changed to Cu2+, activating the CE through recombining with a CB electron (process 3). (b) Optical absorption and steady-state PL spectrum of Cu-doped CQWs in solution under Xeon lamp illumination, green color region indicates the BE and red color region indicates the CE. (c) Conceptual illustration of the reduced Coulomb screening in Cu-doped CQWs. Utilizing the hole-capture ability of Cu dopants, the carrier density in host CdSe CQWs has been diluted and longer Debye length can be expected. Red curve: with copper dopants. Blue curve: without copper dopants. (d) Normalized decay curves of BE for undoped CQWs and Cu-doped CQWs under the pump fluence of 0.4 μJ/cm2. The PL dynamics of both doped and undoped CQWs are probed at 514 nm. The difference in dynamics sheds light on the suppressed intrinsic traps and the hole capture rate of Cu-dopants. (e) Time-resolved photoluminescence of BE and CE in Cu-doped CQWs with the pump fluence of 0.4 μJ/cm2. The black circles show the excitonic dynamics (514 nm) and the red squares presents the copper MLCBCT dynamics (integrated between 650 and 750 nm). Fitting the CE dynamics (blue solid line) yields the time constant of hole capture process (~1 ns), which is much slower compared with the Cu-doped quantum dots19. The copper dopants in CQW host behave like a hole-trapper40-44, as illustrated in Fig. 1a. After excitation, copper dopants irreversibly localize the photo-generated holes in CQW host, Cu+ is promoted to Cu2+ and therefore activated as a radiative acceptor for the conduction band (CB) electrons, which can be expressed as35: [Ar]3d9 + e → [Ar]3d10 + hν (hν is the photon energy of Cu-related emission and e denotes an electron in CB)19,39. As a result, the photoluminescence (PL) spectrum (Fig. 1b) of Cu:CdSe CQWs is composed of two emission bands: (1) a narrow band-edge emission (BE) at ~514 nm with a full-width at half-maximum (FWHM) of ~40 meV, which is not affected by copper dopants; (2) a broad Cu-related emission (CE) peaking at ~700 nm with a very large FWHM of ~350 meV, the broad emission linewidth is arising from a wide distribution of copper energy level rather than size/doping inhomogeneity within an ensemble of Cu:CdSe CQWs, as detailed previously40,41,44,45. The absorption spectrum (Fig. 1b) of Cu:CdSe CQWs is also consistent with the scenario presented in Fig. 1a: two clear excitonic features associated with electron to heavy-hole (512 nm) and electron to light-hole (480 nm) transitions, remain unchanged compared with undoped CdSe CQWs (absorption spectra of undoped and doped CQWs with different copper doping levels are presented in Fig. S2). Notably, there exists a weak and broad absorption tail on the red side of the first CQW excitonic maximum (Fig. S2), which can be assigned as the direct transition of d orbital electrons in Cu+ to the CB of CdSe CQWs39,41,43. However, the contribution of this transition to BE and CE is eliminated since the extinction coefficient -1cm-1 (Fig. S3a) and without the hole-capture process, the of this tail at its maximum is smaller than 1000 MCu extremely low recombination rate (0.002 ns-1, Fig. S3b) between Cu2+ and CB electron cannot compete with the band- edge recombination. 3 In quantum-confined nanomaterials, the screened Coulomb potential for electrons and holes can be reduced with , where <N> is the number of photo-generated longer Debye screening length (in neutral systems, excitons per nanoparticles)46-47. Several groups have demonstrated that the binding energy in nanomaterials can be , where ne is the significantly weakened via free electron injection since the screening length ( number of injected electrons) is smaller compared to the neutral case36,37. Accordingly, in Cu:CdSe CQWs, as conceptually illustrated in Fig. 1c, the hole-localization ability of Cu+ can strongly dilute the hole density and expand is a ratio to consider the widely the screening length in host CdSe CQWs ( , where distributed energy levels of Cu+). Thus, enhanced many-body interaction can be expected in this host-dopant system. In view of this host-dopant system, the diluted hole concentration in CQW host is suggested by strong CE (Fig. 1b). It is important to ensure the hole-capture dynamics can support the radiative many-body states since the ultrafast hole consumption leads to insufficient constituent (holes) for many-body complexes. To figure out the kinetics of hole capturing in Cu:CdSe CQWs, Fig. 1d presents the BE dynamics of undoped and doped CQWs in the sub-nanosecond time scale, which shows that the ultrafast recombination channel (~31 ps) is absent in doped CQWs. The single exponential decay behavior of Cu:CdSe CQWs in a short time window is consistent with the theoretical calculation conducted by Nelson44,48, in which highly localized covalent [CuSe4] cluster can suppress or deactivate intrinsic traps and thus, enhance the quantum efficiency of Cu-doped nanocrystals (Fig. S2). Here, we can assume that Auger can be negligible in low pumping regime (<N> is smaller than 0.1), only hole capture process is responsible for the difference in BE dynamics between undoped and doped CQWs. Thus, we can extract the time constant of hole capture process49 (τhc) from exciton lifetime of doped (τdoped) and undoped (τundoped) CQWs using 1/τundoped +1/τhc = 1/τdoped, which is around 1.1 ns. Furthermore, since the prerequisite of CE is Cu+ localizes a hole, hole-capturing time can be calculated from the rising part of CE43. As shown in Fig. 1e, in a nanosecond time window, a slow rising process in CE indicating a time constant of 977±45 ps is in good agreement with hole-capture dynamics extracted from Fig. 1d. It is worth to mention that, the onset of CE is significantly delayed compared to BE, emphasizing again the proposed carrier dynamics in Fig. 1a. In addition, as shown in Fig. S4, by comparing the gated emission spectra of Cu:CdSe CQWs when CE is absent (integrated time window: 0~1 ns), and when both CE and BE are present (integrated time window: 1~5 ns), we can draw the same conclusion that the time constant of hole-capture process in Cu:CdSe CQWs is prolonged and in the range of nanosecond. Compared with the ultrafast hole-capture process (an averaged lifetime of ~25 ps) in Cu:CdSe CQDs43, we attribute the slow dynamics in CQWs to the larger spatial separation of band-edge excitons and copper dopants due to large lateral size of the structure. The 2D morphology can slow down the hole localization, suppress the Auger recombination and lengthen the lifetime of multiple-exciton, as four-exciton states observed in CdSe/CdTe CQWs using transient absorption spectroscopy50,51. Continuous-wave excited biexcitons We have therefore obtained a colloidal nanomaterial with several unique properties that are crucial to radiative many- body complexes: reduced Coulomb screening, slow hole localization and suppressed Auger. To identify the potential many-body complexes in Cu:CdSe CQWs, PL of the drop-cast film under continuous-wave excitation with different intensity at room temperature has been studied (Fig. 2a and Fig. 2b). With low pump intensity (< 6 W/cm2, the bottom spectrum in Fig. 2b), the BE shows single symmetric peak located at ~514 nm, which is assigned to typical single exciton emission4,5. With gradually increased excitation intensity, a new emission feature appears at the low-energy side (peaking at ~528 nm) and the intensity of this new feature grows much faster than the intensity of exciton emission with increasing excitation intensity (see the recorded movie in Supplementary Materials). Beyond the excitation intensity of ~10 W/cm2, the new feature emerges as the strongest emission channel (see Fig. 2c, the maximum intensity of BE shifts from 514 to 528 nm with higher excitation intensity). To quantitatively analyze the properties of the emerging emission feature, BE is fitted by two separately symmetric emissions using Voigt function22,52, as shown in Fig. 2b. The fitting profiles are in good agreement with the experimental spectra with different excitation intensities. After spectrally separating the exciton emission and the new emission feature, Fig. 2d presents the integrated emission intensities as a function of the excitation intensity. The excitation intensity dependence of these two features can be described adequately by the power-law equation8-12: k. The exciton emission exhibits a slightly sublinear fluence dependence (k = 0.91), due to the hole Iemission IIntensity 4 1/DebyeLN1/DebyeeLNn1/DebyeCuLNncapturing process from host CQWs to Cu+ sites8,43. However, the integrated intensity of the new emerged emission displays a superlinear growth with excitation intensity (k = 1.68). On the basis of nearly quadratic (our case is ~1.85) intensity dependence of the new emission feature with respect to the exciton emission, the new feature peaked at ~528 nm can only be assigned to biexcitons formed from two excitons, the population of each exciton growing almost linearly (~ 0.91) with excitation intensity12. This is similar to the case of biexcitons in other systems, providing a first evidence that Cu:CdSe CQWs support radiative biexcitons at room temperature with extraordinary low excitation intensity (~ 10 W/cm2). Figure 2. Photoluminescence spectra of Cu-doped CQWs at room temperature with varied continuous-wave excitation intensities. (a) Normalized BE map of Cu:CdSe CQWs under different excitation intensities. The white dashed lines indicate different excitation intensities and the ordered numbers correlate to the BE spectra in Fig. 2b. (b) BE spectra recorded for different excitation intensities. Black dots: measurement; green line: the fitting of exciton; blue line: the fitting of biexciton; red line: the overall fitting. (c) Wavelength of the maximum BE intensity as a function of the excitation intensity. Beyond ~10 W/cm2, the intensity of biexciton emission (peaking at ~528 nm) exceeds the exciton emission (peaking at ~514 nm). (d) Logarithmic plot of the integrated emission intensity for exciton (red stars) and biexciton (blue circles) as a function of the excitation intensities. The green and blue solid lines are the power-law fitting for exciton and biexciton, respectively. Furthermore, the biexciton binding energy can be extracted based on the energy difference between exciton and biexciton emissions assuming that the radiative decay of a biexciton will generate one photon and an exciton (see the derivation in Methods) 8,12. The value of ~64 meV derived here is two-fold of the biexciton binding energy in undoped CdSe CQWs22 (see PL spectra of undoped CdSe CQWs in Fig. S5). This is also higher than that of inorganic colloidal perovskites52 and surprisingly comparable to the value in monolayer WSe2 8. Instead of classical explanation for the enhanced biexciton binding energy using reduced Coulomb screening as illustrated in Fig. 1c, we have performed quantum-mechanical calculation using a biexciton Hamiltonian with a perturbed electron-hole interaction potential by copper dopants to obtain deeper physical insight into the biexciton state. The simulation yields an enhancement of biexciton binding energy of ~30% in Cu:CdSe CQWs compared to undoped CdSe CQWs, which is in good agreement with the experimental result (see Fig. S11, Fig. S12 and details of the calculation in Supplementary Note 3). 5  Figure 3. Photoluminescence spectra of Cu-doped CQWs with fixed excitation intensity of 10 W/cm2 at different temperatures. (a) BE map of Cu:CdSe CQWs at different temperatures. The white dashed lines correspond to different temperatures and the ordered numbers correlate to the BE spectra in Fig. 3b. (b) BE spectra recorded at different temperatures. Black dots and red/green/blue lines have the same denotations as expressed in Fig. 2. (c) Integrated biexciton emission intensity (blue stars) as a function of 1/T. The red line is the Boltzmann distribution fitting, which yields an activation energy of ~60 meV. (d) Normalized CE map as a function of the temperature. the suppressed CE suggests the weaker hole capture ability of copper sites at low temperature. To probe the biexciton states in more detail, we have investigated the thermal stability of biexcitons in Cu:CdSe CQWs at the excitation intensity of 10 W/cm2. As temperature decreases from 298 to 125 K, besides the expected spectra change and increased emission intensity (Fig. 3a and Fig. 3b), biexciton emission displays a small intensity increasing rate when T < 100 K (Fig. 3c). This trend can be explained by the assumption8 that the formation rate of biexciton is roughly constant with varied T, while the dissociation rate (γbiex) is enhanced by the thermal energy: γbiex = γconst + a*exp(-Ea/kBT), where γconst is a fixed recombination rate, a is a ratio factor and Ea is the activation energy. Meanwhile, the integrated emission intensity of biexciton could be well fitted by the Boltzmann distribution (see Methods) 6,8,53 with an activation energy Ea of ~60 meV (Fig. 3c), for which exp(-Ea/kBT) will have a negligible effect on γbiex and the integrated intensity of biexciton emission will show a plateau when T is lower than 100 K. The Ea value of ~60 meV is almost equal to the biexciton binding energy derived from steady-state PL analysis (Fig. 2). This agreement emphasizes again the peak located at ~528 nm arises from biexcitons, in which the dissociation rate can be tuned by the thermal energy. Moreover, with decreased temperature, CE exhibits barely blue-shifted and narrowing spectra, companying with decreased emission intensity (Fig. 3d). This trend is another supporting evidence of the spatial separation of dopant ions and host carriers (see the discussion in Fig. 1d and Fig. 1e): with gradually decreased temperature, thermally assisted carrier movement will be suppressed, weakening the hole capture probability and resulting in reduced CE intensity41. Transient photophysics of biexcitons To furtherly justify our assignment of biexcitons, we have investigated the transient photophysics in Cu:CdSe CQWs. Firstly, we conducted time-resolved PL measurement in Cu:CdSe CQW film with varying excitation fluences (see 6 details in Methods). In the bottom panel of Fig. 4a, a narrow PL spectrum is observed with low excitation fluence (0.4 μJ/cm2, the bottom panel); while with high fluence (0.4 μJ/cm2, the top panel), biexciton emission appears and locates exactly at the wavelength observed in continuous-wave excited PL (Fig. 2 and Fig. 3). Following the white and black dashed lines in Fig. 4a, we compare the PL decay trace for X (located at 514 nm) and XX (located at 528 nm). With low fluence in which biexciton emission is absent, the dynamics of X and XX are almost identical (single-decay process with similar lifetime of ~234 ps and ~245 ps for X and XX, respectively), and their onsets are at the same time (see Fig. 4b). In contrast, with high fluence and thus high exciton density (Fig. 4b), the onset of XX is significantly delayed compared with X, accompanying a fast recombination process with the lifetime of ~128 ps while the lifetime of X is ~201 ps. This delayed and accelerated decay provides another strong evidence of the biexciton emission6,54 at XX. Note that the lifetime ratio between exciton and biexciton is ~1.57, which is close to the value reported in traditional epitaxial quantum wells (GaAs or CdS)55,56. The result strongly suggests a suppressed Auger recombination in Cu:CdSe CQWs compared to that in type-I CdSe CQDs20,21, where biexciton lifetime is typically two order of magnitude shorter than exciton lifetime. Moreover, according to the statistics of carrier recombination, the radiative recombination rate of the biexciton should be one quarter of the value in exciton, here, our ratio is smaller than 4 (ref. 21), excluding the efficient Auger recombination in Cu:CdSe CQWs, as discussed in previous studies22,24,25. Figure 4. Time-resolved photoluminescence of Cu-doped CQWs. (a) Streak camera images of Cu:CdSe CQWs under excitation of 100 fs pulsed laser. Top panel: at high excitation fluence (40 μJ/cm2) in which biexciton emission is observed. Bottom panel: at low excitation fluence (0.4 μJ/cm2) in which biexciton emission is absent. Black dashed line is located at 514 nm (X: exciton) and white dashed line is located at 528 nm (XX: biexciton). (b) Decay traces probed at 514 nm and 528 nm and corresponding fitting curves when the excitation fluence is 0.4 μJ/cm2. 514 nm (measurement: black squares, fitting: black solid line), 528 nm (measurement: red circles, fitting: red solid line). (c) Decay traces probed at 514 nm and 528 nm and corresponding fitting curves when the excitation fluence is 40 μJ/cm2. Same color and shape coding as in (b). (d) Normalized PL intensity decay map of BE in Cu:CdSe CQWs (probed at 514 nm) as a function of the time and the excitation fluence. With low excitation fluence, the lifetime of X increases due to the filling of copper sites. Further increasing the excitation fluence, multiple carriers dominate the recombination dynamics of X. (e) Excitation fluence dependent lifetimes probed at 514 nm which is extracted from the 2D contour map shown in Fig. 4d. The red solid line is the fitting based on the coupled kinetic-equation model (see Fig. S9). To shed more light on the effect of copper dopants on band edge emission in host CQWs, we have analyzed the fluence dependent dynamics of exciton (probed at 514 nm). As shown in Fig. 4d, initially increasing fluence results in a slower dynamics and the exciton lifetime will be close to the value in undoped CQWs (293 ps, Fig, 1d), indicating the fully- filled or saturated copper sites. This trend validates the proposed carrier mechanism in Fig. 1a and our method to extract the hole capture lifetime from host CQW to the Cu+ sites (Fig. 1d). With even higher excitation fluence, in addition to the Cu+ saturation, multiple-exciton interaction process is occurring with high excitation fluence and accelerates the exciton dynamics (two-exciton interaction process to promote into biexciton)21,57. We furtherly verify 7 our interpretation by reproducing the fluence dependent exciton dynamics using a copper-perturbed coupled kinetic- equations (see Fig. S8)8,58. Briefly, exciton dynamics is controlled by the interplay between dopant-related exciton decay rate and fluence dependent exciton-exciton interaction process while biexciton formation rate is related to the thermal equilibrium status (see the model in Supplementary Note 2 for details). The quality of the fitting (see Fig. 4e) confirms the validity of our model: with low fluence, exciton dynamics is changing following the filling of Cu+ site; Whereas with high fluence, multiple carrier interaction takes control the decay channel. Moreover, we can also predict the biexciton dynamics and fluence dependent biexciton intensity based on the proposed model, which is consistent with the experimental results (Fig. S10). Figure 5. Transient absorption spectra of Cu:CdSe CQWs, excitation laser: 100 fs @ 400 nm. (a) 2D transient absorption change (ΔA) images of Cu:CdSe CQWs (see the ΔA data with the full wavelength range in Fig. S6b). The measurement was obtained at a fluence of 40 μJ/cm2. Also the negative ΔA beyond the PIA band is only observed in Cu:CdSe CQWs (see the ΔA images of undoped CQWs in Fig. S6a and S6d ). (b) ΔA spectra as a function of wavelength using different delay time at a fluence of 40 μJ/cm2. The negative band around 528 nm exhibits a delayed formation. (c) 2D transient absorption (A) images of Cu:CdSe CQWs at a fluence of 40 μJ/cm2. The red color-coded region indicates the stimulated emission (A < 0) around 528 nm. (d) Time-resolved A spectra (probed at 528 nm) with different fluences. At a fluence of 40 μJ/cm2, the gain can last over a period of >200 ps. The inset shows the dynamics of -A at a fluence of 40 μJ/cm2 which yields a lifetime of 187±5 ps. A most striking observation is the stimulated emission in dispersions of Cu:CdSe CQWs at high excitation fluence in transient-absorption spectra (see Methods), arising from the radiative recombination of biexcitons. As expected, in the ΔA spectra of Cu:CdSe CQWs, we observe a weak absorption bleach (the transition of d orbital electrons in Cu+ to the CB of CdSe CQWs, MLCBCT, which is also discussed in Fig. 1) covering the whole spectrum beyond the red side of the positive photo-induced absorption (PIA) band, as shown in Fig 5a and 5b. This weak absorption bleach exhibits nearly invariant dynamics within the measured time window (> 3ns) which is consistent with the slow recombination between CB electrons and copper-trapped holes (lifetime: ~450 ns, Fig. S3). Notably, after several picoseconds delay, we unambiguously observe the appearance of a strong negative signal (the saturated red region in Fig. 5a) peaked at ~ 528 nm below the PIA band. Here, we can assign the emerged negative signal around 528 nm to a stimulated emission process, because of three reasons as follows. (i) This does not arise from MLCBCT since this process induced bleaching should occur simultaneously with the heavy-hole bleaching (see the curve in Fig. 5b with a delay time of 10 fs)43. Moreover, the strong negative signal exhibits much faster decay than the absorption bleaching of MLCBCT, implying that the negative signal around 528 nm is related to the dynamics of band edge carriers39. (ii) The negative signal peaking at 528 nm shows a delay formation and only appears after several picosecond (Fig. 4b) with respect to the linear absorption bleaching band58,59. (iii) The negative signal can only be observed under high pump fluence (see 8 the images of ΔA in Cu:CdSe CQWs with different pump fluences in Fig. S6e) and the energy location is exactly matching our biexciton emission spectra50,58,59. The stimulated emission from biexcitons is further verified by the absolute absorbance A(λ,t) = ΔA(λ,t)+A0(λ), where A0(λ) is the linear absorption with the same measurement condition59,60. Fig. 5c presents the contour map of absolute absorbance at a pump fluence of 40 μJ/cm2 which codes the positive absorbance with blue color and gain (A<0) with red color. The optical gain region is exactly matching the biexciton emission resolved in previous part (dashed black rectangle) and can last more than 100 ps. The dynamics of stimulated emission with different pump fluences probed at the wavelength of 528 nm is shown in Fig. 5d. with low fluence, no gain can be achieved. With increasing fluence (> 10 μJ/cm2), the value of A turns negative after the initial short-lived (< 1 ps) positive feature caused by hot carriers. Significant amplification is observed at a pump fluence of 40 μJ/cm2 with a minimum absorbance value of ~2.4 mOD (ASE is also demonstrated in Cu:CdSe CQW film with a low threshold of ~53 μJ/cm2, see Fig. S7). At the fluence of 40 μJ/cm2, the optical gain relaxation is fitted to a bi-exponential decay. The slow decay component with a lifetime of ~187 ps matches well with the PL decay probed at 528 nm (~128 ps, Fig 4c), confirming again the long-lived biexciton states support the observation of sustainable biexciton emission in Cu:CdSe CQWs22,50. Interestingly, the gain dynamics shows a fast component (with a lifetime of 8.67 ± 0.2 ps) which is absent in the trPL measurement. This fast decay in absorbance is likely to be assigned to multi-exciton recombination. Discussion We have discovered the four-particle neutral biexciton in Cu-doped CdSe CQWs, which is unambiguously identified by its transient photophysical properties, superlinear emission intensity, and thermal stability. Further studies are necessary to understand formation mechanism and constituent of biexciton, where the roles of a dark exciton was not observed in our Cu:CdSe CQWs. This is in contrast with the biexciton mechanism in WSe2 where slow dark excitons enable biexciton states with continuous-wave excitation12. It is also worth to explore whether biexcitons in Cu-doped CQWs could have a more significant fine structure splitting since the energy difference between photon pair generated from biexciton can reduces the pure polarization entanglement substantially13. More evidences to characterize the electron-hole exchange interaction and optical selection rules could furtherly check the potential of biexciton in Cu- doped CQWs as practical entanglement photon sources. Nevertheless, we have demonstrated biexciton emission in Cu-doped CdSe CQWs under continuous-wave excitation at room temperature. The achievement is attributed to the signatures of the material: large biexciton binding energy enhanced in host-dopant system and suppressed non-radiative Auger recombination mediated in 2D confinement structure. The observation of spectrally separable, radiative biexciton states is a key to study coherent many-body physics, such as condensation and superfluidity. Further, such sustainable biexciton emission at room temperature, using fully solution-processed colloidal assemblies of nanomaterials, offers broad potential applications in practical quantum optoelectronics, including quantum logic gates and high-order correlated photon emission. Methods Linear optical characterization of Cu:CdSe colloidal quantum wells. Absorption spectra of Cu:CdSe CQWs in hexane is measured by a UV- VIS spectrophotometer (Shimadzu, UV-1800). PL spectra of Cu:CdSe CQWs in hexane are recorded using a fiber-coupled ANDOR spectrometer (monochromator: ANDOR Shamrock 303i, CCD: ANDOR iDus 401) with a diode laser excitation (Cobolt 06-MLD, excitation wavelength: 405 nm). Quantum yield (QY) of Cu:CdSe CQWs in hexane is measured with an integrating sphere and calculated as the ratio of absolute emission and absorption photons. The accuracy of the QY measurement is verified using Rhodamine 6G, whose QY of 94.3% measured in our setup, is found in good agreement with the standard value of 95%. Estimation the Cu atomic level. With Inductively coupled plasma mass spectrometry (ICP:MS) we estimated Cu atomic levels with respect to cadmium and selenium. Average dimensions of our 4 ML Cu:CdSe CQWs measured by TEM microscopy are (36±1.9)×(15.0±1.5)×1.2 nm, which suggests 13500 cadmium and 10600 selenium atoms in one CQW. Therefore, using ICP:MS measurements we can estimate Cu atoms per CQW. Time-resolved photoluminescence measurement. Time-resolved PL (TRPL) measurements are performed with a streak camera from Optronics. The 400-nm pump laser pulses for TRPL are generated from a 1000 Hz regenerative amplifier (Coherent LibraTM). The beam from the regenerative amplifier has a center wavelength at 800 nm, a pulse width of around 150 fs and is seeded by a mode-locked Ti-sapphire oscillator (Coherent Vitesse, 80MHz). 400-nm pump laser was obtained by frequency doubling the 800-nm fundamental regenerative amplifier output using a BBO crystal. All measurements are performed in the solid film at room temperature in ambient air (53±2% humidity) conditions. Continuous-wave pumping photoluminescence spectroscopy. Continuous-wave excitation PL study is performed with a diode laser (Cobolt 06- MLD, excitation wavelength: 405 nm) and a fiber-coupled ANDOR spectrometer (monochromator: ANDOR Shamrock 303i, CCD: ANDOR iDus 401). Samples are drop-casted on a glass substrate for both room and low temperature measurements. The measurement is conducted in a surface 9 emission geometry to avoid the spectra shift caused by reabsorption. The excitation spot radius is 50 μm produced by a plano-concave lens with a focal length of 75 mm. For the low temperature CW PL measurement, sample are cooled with a closed-cycle helium cryostat. Calculation of the biexciton binding energy based on emission spectra. The biexciton binding energy is defined as the energy difference between biexciton states and two free excitons: ∆biex = 2Eex-Ebiex, where ∆biex is the biexciton binding energy; Eex and Ebiex is the energy of exciton and biexciton, respectively. Considering that the radiative recombination of one biexciton will generate one exciton and emit one photon: Ebiex = Eex + hνxx = hνx + hνxx (hνx and hνxx is the photon energy of exciton and biexciton emission, respectively). Thus, we can calculate the biexciton binding energy (∆biex) based on the spectra shift: ∆biex = hνx - hνxx. Thermal intensity distribution fitting. As explained in Fig. 2g, assuming the formation rate of biexciton is roughly constant with varying temperature, while the decay rate (γbiex) is enhanced by the thermal energy, the temperature dependent biexciton emission intensity change can be fitted by a Boltzmann distribution function: b/[1+a*exp(-Ea/kBT)], where b is a constant, a is the coefficient, and Ea is the activation energy which is corresponding to the biexciton binding energy. Transient absorption spectroscopy. TA spectroscopy is performed using a HeliosTM setup (Ultrafast Systems LLC) and in transmission mode with chirp-correction. The white light continuum probe beam (in the range of 400-800 nm) is generated from a 3 mm sapphire crystal using 800 nm pulse from the regenerative amplifier as mentioned in TRPL measurement. The pump beam spot size is ~0.5mm. The probe beam passing through the sample was collected using a detector for UV -- Vis (CMOS sensor). All measurements are performed at room temperature in solution (hexane). Acknowledgements We would like to acknowledge the financial support from Singapore National Research Foundation under the Program of NRF-NRFI2016-08, the Competitive Research Program NRF-CRP14-2014-03 and Singapore Ministry of Education AcRF Tier-1 grant (MOE-RG178/17). H.V.D is also grateful to acknowledge additional financial support from the Z4BA. We would like to thank Muhammad Taimoor and Thomas Kusserow at University of Kassel (Kassel, Germany) for carefully reading our manuscript. Author Contributions C.D. and H.V.D supervised and contributed to all aspects of the research. J.Y., M.S, H.V.D and C.D. wrote the manuscript. J.Y. conducted the spectroscopy measurement and proposed the coupled kinetic model. M.S performed the material synthesis and designed them to achieve the best Biexciton lasing performance. S.D and A.S helped in material synthesis and characterizations. M.L. performed lifetime and transient absorption measurement. M.L. and TC.S. supervised the ultrafast dynamic analysis. P.L.H-M conducted the Hamiltonian calculation to predict the biexciton binding energy. Y.A. conducted the material characterization. All authors analysed the data, discussed the results, commented on the manuscript and participated in manuscript revision. Competing financial interests The authors declare no competing financial interests. References R. M. Stevenson. et al. A semiconductor source of triggered entangled photon pairs. Nature 439, 179-182 (2006). N. Akopian. et al. Entangled Photon Pairs from Semiconductor Quantum Dots. Phys. Rev. Lett. 76, 130501 (2006). Daniel Huber et al. Highly indistinguishable and strongly entangled photons from symmetric GaAs quantum dots. Nature Commun. 8, 15506 (2017). Paul G. Kwiat et al. New High-Intensity Source of Polarization-Entangled Photon Pairs. Phys. Rev. Lett. 75, 4337 (1995). Timo Kaldewey et al. Coherent and robust high-fidelity generation of a biexciton in a quantum dot by rapid adiabatic passage. Phys. Rev. B. 95, 161302 (2017). Chen J. et al. Room Temperature Continuous-wave Excited Biexciton Emission in CsPbBr3 Nanocrystals. arXiv:1804.09782 (2018). Yu-Ming He et al. Cascaded emission of single photons from the biexciton in monolayered WSe2. Nature Commun. 7, 13409 (2016). You, Y. et al. Observation of biexcitons in monolayer WSe2. Nature Phys. 11, 477 -- 481 (2015). Shao-Yu Chen et al. Coulomb-bound four- and five-particle intervalley states in an atomically-thin semiconductor. Nature Commun. 9, 3717 (2018). Kai Hao et al. Neutral and charged inter-valley biexcitons in monolayer MoSe2. Nature Commun. 8, 15552 (2017). Zhipeng Li et al. Revealing the biexciton and trion-exciton complexes in BN encapsulated WSe2. Nature Commun. 9, 3719 (2018). Ziliang Ye et al. Efficient generation of neutral and charged biexcitons in encapsulated WSe2 monolayers. Nature Commun. 9, 3718 (2018). Alexander Steinhoff et al. Biexciton fine structure in monolayer transition metal dichalcogenides. Nature Phys. 14, 1199 -- 1204 (2018). Jiajie Pei et al. Excited State Biexcitons in Atomically Thin MoSe2. ACS Nano 11, 7468 -- 7475 (2017). Ankur Sharma et al. Highly Enhanced Many-Body Interactions in Anisotropic 2D Semiconductors. Acc. Chem. Res 51, 1164 -- 1173 (2018). Wonbong Choi et al. Recent development of two-dimensional transition metal dichalcogenides and their applications. Materials Today 20, 116 -- 130 (2017). Jiajie Pei et al. Many-Body Complexes in 2D Semiconductors. Adv. Mater. 31, 1706945 (2019). Ithurria, S. et al. Colloidal nanoplatelets with two-dimensional electronic structure. Nat. Mater. 10, 936 -- 941 (2011). Debasis Bera et al. Quantum Dots and Their Multimodal Applications: A Review. Materials. 3, 2260 -- 2345 (2010). Dang, C. et al. Red, green and blue lasing enabled by single-exciton gain in colloidal quantum dot films. Nat. Nanotech. 7, 335 -- 339 (2012). Klimov, V. I. Quantization of multiparticle auger rates in semiconductor quantum dots. Science 287, 1011 -- 1013 (2000). 10 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. Grim, J. Q. et al. Continuous-wave biexciton lasing at room temperature using solution-processed quantum wells. Nat. Nanotech. 9, 891 -- 895 (2014). Avidan, A. & Oron, D. Large blue shift of the biexciton state in tellurium doped CdSe colloidal quantum dots. Nano Lett. 8, 2384 -- 2387 (2008). Jaehoon Lim et al. Optical gain in colloidal quantum dots achieved with direct-current electrical pumping. Nat. Mater. 17, 42 -- 49 (2018). Kaifeng Wu et al. Towards zero-threshold optical gain using charged semiconductor quantum dots. Nat. Nanotech. 12, 1140 -- 1147 (2017). Yongfen Chen et al. " Giant" Multishell CdSe Nanocrystal Quantum Dots with Suppressed Blinking. J. Am. Chem. Soc. 130, 5026−5027 (2008). Park Y-S et al. Effect of Auger Recombination on Lasing in Heterostructured Quantum Dots with Engineered Core/Shell Interfaces. Nano Letters 15, 7319-7328 (2015). Ithurria, S. et al. Continuous transition from 3D to 1D confinement observed during the formation of CdSe nanoplatelets. J. Am. Chem. Soc. 133, 3070 -- 3077 (2011). Son, J. S. et al. Colloidal synthesis of ultrathin two-dimensional semiconductor nanocrystals. Adv. Mater. 23, 3214 -- 3219 (2011). She, C. et al. Low-threshold stimulated emission using colloidal quantum wells. Nano Lett. 14, 2772 -- 2777 (2014). Guzelturk, B. et al. Amplified spontaneous emission and lasing in colloidal nanoplatelets. ACS Nano 8, 6599 -- 6605 (2014). Flatten, L. C. et al. Strong exciton -- photon coupling with colloidal nanoplatelets in an open microcavity. Nano Lett. 16, 7137 -- 7141 (2016). Olutas, M. et al. Lateral size-dependent spontaneous and stimulated emission properties in colloidal CdSe nanoplatelets. ACS Nano 9, 5041 -- 5050 (2015). She, C. et al. Red, yellow, green, and blue amplified spontaneous emission and lasing using colloidal CdSe nanoplatelets. ACS Nano 9, 9475 -- 9485 (2015). Pinchetti, V. et al. Excitonic pathway to photoinduced magnetism in colloidal nanocrystals with nonmagnetic dopants. Nat. Nanotech. 13, 1145-151 (2018). Kin Fai Mak et al. Tuning Many-Body Interactions in Graphene: The Effects of Doping on Excitons and Carrier Lifetimes. Phys. Rev. Lett. 112, 207401 (2014). Matteo Barbone et al. Charge-tuneable biexciton complexes in monolayer WSe2. Nature Commun. 9, 3721 (2018). Yuxuan Lin et al. Dielectric Screening of Excitons and Trions in Single-Layer MoS2. Nano Lett. 14, 5569 -- 5576 (2014). Sharma, M. et al. Near-Unity Emitting Copper-Doped Colloidal Semiconductor Quantum Wells for Luminescent Solar Concentrators. Adv. Mater. 29,1700821 (2017). Sharma, M. et al. Understanding the Journey of Dopant Copper Ions in Atomically Flat Colloidal Nanocrystals of CdSe Nanoplatelets Using Partial Cation Exchange Reactions. Chem. Mater. 30, 3265-3275 (2018). Kathryn E. Knowles et al. Singlet−Triplet Splittings in the Luminescent Excited States of Colloidal Cu+:CdSe, Cu+:InP, and CuInS2 Nanocrystals: Charge-Transfer Configurations and Self-Trapped Excitons. J. Am. Chem. Soc. 137, 13138−13147 (2015). Kira E. Hughes et al. Copper's Role in the Photoluminescence of Ag1 -- xCuxInS2 Nanocrystals, from Copper-Doped AgInS2 (x ∼ 0) to CuInS2 (x = 1). Nano Lett. 19, 1318 -- 1325 (2019). Hughes, K. E. et al. Photodoping and Transient Spectroscopies of Copper-Doped CdSe/CdS Nanocrystals. ACS Nano 12, 718−728 (2018). Nelson, H. D. et al. Mid-Gap States and Normal vs Inverted Bonding in Luminescent Cu+- and Ag+-Doped CdSe Nanocrystals. J. Am. Chem. Soc. 139, 6411−6421 (2017). Ward van der Stam et al. Tuning and Probing the Distribution of Cu+ and Cu2+ Trap States Responsible for Broad- Nanocrystals Band Photoluminescence in CuInS2. ACS Nano 12, 11244−11253 (2018). Eric Stern et al. Importance of the Debye Screening Length on Nanowire Field Effect Transistor Sensors. Nano Lett. 7, 3405 -- 3409 (2007). Alpha A. Lee et al. Scaling Analysis of the Screening Length in Concentrated Electrolytes. Phys. Rev. Lett. 119, 026002 (2017). Nelson, H. D. et al. Computational Studies of the Electronic Structures of Copper-Doped CdSe Nanocrystals: Oxidation States, Jahn−Teller Distortions, Vibronic Bandshapes, and Singlet−Triplet Splittings. J. Phys. Chem. C 120, 5714−5723 (2016). FT Rabouw, et al. Delayed exciton emission and its relation to blinking in CdSe quantum dots. Nano Lett. 15, 7718 -- 7725 (2015). Li, Q. et al. Low Threshold Multiexciton Optical Gain in Colloidal CdSe/CdTe Core/Crown Type-II Nanoplatelet Heterostructures. ACS Nano 11, 2545− 2553 (2017). Li, Q. et al. Efficient Diffusive Transport of Hot and Cold Excitons in Colloidal Type II CdSe/CdTe Core/Crown Nanoplatelet Heterostructures. ACS Energy Letters 2, 174−181 (2017). Wang Y. et al. All-inorganic colloidal perovskite quantum dots: A new class of lasing materials with favorable characteristics. Adv. Mater. 27, 7101 -- 7108 (2015). Fujisawa, J.; Ishihara, T. Excitons and biexcitons bound to a positive ion in a bismuth-doped inorganic-organic layered lead iodide semiconductor. Phys. Rev. B 70, 205330 (2004). Bacher, G., et al. Biexciton versus exciton lifetime in a single semiconductor quantum dot. Phys. Rev. Lett. 83, 4417-4420 (1999). Phillips, R. et al. Biexciton creation and recombination in a GaAs quantum well. Phys. Rev. B 45, 4308-4311 (1992). Woggon, U. et al. Huge binding energy of localized biexcitons in CdS/ZnS quantum structures. Phys. Rev. B 61, 12632 -- 12635 (2000). Malko, A. V. et al. Pump-intensity- and shell-thickness-dependent evolution of photoluminescence blinking in individual core/shell CdSe/CdS nanocrystals. Nano Lett. 11, 5213 -- 5218 (2011). Soavi, G. et al. Exciton -- exciton annihilation and biexciton stimulated emission in graphene nanoribbons. Nat. Commun. 7, 11010 (2016). Sutherland, B. R. et al. Perovskite thin films via atomic layer deposition. Adv. Mater. 27, 53 -- 58 (2015). Geiregat, P. et al. Continuous-Wave Infrared Optical Gain and Amplified Spontaneous Emission at Ultralow Threshold by Colloidal HgTe Quantum Dots. Nat. Mater. 17, 35-42 (2017). 11 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60.